>AMD Threadripper DELIDDING Threadripper is Epyc with 2 fully

Source? Their yields must be near perfect if true.

What's that silver stuff between the spreader and the dies, where the cheap thermal paste should be?

Sounds ridiculous. While sharing the same PCB is sensible, I'd just stick in two dies and leave the other two empty.

>>61589218
solder pads.

>>61589162
Fully working dies disabled? What the fuck?

>>61589228
Solder? Lol, sounds shit, Intel uses real paste.

How much money are they losing by using that solder instead of Intel™Recommends paste

>>61589218
Solder.
The yellow stuff is gold coating for the solder to stick to the heat spreader better.
The cum over the capacitors is some kind of glue to protect them.
Black stuff is glue to fix the heat spreader to the PCB.

>>61589229
They are just epyc renamed with overclocks.

>>61589251
10 cents.

>>61589162
Sauce?

File: the moneyshot.png (402KB, 617x463px) Image search: [Google]

402KB, 617x463px

>>61589248
>paste

>>61589194
We've already heard what I guess you call rumors that Zen die yields were very very good. I guess this is exactly just how good.

If Threadripper CPUs could be unlocked to extra dies, this is only making me want it harder.

Looks absolutely disgusting
Good thing Intel dropped this ancient technology for more advanced one

>>61589270
>dellid jew
https://www.youtube.com/watch?v=ZoVK6rJR5VE

>>61589162

>inb4 AMD sandbagged their 20 core threadripper CPU that will be way cheaper than Intel's 18 core housefire cumstain

>>61589283
tyvm

File: 1500345331078.jpg (68KB, 720x543px) Image search: [Google]

68KB, 720x543px

>>61589282
>le mesh architecture

>>61589162
>Threadripper is Epyc with 2 fully working Dies disabled

I thought for their CCX design there must be an equal number of disabled cores on each die?

So in this case, for Threadripper it's 4 cores per die. It would be extremely wasteful to throw perfect 8 core dies away like that

>>61589303
They may just be dummy dies or dies with passthrough and no working cores.

>>61589162
Ignoramus here. What is happening there? Do they arbitrarily gimp the processors to not use all cores?

File: IMG_2644.jpg (2MB, 4032x3024px) Image search: [Google]

2MB, 4032x3024px

>gold plated
REEEEE

>>61589381
Oi vey what a waste. I mean for the environment, of course.

>>61589162

Der8auer<333 German engineering<333

>>61589283

"So, are they using two dies with 8 cores each or 4 dies with half of the cores disabled on each die" *smirks*

>*smirks*

32 core threadripper confirmed but delidman is under NDA

>>61589371
Disabled cores has been a thing since the concept of multi-core CPUs was a thing.

Usually it's CPUs with one or two cores that didn't pass testing, so you disable them and sell it as a lower-end model to save waste and money, sensible.

What happens when your yields are so high that you have very few failing the tests though? Well you have to disable good cores to make up stock then. It's happened in the past, everyone remembers unlocking their athlons and phenoms for more perfectly serviceable cores.

File: 1491601973605.jpg (66KB, 720x707px) Image search: [Google]

66KB, 720x707px

>>61589162
I like how OP isn't posting source sauce but is stringing people along w/ his pictures.

Here is the source sauce anons :
https://www.youtube.com/watch?v=ZoVK6rJR5VE

What I have to say at this point is .. (wow) AMD.
You've shipped your flagship CPU to a bunch of E-celebs who are hacking away at it and monkeying around yet deliver your consumers up to OEMs/Pre-built Manufs before launch and before you properly detail the :
> CPU
> Motherboards
> Cooling solution

I have no clue what kind of NDA these people are on but clearly not much as one of them was allowed to butcher and destroy a thread ripper CPU and post it to his youtube channel. What exactly does this do for AMD?

Now, to the hardware... 4 Dies w/ 2 disabled under the hood? Wow guys...
Is this why the Wattage is 180W? The detailing before was 2 dies. However, clearly looking at the socket on the mobos and pinouts therein it wasn't. Could they not have just detailed the obvious before and say it was 4 dies?

So, which dies are disabled? random ones?

So, Threadripper is binned and overclocked Epyc Processors
If so, is the infinity fabric interconnect that would have normally connected the 4 and traces therein are likely sitting there dormant or sucking juice?

Epyc has 128 pci-e lanes ... Threadripper 16/32 has 64. So, there is interconnect disabled on the 2 active dies.

I don't know what to make of this desu.. On one hand, sure.. 32/64 future. Everyone knew that though looking at the socket/pin-out. For now though, instead of 2 properly connected complexes, we have 2 randomly disabled ones and binned epyc serving as a threadripper? Going to have to think about this for a while.

Maybe will revisit this platform up the road once its proper.
That being said, I have to say that they are incredibly fumbling these launches.

Really not a good look on your flagship line.

>>61589417
Better announce that 28 core i11.

File: 1351433114003.png (73KB, 1000x1000px) Image search: [Google]

73KB, 1000x1000px

>>61589462

>>61589462
It's 4 cores on each die. Doesn't make sense otherwise. They're using whatever is left from Epyc dies.

>>61589475
No. Its 2x full 8c dies working and 2x8c dies disabled.

>>61589303

It isn't required, per se, but disabling an equal number of cores per die helps to keep latency consistent.

>>61589463
They'd have to partner with CERN to power it.

>>61589483
How can you be sure?

>>61589251
About $2-4

>>61589462

>this fucking wall of text

how many X299 raid keys did intel pay you for this

>>61589494
fuck off loch ness monster you crafty fuck

>>61589492
>source: amd
2x dies - 64 pcie lanes and 4c memory
4x dies - 128 pcie lanes and 8c memory.

File: c4713340f3bee2e136da964a67fa7234.jpg (43KB, 640x640px) Image search: [Google]

43KB, 640x640px

>>61589462
This is a perfect example of people that should not post here.

>>61589483

>source: my dad forks for nintendo

>>61589519
Watch the video cunt

>>61589519
As Der8auer(dellid jew) said this 8C+8C for 1950X and 8C+4C for 1920X.

>>61589229
no, if the dies were working properly they would be sold as 24 or 32 core epyc.

>>61589225
These are probably lower binned parts using one ccx per die.

>>61589551
He said AMD said it's 8+8 for the 16 core, but he speculated about the 12 core.
Personally I bet it's 6+6

File: 1494911910249.jpg (41KB, 420x420px) Image search: [Google]

41KB, 420x420px

>>61589473
>>61589499
>>61589513
I call it like it is...
I got 0 raid keys and have serious critiques of Intel's x299 platform.
I was literally set to pre-order the Gigabyte mobo here :
> http://event.gigabyte.us/x399promo/#
Until I discovered they aren't actually having mobo pre-sales...
It was an elaborate bait and switch for pre-built systems similar to how an OEM is being allowed to front run the product.

Also I was set to buy a threadripper as soon as it was available.
Given this kind of detailing, I'm more likely to wait until the platform, from all perspectives is polished.
> Motherboard manufacturers start taking it seriously as a workstation CPU and offer workstation motherboards and not gaming RGB light shows at top dollar
> Bugs get worked out and lord knows there will be many
> etc etc..

>>61589475
Watch the video. AMD's e-celeb pre-launch delidder said he called AMD and they confirmed its only 2 active dies... Which means there's all sorts of dead interconnects sitting latent on the chip and on the active dies.

On the active dies, there is dead infinity fabric interconnects leading to the other 2 dies... Allowing an e-celeb to hack away at your pre-launched product and detail this is not the way you get you showcase a flagship product.

It was said there would be happenings today...
Funny .. the timing of this.
Bright and early before this mobo pre-sale fiasco gets sorted.

It's time for bed. I've seen what I needed to see.

dumb frogposter

>>61589698
>Muh feelings
>Muh implications
>Muh innuendos
>Muh FUD
|
|>
|
|3
|

File: epyc.jpg (86KB, 700x423px) Image search: [Google]

86KB, 700x423px

>>61589698
Even more of a reflection for consideration is the fact that you can't do much beyond what is being done with the X399 platform....

There is no more room on the ATX/E-ATX motherboard. So, the current mobos already show the limit of what can be done here.

16/32 would have been solid AF at 2 dies and a smaller socket to make some room on the motherboard.

Instead, they put a server class processor socket taking up yuge amounts of space and making the mobo layout look comical to save money on binning Epycs.

People are talking about 32/64 but ignoring the huge waste that would be as you can't scale the memory to be in support of it nor can you scale PCI-E slots...
There's no more room. That's it. So, barring a new mobo standard for the desktop, this is literally it right here albeit with a gimped out/binned chip.

> No 16 dimm slots to feed 4 dies properly
> No more room for 128 PCI-E slots on an ATX/EATX

Unless they're going server style w/ riser slots and changing the whole game of desktop computing w/ a new motherboard standard... Threadripper ends here.

Epyc is the real variant with real utilization of the socket.
The habbenings of the 27th have begun in epic fashion

File: 1483167353491.jpg (23KB, 439x290px) Image search: [Google]

23KB, 439x290px

>>61589462

>>61589698
Go clutch your pearls somewhere else, fagtron

>>61589642
>no, if the dies were working properly they would be sold as 24 or 32 core epyc.
Not true, if yields are high there's not always enough failed dies for that to work. How do the ram channels work? If it's using two dies with 2 channels each then activating disabled cores is likely impossible. If it's one die using 1 channel maybe it can be done.

>>61589462
>>61589698
>>61589839
Holy fucking wall of text.

>>61589962
The sad part is he had all that typed out in advance.

File: confus.jpg (19KB, 346x300px) Image search: [Google]

19KB, 346x300px

>>61589162
Anyone else think the 8+8+0+0 configuration may make these things weird to cool?
What if all the heat is down in that end. Or over on that side?

>>61589839
I would buy an E-ATX motherboard and $2000 1p epyc if they made decent workstation boards and either unlocked epyc or sold 32 core threadrippers. But nooo this is X86 products must always be crippled and limited somehow, can't have customers being allowed to have high clockspeeds and lots of cores. But 400 watt videocards are completely normal.

>>61589988
Thank fuck we've got good contact between the dies and the IHS. This chip would be an X299-tier housefire if it used Jewtel thermal jizz.

Hopefully the baseplates on the new coolers will compensate for any oddities.

File: epyc.png (252KB, 1462x480px) Image search: [Google]

252KB, 1462x480px

>>61589823
No feels lad.
Pure technical critique...
If you disagree with any particular point, speak up.

...Just able to see where the thread ripper road ends (for now) .. now that I know what's under the hood.

If you were technically informed at how zepplin dies are configured and how the die interconnect works, you'd understand this...
(see pic)

This is what 32/64 looks like :
http://b2b.gigabyte.com/Server-Motherboard/MZ31-AR0-rev-10#ov
It's server standard connections w/ riser cards to service the 128 PCI-E
slots. It's available right now .. It's what thread ripper 32/64 looks like down the road. So, if you want that, you buy epyc.


You can't do anything with pushing threadripper to 32/64 on ATX/EATX form factor.

Feels is what had me ready to invest in this platform at pre-order/launch when I thought it was 2 dies but, given the silly size of the socket and pinouts, had obvious suspicion it wasn't.

The feels are gone lad. They let their processor out into the wild and let it get stripped down to by a youtube e-celeb before launch. Just think about this for a minute. This is the first unveiling their CPU got.. Some clown deliding it, drilling holes into the heat spreader, and roasting it... Only to confirm its a binned epyc and no performance numbers. This is what's supposed to excite and hold people over until launch.

Indium solder again? AMD lost itself a buyer, I run servers for 10+ years and solder cracks and shits itself in under a few years, Intel has the sensible professional choice

File: mitx threadripper.jpg (114KB, 492x483px) Image search: [Google]

114KB, 492x483px

>>61590062
It was obviously repurposed Epyc shit when the sockets were exactly the same size.

File: Your+bait+is+bad+and+you+should+feel+bad.png (4KB, 622x626px) Image search: [Google]

4KB, 622x626px

@61590070 (You)

>>61589973

The even sadder part is it's all a big song and dance circling around the topics and absolutely nothing concrete.

>>61590070
*looks at his still working K8, Phenom and C2D*

Right

>>61590062
>The feels are gone
Nice. Another wall of text of muh feelings, muh innuendos, muh implying implications.

Kill yourself FUD shill.

>>61590041
But you know what I mean right.
"Three of my seven heat pipes aren't touching a hot part of the IHS. Oh well"

>>61590070
Are you payed per post or by the hour?

>>61590062
This is all feels, "lad"
>muh DEAD interconnect
>muh eceleb teardown
>muh riser cards

seriously the most pathetic criticism I've ever witnesses. Go to bed, Pajeet.

>>61590070
>>61589248
Samefag

Motherboards that enable the other 2 dies, when. I fucking remember the old days with the AMD chips that had disabled cores and good guys like ASRock came forward and enabled them in a partnership with AMD to make AMD a shit ton more worth it than Intel due to paying piss cheap for 2 cores and getting 2 cores for free.

Apparently it's just cheaper doing EPYC/TR baking and assembly in one place if they can throw dies away.
TR isn't a high volume product so it's probably cheaper this way too.

>>61589162
>waste solder and gold for disabled die
the kikes at intel are rolling in their ovens.

>>61589854
>>61589930
I take no sides. I'm a neutral party.
If you had counter points to what was stated, you would have declared them. Given that you don't you resort to idiotic (muh shill) replies.

If you're gonna rub salt and be fair.
Salt goes in both wounds ... x299/x399 now.

All of this could have been resolved by a smaller socket to fit 2 proper dies.
Instead, you get an obnoxiously yuge socket, wasted space. Yuge coolers and cost. Wasted mobo space and cramming of components that are going to roast each other if you push the platform to its PCI-E and dimm slot limit...

And for what? I'd love for someone to detail how you scale threadripper on x399 to 32/64 without a gerrymandered mess on the Zepplin die PHY interconnects.

>>61590219
You don't have any point tho. It's all "this is not what I personally envisioned." If you had EVER ordered a server, you'd realize riser cards are a necessity. Pure trash, Pajeet.

>>61589551
if it's 8 + 8 why the hell amd waste money glueing the other 2 dies?

>>61590277
Maybe yields are so good it's cheaper to throw working dies away than create another socket.

File: 4d9.png (124KB, 382x491px) Image search: [Google]

124KB, 382x491px

>>61590219

>because I say so
Isn't a valid argument

>>61590277
Because it comes from the same assembly line as EPYC obviously, it's probably cheaper.

Zeppelin dies are incredibly cheap to produce, and who knows if AMD will enable a die latter on.

>yfw unlock cores is enabled

>>61589933

See pic on :
>>61590062

The DDR slots are hardwired to each die. So, 4 dimm slots go to 1 die and the other 4 go to another die. If what was said is correct (AMD said its only 2 active dies) then scaling beyond 16/32 on x399 is not gonna happen and you just have a silly big arse socket taking up space that can't be used and likely dead pin-outs.

Given the fixed nature of the socket pin-out interface to the motherboard, you just can't do anything beyond what is done. The PHY leading from the die to the DIMM slots is the PHY that leads there. If you've set it to connect to two dies, it's set. I'm not aware that you can mux, switch or play with it after its set. So, I don't believe, with x399 set that you can do something like go : 32/64 and only feed 2 dimm slots to a die and thus alter how the DIMMs feed to processor pinouts without respinning a new motherboard..
> Someone correct me if i'm wrong.

So, I'd love to know what in the world is going on at the mobo/socket interface w.r.t to these dead dies and subsequent pinouts...

Also, I'd love to know if the physical location of the disabled dies are random. I'd think not. So, literally there are dead traces for that section of pins on the board.
> Again, someone correct me if I'm wrong.

>>61590219
>If you had counter points to what was stated

Considering the unbridled idiocy of implying that a 180w TDP is unexpectedly high for the 16-core Threadripper when the 1800x has a TDP of 95w and the 16-core TR is just two 1800x dies at a slightly lower base clock (incidentally proving no increased power usage from the two unused dies) I see no reason to waste my time bothering to refute the rest of your inane "MUH FEELZ THO!" FUD nonsense.

In sum: You are a shill. Fucking kill yourself.

>>61590260
I think he's asian, keeps going on about ecelebs, and how AMD is "loosing face."

>>61590399
The best bet is that all pins are wired, and the mobo autodetects which dies are active.

>>61590277
It's called binning. They make 9000 epycs, and then they only need one factory. Then if 2 of the 4 cores don't work, sell it as a threadripper.

>>61590277
Who says they are good dies? They probably can't test them before they glue them together.

If there is a 50/50 chance that a die is good, then only 6% have 4 good dies, 25% have 3 good dies and 37.5% have 2 good dies.

>>61589988
It does. However, I don't believe, given that you have to set where the DIMM slots feed on the mobo pinout, that the positioning of the dies in random. As such, they've likely done a diagonal positioning of the active dies if it is 8+8 / 0+0. That being said, it makes this a hellavuh lot more of a snafu that they know they're putting dead dies on the chip before its stamped.

They'd have to have something quite strange planned to scale this beyond 8+8 / 0+0 on x399.

>>61590428
Yeah, enabling all 4 dies would be piss-easy then.

>>61590399
>silly big arse socket taking up space that can't be used and likely dead pin-outs.
I personally don't care about silly things. As long as it doesn't actually hurt.

>Also, I'd love to know if the physical location of the disabled dies are random. I'd think not. So, literally there are dead traces for that section of pins on the board.
Well we don't know if we'll see a 32 core Threadripper in the future. Or if motherboard manufacturers have been told to just leave half the socket alone.

>>61589194
they had 80% yields last i saw.

File: 1282036994623.jpg (155KB, 550x530px) Image search: [Google]

155KB, 550x530px

>>61590445
>Who says they are good dies? They probably can't test them before they glue them together.
This suggests the positioning of the two good dies may be random

>>61590006
My thoughts exactly. For anything beyond 16/32+, it seems epyc is in order with proper workstation motherboards, form factors, and riser card PCI-E configurations.

16/32 is the limit of the mobos rolling out.

What's worse is the fact that they did nothing to reduce the space threadripper takes up on the motherboard by sticking with 2 solid dies knowing full well they can't scale this beyond 16/32. I'm sure all of this commentary is going to come in the form of questions about threadripper's future come launch time.

>>61590445
80%

>s-socket is t-too big!

Is this really an argument or just nitpicking?
Personally I always wanted a POWER8 size CPU in my case

>>61590551
I just want an EK monoblock

>>61590551
I think this one guy spamming the thread with concern trolling needs to get his head checked.

File: 1491292548989.jpg (3KB, 96x96px) Image search: [Google]

3KB, 96x96px

>>61590217
kek..

The more I think about it, I am starting to believe that this :
8/8 - 0/0 die configuration is false... It just makes no sense unless they're having problems with the interconnect region of the dies. Otherwise, they'd just disable cores per CCX and open up the memory throughput by having 2 dimms per die wiring.

Something smells fishy about letting this e-celeb butcher the chip, show the innards and losely claim he confirmed its 8/8 and 0/0 before anything of value has been published on the processor.

mfw you sense something else is afoot.

>>61590551
Only mITX babies are angry, not like there aren't other motherboard sizes that can fit the socket, probably some wizardry with SODIMMs and using the underside might fit the socket in a regular mITX.

Anyhow, nothing of consequence because these people fap to the thought of having 16 cores in a case the size of a bottle and wouldn't mind throttling at all because there's no way to put a real cooler in there.

>>61590636
You're retarded.

Dumb frogposter.

>>61589162
>>61589229
>Threadripper is Epyc with 2 fully working Dies disabled.
He doesn't actually say that. It's all speculation.

>>61589194
their yields are high, but more importantly because of infinity fabric design, they can source sub-par ccx modules into their cheaper ryzen line, and the perfect ccx modules for EPYC

>>61590636
>8/8 - 0/0 die configuration is false.
That's what he said AMD told him. That's one of the few things in the video that isn't just conjecture.
I believe him and AMD

>>61590277
You've built a factory that makes a specific kind of widget. You know some people on the market will buy your widgets at a 40% profit margin, but most of the market will only pay 20%. How do you make the most money? Market the product at 40%, and market the product again with a "lite edition" sticker on it at 20%. Boom, market saturation and max profits.

"But wait!", you say, "you could make more off of the lower margin sales by cheapening the product a bit". Ah, but you already built $factory which cost $hugemoney. Your floor boys say modifying the line will lower efficiency, and your bean counters tell you it would actually be less expensive to just slap on a sticker and pretend the product is lower quality.

This happens everywhere in business. Brand name food/drugs vs generic, software limited test equipment, signature/designer and discount clothing. Everywhere.

>>61590260
Points are :
> If it is 8/8 and 0/0, X399 mobos can't scale past 16/32
> If point #1 is correct, than this is a big waste of space on the motherboard and causes components to be cramped and heated. Not to mention the joke that will result of the cooling solution
> 16/32 is likely the limit here
> Dead pinouts and foolery to cut cost on manufacturing from binning epyc processors

They hit the 16/32 mark and are providing yuge processing power at an excellent price. This is likely the way they did it.

As the processor is officially detailed, questions are answered, and technical experts detail the implications (if 8/8 - 0/0) is the configuration, you'll see what my points mean.

The fact that you don't and claim shill means you have no technical understanding of the matter at hand. So, you're projecting your weak hand on others. Quite the tell shill.

>>61590698
Is this Nazi symbolism?

>>61590698
He killed it, too. RIP.

>>61590711
It's not his and he'd have to return it anyway.

>>61590729
He still killed it

>>61589462
>>61589698
>>61590062
>>61590219
>...
>..
>...
>...
Jesus christ stop with the fucking ellipses and off yourself you fucking shill. Nobody on this site unironically types like this.

>>61590663
No he said in talking with AMD 2 of the dies are enabled, 2 are disabled

>>61589229
I'm going to guess that the dies were working fine, and in the process of building the Epyc chip they died or didn't pass QC

>>61590711
He knocked out 2-3 capacitors. It can be repaired.

>>61590704
>why is this a problem
>who cares
>boo hoo
>who cares
go buy a $3k incel Xeon if dead interconnects trigger you so much

>>61590428
Can you provide some source sauce on how exactly this functions at the motherboard/socket level especially as it relates to rerouting/dynamically routing DDR channels to different pins? I'd love to understand how this (switching) can occur and what asics/chips/bios do it.

>>61590762
>talking with AMD 2 of the dies are enabled, 2 are disabled
Exactly. No word of them being "fully working" that's just OP's bait

>>61590514
Now people are starting to see the shit show I saw when the lid came off and my comments flowed in my earlier posts. Quite the idiotic way to allow your product to be showcased before launch.

File: APU.png (1MB, 2433x1309px) Image search: [Google]

1MB, 2433x1309px

>>61589462
Do you not know who he is?

ALSO FUCKING APUS HELL YEAH MOTHERFUCKER, NEED MY FANLESS GPULESS ITX BOX

https://www.scan.co.uk/shop/computer-hardware/cpu-amd/2609/2610

>>61590799
Something something all exposure is good exposure?

>>61590811
nigger, those are just bulldozer/excavator APUs nigger hacked to AM4, those are NOT raven ridge

>>61590636
It could be a variety of things on the interconnect or dies though.
>The mcm silicon not being perfect for epyc.
>dies with some failed interconnect silicon
>low volume product so they just send spare 32/24core epycs as they're going to be making millions of those, possibly chips which are too leaky as all the epyc are low speed efficient parts.
>Absolute die failure when mounting to the mcm. They could be assembled in the same order so first two dies go on fine, then die 3 or 4 fails, chip is a TR instead of trashed.

But it has to be the same silicon in use for each TR, the motherboard isn't going to be able to reroute 2000pins on the fly for the 6 possible variations of tr layout.

I don't see a reason why there can't be a silly expensive 32core thread ripper though, yes it won't have the extra memory or pcie slots and the meory channels will be unbalanced, I'm sure there are a few situations where 32cores can be utilised or even just make it a limited edition part for marketing.

>>61589462
People like you are what's wrong with this world and thid board. U just can't accept anything. If you are buying a cpu like Ryzen you are probably gonna buy a seperate gpu so no need for igpu, are u literally retarded, igpus are just a waste of space on the cpu nowadays. You are so fucking stupid, if I saw u in front of me I would fucking rage so hard. U are fucking mentally retarded to be asking shit like that fucking Intel jew, go suck off that 6700 igpu or whatever it is you'll never use it anyway go pay that 400$ for it u fucking retard. I am so pissed rn at how dumb u are, u are ruining this board and just making an idiot out of yourself. Please just stay off the internet in general if u are gonna ask questions like these u fucking attention seeking intel bastard. All u guys want is attention now because amd is the only one doing stuff. I bet you still live in your mum's house and suck ur dad off for that nice 6700 igpu. ITS SO NICE HAVING AN IGPU LOOK AT ME AMDFAGS HAHAHA AMD IS FINISHED NO IGPU, AMD BTFO. SHUT THE FUCK UP I WILL KILL YOU YOU STUPID FAG, STOP POSTING USELESS THREADS LIKE THESE U FUCKING RETARD. U made me so mad I wanna punch the wall and hope there's you on the other side so it would shatter your dumb skull you stupid fuck. I bet you are gonna make another shit thread after this one gets removed because it is SHIT just like your igpu and your shit intel cuck cpu. So fuck off to another board or off 4chan, because u can't handle us or even basic logic like why there isn't an igpu in ryzen cpu's. Can't wait for the release to see you cry and be mad at yourself for spending so much of your dad's money on a shit intel cpu with an igpu u won't ever need. JUST FUCK OFF!! AND IF U EVER MAKE A THREAD LIKE THIS AGAIN, I WILL MAKE SURE YOU WILL NEVER VISIT THIS SITE AGAIN. So bye intelfaggot and I hope I'll never get to see you or one of your threads ever again!!

>>61590399

It really doesn't matter about how AMD bins Threadripper as long as they hit the required feature set that they set out for the platform. Yes, it's a "waste of space" but what binned product isn't?. If you don't like it, don't buy it or step up and buy Epyc where you can get all you want and more.

>>61590848
Nice pasta.

>>61589162
>AMD's 8 core Ryzens have 95W TDP
>AMD's 16 core Ryzens have 180W TDP
Imagine what kind of housefire 32 cores would be.

>>61590522
>My thoughts exactly. For anything beyond 16/32+, it seems epyc is in order with proper workstation motherboards, form factors, and riser card PCI-E configurations.
>16/32 is the limit of the mobos rolling out.
The problem is epyc 32 core is limited to 2ghz, epyc at threadripper clocks would be a housefire but no more than 2p xeons are. So you're left with the 32 core epyc that is only slightly faster than threadripper and has no decent motherboards. Threadripper has 16 deactivated cores wasted, is still slower than 2p xeons though much better value. An e-atx epyc workstation board with huge VRMs and either threadripper workstation versions or unlocked epyc 1p would have been the easiest product ever requiring only a motherboard design to make the product.

The workstation market would eat up 32 highly clocked cores + 128 pcie lanes but as usual AMD has zero business sense.

>>61590829
Me again, the reason I can see for not having 32core chips is if the x399 specs don't lay power lines to the normally disabled cores, the interconnect is all on package.

>>61590811
Quad?

File: tdp.jpg (56KB, 1030x579px) Image search: [Google]

56KB, 1030x579px

>>61590403
You took the bait. Now explain to me why this magically matches EPYC's TDP and what will occur if they try to push threadripper beyond 16/32. What happens to Threadripper when it becomes an EPYC chip? Hint : it gets eypc clocks.

So, to my overarching point : Threadripper doesn't scale beyond 16/32 on X399 ATX/EATX. So, what you have is a big ass fucking socket taking up useful space, dead pinouts, and two dead dies on it because (muh reuse/binning).

AMD could clarify this is not the case and outline a deeper roadmap, but they've let some e-celeb hack their processor up before they've done an official launch.

Whose the shill friendo?

>>61590876
180W tdp @2.1ghz

>>61590811
You're better off buying a Celeron and a GT1030 than that shit.

>>61590876
Well the TIM isn't mayo, so you know

>>61590828
Look at the TDPs, no way those are 28nm last gen APUs

last gen was over 95W for 4 cores

>>61590918
They've been available through OEMs only for months now. http://cpu.userbenchmark.com/Compare/AMD-A12-9800-RADEON-R7-12-COMPUTE-CORES-4C-8G-vs-Intel-Pentium-G4560/m178624vs3892

>>61590857
>It really doesn't matter about how AMD bins Threadripper as long as they hit the required feature set that they set out for the platform. Yes, it's a "waste of space" but what binned product isn't?. If you don't like it, don't buy it or step up and buy Epyc where you can get all you want and more.
There is no epyc to buy, overclocked threadripper will likely beat the 32c epyc because it's locked to 2ghz. Completely ignored the workstation market because their product planers are clueless.

File: HR 22.jpg (150KB, 602x635px) Image search: [Google]

150KB, 602x635px

>>61590900
I don't want any GPU, I want one big ass cooler on my APU to run it fanless in my ITX case

If those aren't Ryzen APUs though they're trash

>>61590938
RIP

>>61590951
you know that EPYC can turbo to 2.7GHz on all cores, and 3.2 on <12 cores right?

>>61590951
Do you really not understand why someone would want Epyc over Threadripper? They're different markets. Most people don't even need more than an R7 1700

>>61590964
Yeah it's basically a G4560 with meh single core, but on the other hand the iGPU in the A12-9800 will hand the G4560 it's ass on a platter.

>>61590918
The excavator cores were rimarily designed for mobile so they're pretty power efficient. There's been little coverage on them but I remember some guy getting 4.8ghz on the 9800 with little issue.

>>61590973
>you know that EPYC can turbo to 2.7GHz on all cores, and 3.2 on <12 cores right?
No, he doesn't.
He's pulling this all out of his ass based on Intel's definitions and history.

File: 1494698005793.jpg (92KB, 447x650px) Image search: [Google]

92KB, 447x650px

inb4 somebody figures out how to enable all cores phenom-style

>>61590871
>fight fire with fire

>>61590986
The issue is that really no one can recommend them, unless you really want to get on AM4 with a $50 CPU that's shit

Ryzen APUs should actually be pretty good though, bringing back $300 console killing APU rigs

>>61591013
>24c for 1k usd.
Oy vey

>>61590894
>So, to my overarching point : Threadripper doesn't scale beyond 16/32 on X399 ATX/EATX. So, what you have is a big ass fucking socket taking up useful space, dead pinouts, and two dead dies on it because (muh reuse/binning).

Why should it scale beyond 16/32? When did AMD announce that it would or could? What do you need that space for on a botherboard that you would use on a workstation and not just buy an EPYC cpu/board if you needed more I/O? Do you realize that AMD would have to charge more for this product if they couldn't make it this way? Why should AMD cater to your specific needs, as opposed to offering the rest of the HEDT market an unbelievable value?

>>61591013
see the capacitor on the ring around the package? see how there aren't any on the outer ring, and the ones on the inner ring are covered?
the only way to turn on these other cores is to install the proper capacitors on the proper slots, which is just impossible for basically everyone

>>61591063
>buy 1 epyc 32c/64t
>delid
>??????
>profit

>>61590951

>overclocked threadripper will likely beat the 32c epyc because it's locked to 2ghz

Disregarding turbo speeds and whatnot for a minute, why does anyone needing servers value single core performance more than threads? Enterprise will choose double the threads any time over nearly double the clockspeed, as in the case here.

>>61590894

Who makes you the expert on what is possible on X399 motherboards? You're not a motherboard expert and AMD could easily in the future be able to offer something more with a 7 nm shrink and stuffing more cores per CCX.

>>61591080
nice job, you killed a $4000 cpu

>>61590465
32 core threadripper is epyc. So, we know exactly what it is down the road and we know what clocks/TDP match such a scaling. There is no magic pixie dust. These are server grade processors and core complexes. Thus, when you scale them, you get server grade clocks.

Threadripper comes in at an excellent price/performance. However, I think a ruse is occurring suggesting this can go beyond what has been detailed. Beyond the lack of physical room on ATX/E-ATX for more DIMMS/PCI-E lanes, is the conundrum of reduced clock speeds.

7nm could unlock things w.r.t to increased clock speeds but you're still facing lack of DIMM/PCI-E lane to match the increased core count.

This is where skylake-X possible re-rears its head and why intel held back on its 12/16/18 core offering. A game of chess is occurring at the hardware offering level and I'm neutral to all parties. I see what I see on all platforms and will make decisions accordingly.

Expect a strong salvo to come as a result of this fumbling IMO. Although this has put a bad taste in my mouth, I look forward to technical review outlets like anandtech grilling AMD on these details.

>>61591063
Bullshit.
Why waste time connecting all dies in the first place, if you know that 2 of them are not good? You need capacitors to test all dies in the first place.
Put them on, test, and remove them after? Sounds stupid.

File: AMD-EPYC-Four-Die-with-Infinity-Fabric.jpg (36KB, 957x539px) Image search: [Google]

36KB, 957x539px

>>61591111
>>61591063
>>61591080
There is no need to dellid a 4k CPU. pic related.

>>61589162
>2 fully working Dies
seems a bit optimistic desu

if it's true it's going to be Athlon X3 allover again

>>61590973
It's still not enough for 2x the price, if I pay $2000 for your companies high end CPU is it really too much to ask to be allowed to select voltage and clock speed like you allow people who buy a $200 ryzen chip do? Intel did this crap because they charge ridiculous amounts for higher clock speeds on xeons, AMD doesn't have a highly clicked epyc product to sell in the first place. AMD opterons even use to be able to overclock the last time AMD was relevant. Most other CPU arch don't even artificially restrict CPU clocks either and will happily run at maximum speed as long as you keep temperatures within requirements. It's all just X86 only bullshit intel pulled when they forced AMD out of the server market.

>>61591184
Softblock or hardblock? We need to know.

File: the90sarecalling.jpg (63KB, 655x1000px) Image search: [Google]

63KB, 655x1000px

>>61590639

>>61591177
You're a big chip

>>61590811
Those are still Bristol Ridge
Pretty decent budget chips, my laptop has a 9600p

>>61590759
if you're actually bothered by...this you seriously need to actually get checked for autism

>>61590759
This. And also stop using exclamation points. How fucking new to the internet do you have to be to not now these things? "Redditspacing" wasn't actually redditspacing until what I assume was a couple of months ago when reddit diaspora found another way to distance themselves from other diaspora.

AMD is sandbagging intel. Next threadripper will have 24c/48t @ 3.2ghz 260W TDP

File: RYZEN_THREADRIPPER_DELIDDING_(en)_-_YouTube_-_2017-07-28_00.43.24.png (12KB, 231x69px) Image search: [Google]

12KB, 231x69px

he is monitoring this thread

>>61591109
>Disregarding turbo speeds and whatnot for a minute, why does anyone needing servers value single core performance more than threads? Enterprise will choose double the threads any time over nearly double the clockspeed, as in the case here.
It's common for workstation tasks, you alternate between poorly threaded tasks and needing as many cpu cores as possible. 3ghz turbo on epyc 32c 1p isn't fast enough, AMD forces these users to either buy a cheaper threadripper they don't really want or dual xeons if money is no issue. There's no epyc cpu or motherboard well suited for workstations. They're ignoring a market that will happily pay 2k per CPU where epyc would be ideal if they released a suitable version.

File: 2.jpg (703KB, 2048x1536px) Image search: [Google]

703KB, 2048x1536px

>oh no they delidded a chip... this is a bad look...!
>... they're... get this... reusing epyc dies...
>w-w-what about all that motherboard space...!!!!
>i'm............... s-s--s-scared.............

loving every laugh from this absolute fucking retard

>>61591353
Gold Plated. OYYYYY VEYYY

Theres literally nothing wrong with giving it to der8auer. Threadripper is an enthusiast CPU and hes popular with enthusiasts while being in the news recently criticising x299.

>>61591129
Please detail when and where this "ruse" of "it can go beyond what was detailed." started.

If I were intel this would make me a little nervous about selling an 18 core on HEDT

>>61591258
The CPU performance is trash though

>>61589194
https://www.youtube.com/watch?v=ZoVK6rJR5VE

>>61591323

Enterprise focuses on "embarassingly parallel" tasks, and while single core is important, who are you to say that "3ghz turbo on epyc 32c 1p isn't fast enough" when they have already shown that they can beat dual Xeons in the same price range in those tasks? And what does a "suitable version" of Epyc look to you?

Server chips can be overclocked but given the constraints of how the server thermals are, you would be silly to even try and possibly damage your hardware without the thermal headroom.

AMD and Intel WILL NOT give you custom silicon unless you buy in the thousands and give them support contracts in the millions and your company has some reputation in the server world or is one of the big tech companies. An enthusiast or some small time entrepreneur needing 10 servers or something won't get them.

File: 1496325650300.jpg (200KB, 1919x1080px) Image search: [Google]

200KB, 1919x1080px

>>61591232
>You're a big chip
FOR YOU

File: RkGZXrV.jpg (174KB, 1166x925px) Image search: [Google]

174KB, 1166x925px

>>61591518
>threadripper III

>>61589248
yeah call me when intel use something as perfect as indium

>>61591518
>>61591538
BIG

File: 1496953720156.png (185KB, 409x409px) Image search: [Google]

185KB, 409x409px

>>61590811

>>61589248
>real paste
Jew cum is not real paste.

File: wideripper.png (370KB, 800x627px) Image search: [Google]

370KB, 800x627px

>>61591518

File: Amada 1.jpg (4MB, 4777x2361px) Image search: [Google]

4MB, 4777x2361px

How can intel compete?

>>61591129

Intel has yet to even match the X399 platform in core count or DIMMs and PCIE lanes. They are also missing ECC support and need RAID keys because they were scared of cannibalization of servers. August 10th is coming soon and Intel's response is too slow since nothing has come forward about any of Intel's >10 core processors.

File: 1494913356043.jpg (195KB, 577x534px) Image search: [Google]

195KB, 577x534px

>>61590829
If you look at how the EPYC line is configured with extended core counts from 8 all the way up to 32, you see where they're using the Dies that have no interconnect issues. Thus, economically the ones used in Threadripper had interconnect issues and couldn't be used in any configuration on EPYC.

> But it has to be the same silicon in use for each TR, the motherboard isn't going to be able to reroute 2000pins on the fly for the 6 possible variations of tr layout.


The rest of your commentary is quite sound desu and is pure bread. Thank you for detailing this and confirming my thoughts on a technical level. I was getting slammed by a bunch of boneheads who have no technical understanding and wasn't sure someone with technical sense would respond.

You responded. Your commentary is sound and I thank you for that.

>>61591616
Is that young Lisa Su?

File: TR.png (200KB, 681x695px) Image search: [Google]

200KB, 681x695px

THREADRIPPER SI PRE-ORDERS
https://www.cyberpowerpc.com/page/AMD/Ryzen-Threadripper/

Can anyone spare about $5000?

>>61591665
You've yet to detail why any of this is an issue.

File: jontron-puking-o.gif (522KB, 320x180px) Image search: [Google]

522KB, 320x180px

>>61591699
>OEMs

>>61591699
>prebuilts
absolutely disgusting

>>61591699
None of them are using a stock AIO cooler, though maybe that's to be expected anyway.

>>61590876
See lower clocks.. The same lower clocks intel is hitting in their detailed 12 core. No one can defy the laws of nature of physics. 14nm has its limits and these processors are at it. What this shows though is AMD can't scale beyond this limit nor can intel. 16/32 is quite a comfy count. Beyond that, you're going to start facing headwinds which is why this is most definitely and likely 8/8 and the limit therein. If it were 4/4/4/4, they either wouldn't be able to get the clocks threadripper has or TDP would be through the roof.

With that being said.. were back to the : WTF socket size. It's not going beyond 2 dies without headwinds. So, here we are with this yuge ass socket and nothing to do w/ 2 dies worth of space. This could change down the line but those pinouts on the mobo sure aren't.

>>61589162
hmmm i wonder where i saw such a thing in the past....

oh yeah athlon xp thoroughbred b

>>61591207
AMD tends to go with softlock usually. You could unlock a large number of ATI gpus, and even recently there were some R9 290s that could be unlocked to 290Xs or RX480 4GB that could be unlocked to 8GB.

>>61591787
Cyberpower's stuff comes default with a 120mm AIO, which can't cool threadripper, they also likely have the wrong PSU as the default, I don't even think most of their PSUs are going to have 2 CPU 4 pins

Wait isn't X399 2 8 pin CPU connectors?

File: images (17).jpg (11KB, 299x168px) Image search: [Google]

11KB, 299x168px

>>61590062
Muh infinity glue.
Stay mad, performance will be shitty considering 4 dies with broken ccx inside. Will be nice to have such latency accessing l3 like mad from other dies

>>61591890
Asscock mobos are 8+4, Zenith Extreme is 8+8.

>>61590883
Yep, there is a lot of detailing and questions that are going to arise as a result of this... Good answers better exist.

> AMD has zero business sense
They're playing a certain hand here and it relates to binning/etc. My only question is. given the other hands they could have played why? It appears they are unaware of what lurks on the horizon as do many.

Interesting times nonetheless.
Marketing and upper level executives had to give the approval on letting this detail come out like this.
So, we'll see soon what in the world they were thinking.

>>61590894
what is the point to scale bigger than 16c? its a fucking HEDT 16c is more than enough for everything..

File: intel ceo JUST.jpg (119KB, 1000x1000px) Image search: [Google]

119KB, 1000x1000px

>>61591914
>unironically using Intel marketing slide shitposting terms
>pretending to know how l3 access works across dies when there's connections to and from every core

Go away, Brian.

>>61591914
>threadripper mobos supports 3600+
Fuck off brazilian monkey.

File: regular goy.jpg (16KB, 230x219px) Image search: [Google]

16KB, 230x219px

>>61591921
>Sixteen cores? What you need more than eight cores for? Here, take this quad core and shut your mouth.

>>61591914
Hi Brian.
Please neck yourself.

>>61591921
Because what scales on 16 cores also scales on 32 cores.

>>61589162
>fully working
No this is a common tactic among hardware manufacturers. Sell a CPU with defective cores as a lower core variant.

>>61591054
> Why should it scale beyond 16/32? When did AMD announce that it would or could?
Do you deny that this is the perception leading into launch? Do you understand what occurs when this obvious question is asked and then the subsequent follow up : Then why the big ass socket?

> What do you need that space for on a botherboard that you would use on a workstation and not just buy an EPYC cpu/board if you needed more I/O?
That's a question for AMD to answer. I have no doubt they're going to get pressed on it as they should.

> Do you realize that AMD would have to charge more for this product if they couldn't make it this way?
I do and they are going to officially have to state this soon to justify why the socket is so big

> Why should AMD cater to your specific needs, as opposed to offering the rest of the HEDT market an unbelievable value?
I never said they should. This is an incredible value and now we know partially why. Soon, AMD will have to confirm this and what their thinking was... Not just for now but down the road w.r.t x399 as a socket.

>>61591972
yeah and then amd wont sell any epyc cause no one will buy a 40c+ cpu considering that the software is PER CORE on the server market..

also about the AM4 apu's
http://anandtech.com/show/11669/amd-releases-bristol-ridge-to-retail-am4-gets-apus

>>61592024
It's usually per-socket.
Kill yourself.

>>61592077
its per core most of the times at least in the software that people ACTUALLY use

>>61592115
No, it's per-socket.
Please kill yourself.

>>61589162
So why add the extra 2 dies and disable them? just wasting money?

>>61592013
Precisely, in a cogent and segacious fashion, detail when and where it was implied TR scaled as *you* imply. Your main argument seems to be "le future proofing."

>>61592128
Read the thread faggot

>>61592128
It's a fucking ES.

>>61592147
>>61592147
why even add the dies, there wasnt any answer in the thread actually.
If its ES and they just used EPYC and disabled 2dies it would make sense.

>>61592189
we know that their yields are above 80% so they dont really care
we might even see binned TR after some time

>>61592189
the only way to know if they're fully working (all the interconnects, memory channels and other stuff) is to slap it on a package and test it

>>61591914
THREE THOUSAND FORTY SIX :^)

>>61592121
maybe it depends on the situation?? server market sure is pretty big for everything to be one way or another

>>61592220
sure i understand they dont care, but why even add them then? if they are disabled just leave the slot empty. Or actually you know just fucking use the design with any unnecessary slot for die or dies.
Really amd always manages to fuck something up. Just so you know im all amd on cpus, but 100% they will be slammed in reviews if TR doesnt destroy intel offerings and one of the things they will be slammed for is 2 disabled dies on the cpu.

>>61592300
its cheaper to just put dies and disable them than to cut them down for amd
the real question here is how amd managed to actually disable them
1)is it like phenom?
2)is it like athlon xp?
3)they actually removed capacitors?

if its the 1 or 2 that means someone will find a way to enable them
if its the third well...

>>61591172
Maybe they put defective modules on to make sure the heatsink/IHS aren't under uneven pressure, it's a big package for 2 dies.

File: 1494546940624.jpg (21KB, 417x405px) Image search: [Google]

21KB, 417x405px

>>61591109
Don't be the kind of person who doesn't know jack shit so critiques anyone who does and then goes on to appeal to authority for one that doesn't exist nor has spoken yet.

If you've actually had formal education in the field of study related to this or just common sense, you wouldn't be making such brainlet responses that expose how limited your real understanding is.

So calm down, all you're doing is impressing other brainlets like yourself who get off on someone who brings nothing to an argument besides critique those making critical comments. There have been well informed responses to my commentary substantiating what has been stated. You, by comparison appear to be the detractors.

Don't worry about who I am. Worry about the content of the my commentary and whether or not you can contribute in some way to the general discussion.

Now, since you want the spotlight...
> You're not a motherboard expert and AMD could easily in the future be able to offer something more with a 7 nm shrink and stuffing more cores per CCX.
7nm doesn't fix a motherboard that has fixed wiring from the PCIE and DIMM slots to specific non-fucked dies on the socket dummy... which is why it was stated.. since 8/8 0/0 is the confirmed (AMD) configuration that, 16/32 is the limit.

> mfw a brainlet who knows fuck-all claims he can judge whose an expert
Know your place !

>>61591616
upload pls, i need those amadas

>>61592412
https://drive.google.com/open?id=0B-JeQd4atb7vYzZmSWhpeHVad00

>>61590087
That motherboard is sexy as hell.

>>61591616
well they did try
https://www.geek.com/geek-cetera/intel-asks-japan-to-personify-its-computers-blissfully-unaware-of-the-inevitable-1626366/

and as it was obviously they failed

>>61591665
Everything I've said was purely conjecture though, there's a chance it's all wrong.

But it's not bad to think of TR as EPYC-LITE, the only wastage is the extra size on the socket and package, none of those extra pins even need trace connections on the motherboard and the size it takes up.

The issue you (was it you?) brought up was the lower clock speed on 32core parts, if yoy do some napkin math 2.7ghzx32 = 86.4 arbitrary performance and 3.6ghzx16 = 57.6. That's 50% more arbitrary theoretical performance from epyc, you bring up the single core turbo speed for workstations as well, which is probably why epyc isn't being marketed as such very often and remember it's only been with skylake server and turbo boost 3 that even intel has had high clock speeds on single cores for xeons, which are also not marketed to workstations for the most part either.

File: 1465702164586.jpg (277KB, 1616x2048px) Image search: [Google]

277KB, 1616x2048px

>>61592438
Thanks anon

File: 148952094964.jpg (53KB, 600x800px) Image search: [Google]

53KB, 600x800px

>>61592395
>rips Anon for "appeal to authority"
>listen to my words, I'm an expert

Does your """"expert, educated"""" commentary consist solely of "it's not future proof"?

File: 1449446148255.gif (2MB, 380x238px) Image search: [Google]

2MB, 380x238px

>>61590087
I need dis

>>61592370
so you saying its was cheaper to just disable them and produce proper design and not castrated cuck

>>61591377
The ruse is in what's implied and what gets strategically leaked out and detailed vs what doesn't. Do you think this e-celeb delider didn't have the approval from AMD to unveil these details? He did and they did it for a reason. The reason being the false perception that has lingered about there only being 2 dies. So, they've engaged in controlled/calculated demolition before the launch to reset expectations.

The ruse is in the implying/lingering belief that there is a future for the socket to expand to EPYC whereby current mobos can't. So, this is peak threadripper on current mobos with room for speed increases/better power profits come 7nm.

Marketing isn't stupid nor am I.

>>61592438
thanks pal

>>61590767
>He knocked out 2-3 capacitors. It can be repaired.

He heated it up so high that the solder melted. That's what, 3-400C? The CPU is dead either way.

And before you ask how they applied the solder in the first place, they use indium solder (way lower melting point) with a special type of heating device so the chip is exposed to the required heat as little as possible. Not a fucking heat gun.

>>61592492
i would have done the same with those yields senpai

they literally can produce them like a chicken laying eggs

File: images(7).jpg (12KB, 275x183px) Image search: [Google]

12KB, 275x183px

>>61592463
>adding cpu speeds together

>>61591184
The dies that have core issues but good interconnects are used in the EPYC line which is why EPYC can drop all the way down to 8 cores yet still support 128 PCI-E lanes.

Threadripper contains the dies that had interconnect damage .. Thus, no matter if the cores are good, can't be used in an interconnected die configuration.

Maximum reuse

>>61590883
and instead of doing that stupid thing, they put out a platform that still eats Intel's HEDT lunch while sharing manufacturing with both their desktop and server line.

Acting as if this is just a total loss is quite possibly the stupidest stance i've ever seen. They had these chips lying around that were useless since they had 1-2 entire dies dead after being MCM'd and instead of throwing them out they made this.

>>61591172
So they can make multiple product lines in the same factory line. It makes manufacturing easier than if they had a completely separate one.

>>61590070
I'm running a Phenom II X4 for 10 years and this shit is going nice.

>>61591617
Watchful eyes are looking for specific things...
> Intel has yet to even match the X399 platform in core count or DIMMs and PCIE lanes.
There are things that lie beyond this that matter

Intel has seemingly delayed the detailing of their higher core count offerings for a reason. With this detail coming out in the fashion it did, I can see why. I have no doubt that their salvo will quickly follow AMD's official launch.

File: images(32).jpg (17KB, 285x177px) Image search: [Google]

17KB, 285x177px

>>61592497
There two separate lines: TR and Epyc, how was it implied that you could upgrade to epyc later?
>>61592640
Jesus the drama and intrigue is hilarious with you.

which x399 motherboard should i get out of the ones that have been unveilved, anons? :)

>>61589162
>>61591498

If we already know what CPU it is, why is he covering the front with black tape?

>>61589462
>▶
>desu

Choke on a dick

>>61590070
i dont think you know much about indium eh

>>61592640
You really think intel has something hidding?

>>61591920
>They're playing a certain hand here and it relates to binning/etc. My only question is. given the other hands they could have played why? It appears they are unaware of what lurks on the horizon as do many.
>Interesting times nonetheless.
>Marketing and upper level executives had to give the approval on letting this detail come out like this.
>So, we'll see soon what in the world they were thinking.
No it's just them being stupid. Back in the single core days there were few differences between the phenom and opteron it was even common for people to pay more for the opterons as they were binned better. Then AMD came out with socket g34 and mangy cores 12 core opteron, people had engineering samples with unlocked multis and were overclocking them to 3ghz, AMD locked the retail ones and said no one needs more than 12 cores at 2ghz anyway, buy the 6 core x6 phenom if you want higher clock speeds. Intel then proceeded to mercilessly rape them in desktop, server and workstation sales. EVGA actually sold a 2P xeon motherboard that supported overclocking and high ram speeds, intel only put an end to it when they had completely crushed AMD, at one stage the motherboards were going for 2 or 3 times their original price, the overclocked xeons were the fastest workstation you could possibly build until the locked xeons edged ahead generations later.

It's just idiocy that comes with the X86 duopoly, if you look at IBM power, ARM servers and other CPU arch they almost never include any sort of artificial clock speed limitation on products, generally if you can cool it you can run it at the highest speed they sell that core design as.

File: 1494387422977.jpg (181KB, 888x1120px) Image search: [Google]

181KB, 888x1120px

>>61591722
This is a day of habbenings. This is one of many habbenings. It matters and matters for reasons of old and reasons that will go unspoken.

>>61592540
I'm just using the numbers in an arbitrary sense to show that the epyc is up to 50percent more powerful than TR as the previous anon seemed to think it was twice the cores at half the clock speed giving no net gain.

>>61592395
>since 8/8 0/0 is the confirmed (AMD) configuration that, 16/32 is the limit
maybe for zen, but zen2 may get 10 core ccx so you would have 10/10 0/0 for a total of 20/40

>>61592779
You mean 10core Zeppelin successor?
That's impossible.

File: 1454819517578.jpg (52KB, 780x688px) Image search: [Google]

52KB, 780x688px

>>61592756
ehhh some people seem to think moar corez for """enterprize"" is better, but sometimes, GHz matters just as much, and ""enterprize"" sometimes needs something over 2.6ghz, but can't afford it bcuz Jews, and have to settle for more, slower corez.

>>61589162
>with 2 fully working Dies disabled
you don't know if they passed all tests under heavy load

probably not

>>61592142
> Precisely
> cogent
> s*agacious fashion
OK...

> While many rumors swarm about your flagship product..
> One being that it is composed of two dies....and many such implications therein like : the ability to scale it to 4 functional dies and utilize all of the pins on the socket....

Maybe its not the best idea to resolve such rumors by letting an e-celeb hack your flagship product to bits before you do an official and professional launch of it.

But hey, maybe your boneheaded marketing team thinks this slick controlled demolition pre-launch via a third party you can distance the company from is the correct way to fix an incorrect rumor that has been festering and that you have capitalized on to win attention for your processor..

But maybe, given how :
> Tight lipped you've been on the product
> How strictly you control product sample information leaks via NDAs

Maybe people know exactly what stunt is being pulled by the feigned amateur hour deconstruction of your chip. Maybe your marketing team thinks this stunt was a good approach. Maybe they don't realize what kind of shit storm this is going to kick up days if not weeks before you officially get in front of it. Maybe they've told you bad publicity is good publicity. Maybe the game has changed. Maybe they're wrong.

My argument is not an argument because I have not nor will state what is the nature of my critique.

Hope that was, at least, s*a*gacious enough for you.

>>61592995
...so no critique, just bloviation?

>>61592514
The Indium is the solder and it melts at 160°C
and actual solder quality heat guns, like the guy used on the video can give you really specific temperatures
He also never got past 200°C
Why are you spreading bullshit when you didn't even watch the video?

Im putting my money on those extra dies not being lazer disabled so maybe we will see them unlocked on lower clocks.

>>61592995
Jesus fucking Christ. AMD at no time suggested, implied or claimed that Threadripper could have more than 16 cores and 32 threads, nor would they have any reason to do so when Epyc already fills that niche.

Just fuck off already with your fabricated muh feelz nonsense.

File: boi-cage.gif (1MB, 300x150px) Image search: [Google]

1MB, 300x150px

>>61592514
Controlled third party demolition ahead of launch approved by AMD with the purpose of addressing the 2-die perception and expansion to 4-die possibility therein.

Effectively, they've disclosed while not being tied to it that the buck stops at 2 functional dies and 2 fucked dies w/ threadripper. Having used a back channel, they can manage the fallout without tainting their flagship branding of threadripper.

He knew exactly what he was doing and had approval from AMD to disclose it. I wouldn't be surprised if he was instructed to do it.

> Ohhh woops, processor is blown. So, we'll never know besides AMD's word..
> Ohhh and ofc I can't go any further w/ the video because muh sample unit is dead .. because year, I just do this kind of shit to my sample units and disclose it under (strict) NDAs.

File: 1495495069551.jpg (7KB, 200x200px) Image search: [Google]

7KB, 200x200px

>>61592547
Takes the shine off of it doesn't it?
Now, what do you think the marketing group did when faced with this reality?

>>61592995
Thank God you pointed out a spelling error, now I know I should listen to you because you are educated, wise, and vastly superior to a lowly pleb like me. No matter that that was neither cogent nor segatious. teehee

File: 14821.png (21KB, 200x194px) Image search: [Google]

21KB, 200x194px

>>61592711
I know they do. The only question is how effective it will be. Look for the salvo soon after Threadripper's official launch.

>>61591914
Oh my god you're literally just spewing Intel marketing slide garbage

File: 1497296900748.png (57KB, 500x382px) Image search: [Google]

57KB, 500x382px

>>61592713
Thank you for this most valuable bread and history desu.

>>61593309
New shills on the block. https://rbt.asia/g/search/image/zlYXrSRpO_SPE-3jRon0vw/

File: 1497283296055.jpg (246KB, 1280x1233px) Image search: [Google]

246KB, 1280x1233px

>>61593097
>>61593193
>>61593245

My work here has come to a close.

>>61593211
>fallout
There was no and will be no fallout. There is not now and never was anything in AMD's information about Threadripper that suggested anything more than 16c/32t was possible. The motherboards were all from their initial announcement said to have 64 PCIE lanes, consistent with 2 dies being enabled, and incompatible with the assumption that the same motherboards would somehow also support 128 PCIE lanes.

Fuck right off you LARPing faggot

File: TB2f4vIvrJmpuFjSZFwXXaE4VXa_2341477237.jpg (335KB, 1060x1047px) Image search: [Google]

335KB, 1060x1047px

>>61593261
Oh yeah, they're definitely hiding their sekret speciulz chpzes just waiting to attack the HEDT market instead of doing something about getting utterly eaten alive in the server space.

They will bust out the sekretz weapons instead of just rebranding this abortion of a Pentium Gold into at their 18-core at 2ghz so it doesn't literally melt the socket on their joke of an x299 platform, right?

>/g/ bitches about stuff

Meanwhile all cpus are compromised by some national agency

>>61593447
I heard the agent assigned to monitor you killed himself after watching you shitpost on /g/ for the 4th year straight.

File: 503.jpg (246KB, 881x848px) Image search: [Google]

246KB, 881x848px

>>61593378
so what your entire blog post consists of: 4 glued together dies!!!, no upgrade for pajeet from tr to epyc when he saves enough rupees, DEAD INTERCONNECTS, muh shameful product review. Go to bed pajeet.

>>61591699
hmm let's see:
>AMD Ryzen Threadripper 1950X Processor [Pre-Order]
>GeForce® GTX 1080 Ti 11GB GDDR5X
>64GB DDR4/2400MHz RAM
>ASRock Fatal1ty X399 Taichi Motherboard
>1TB SAMSUNG 960 EVO M.2 PCIe SSD
>5TB SATA3 7200RPM HD
>LG 16X Blu-Ray Combo Drive

one does not belong here

>>61593485
Rlly? phew that asshole was becoming annoying lately

>>61593211
Clearly this is an abortion of CPU.
Bet it perform like shit

>>61589162
Yeah. Didn't cores get re-enabled on GraveDigger or something?

>>61590767

Send it to Louis, would make for a comfy video.


>>61592514
>He heated it up so high that the solder melted. That's what, 3-400C?


U wot, you mong?

>>61593498
>2400Mhz RAM

For what purpose?

>>61593498
Is it the slow RAM or am I missing something?

>>61592024
Per socket for most of the software I've bought in last five years

>>61593498
It will not support such "faster" memory on such amount on toy ripper.
Print this 64gb will not work on this platform with 2400

>>61594263
t. brian

I'll invest in thread ripper once I can run 32gb+ of ddr4 at 3200+mhz

>>61591914
Stay butthurt forever intelcuck.

File: 1492161489160.png (172KB, 400x382px) Image search: [Google]

172KB, 400x382px

>>61591177
>>61591232
>>61591518
>>61591538
>>61591564
>>61591580
This is the phenotype ashkenazi girls crave

B I G
  I  
     G
[\code]

>>61590087
>no SODIMMs
No.

>>61589162
>>61589173
>>61589206

HOLY SHIT NO WONDER IT'S EXPENSIVE YET CHEAPER THAN INTEL EQUIVALENTS
THEY USE GOLD
THAT YELLOW SHIT IS GOLD
ACTUAL GOLD
JEWS BTFO

>>61595257
OY VEY

>>61595094

Yeah, how come? Can TR run ECC?

>>61595551
>Can TR run ECC?
Yes.

>>61590399
Okay retard. The pins on the board are mostly VCC, DDR and PCI-E pins. Threadripper has 64 PCI-E lanes, which is the maximum supported by both threadripper and epyc CPUs. Epyc has an additional 64 PCI-E lanes for socket to socket communication and those are dead on threadripper and single socket Epyc. Those pins will also be dead on single socket boards.
All memory channels are active. So the only dead pins on the socket are socket to socket PCI-E lanes. All the dies on the threadripper CPU are gonna be partially active. No die is going to be 100% turned off.

>>61595580
WUT?
WUUUUUUUUUUUUUUUUUUUUUUUUT?
Single-socket EPYC still has all 128 PCI-E lanes.

>>61595603
Does it?

>>61595641
Are you fucking retarded?
Like holy fuck how can one man be THAT stupid?
Just fucking google it.

>>61595654
Why would I google it when I can have you tell me.

>>61595677
>spoonfeed me
(You).

>>61595686
Well you did, so you can't exactly say my strategy is flawed.

>>61595580
Just a nightmare of uncompleted shitty dies
Mean while Intel has perfect dies

>>61595711
Hi Brian!

>>61595711
Perfect in what way? The new architecture is a mess. Atrocious rushed launch. Intel is not going to be viable until they launch a successor for Skylake-X which fixes all the problems.

>>61595742
>a successor
Cascade Lake just adds Apache Pass support and some other small fixes.

>>61595757
Then we'll have to wait until the successor for that for a non-shit arch from Intel, I guess.

>>61595562

Neat. DDR4 SODIMM ECC is fairly rare, no?

>>61595603
>>61595641

No.

>>61595765
Yes/

>>61589462
If a core in underpreforming or broken they will disable it and then sell it. threadripper = broken EYPC.

>>61593378
Making everyone loathe faggot frogposters even more than they already did?

>>61590514

That can't be. They have to know which pins on the socket to wire to the active dies for the memory at least.

Maybe they are just using known bad dies to use as filler in the other 2 positions, allowing them to reuse the epyc package. Depending on threadripper demand they can respin the package later to use a different spacer..

File: 29a.jpg (565KB, 723x600px) Image search: [Google]

565KB, 723x600px

AMD is planing to enter the server/workstation market again?

For a while aging opteron platform was the only option if you went with AMD.

>>61596132

Is dis nigga srs

>multiple dies
Remember when everyone laughed at Pentium D?

>>61596132
>AMD is planing to enter the server/workstation market again?
How's the Balrog been lately?

>>61596132
Something about your post smells like Intel shill. Maybe it sounds too much like one of the Intel-sanctioned shitpost presentation slides.

>>61596151
Intel's attempts at MCM are a lot closer to "two CPUs glued together" than AMD's Infinity Fabric and CPUs designed to be in MCM configuration from the start.

>>61596212
Causality only works at the speed of light. The more distance you have the less data you can process.

>>61595580

no way. EPYC has shit memory performance when you go off die. (sorry guys). AMD could never release threadripper with the scheme you describe because some cpus would perform great and some not-so-good. This is way beyond clock binning.

>>61596252
Intel used the frontside bus.

How does AMD's multi die approach negatively affect performance vs a single die? It seems to really bring down the price vs what Intel can do and I'm curious about if it's as applicable in normal consumer lines, such as an 8 core processor using 4 2 core modules.

>>61596343
Memory latency suffers the most. If you have throughput this doesn't matter. If memory latency is critical to your application, it won't be as fast.

>>61596363
So it might not happen in consumer processors because games run better with lower memory latency?

>>61596463
That is one of the issues Ryzen runs into when it has to cross CCX's and one of the big reasons memory speeds and timings improve Ryzen's performance dramatically in games.

>>61596343

I think people are still trying to work this out.

For embarrassingly parallel tasks, like compiling with gcc or running virtual machines EPYC is amazing. As you move to a highly threaded workload where there must be a lot of IPC EPYC performance falls off.

For tasks where you need a lot of pcie-io or a whole lot of memory (with embarrassingly parallel access), EPYC is a natural win.

These are $50k computers, so finding good information on enthusiast websites is thin.

I have read a couple references that the AVX-512 is insane if you can use it and people are getting better than xeon phi performance from normal desktop chips, but have also read that possibly because of the low clocks required when the unit is spun up performance on other code can suffer.

These chips are just very new and different and it is going to take a year to digest. My feeling is that intel designed its cores (with avx-512) with smaller feature sizes in mind, and what we are seeing now is a rough cut. Not that this matters to anybody that has to make a purchasing decision now.

after the fanboy shit dies down, it is gong to be super interesting benchmarking this stuff seriously. These chips are very, very different beasts.

>>61595742
>Muh Intel shitty arc
Meanwhile amd poo ripper
>Incomplete dies leading to several latency across dies
>Infinity glue tied to horrible DDR speeds
>Infinity latency > memory access
>4 faux memory channels ( not a single bus addressed apart )
Shitty clocks at 200+w (remember kids amd always lie about tdp)

>>61596278
You really believe they care? Performance will be universal poor and mediocre latency across boards but fanboys will line to get one.
Don't forget to also get yours Vega fe, just use crossfire X3 to have a decent performance

>>61589162
>2 fully working Dies disabled

This shit is so fucking wasteful. Same thing happened with the RX 480 with the 4GB version having 8GBs with half disable.

They should push firmware update that can re-enable the cores after a few years.

>>61597232
>>61597258
Too bad all of your bullshit means nothing since Epyc crushed Intel's latest Xeons. http://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade

>>61590277

1) You can't test the dies easily preglue.
2) Gluing has it's own failure rate.
3) They're slowing stocking up for 1980X 32 core sku later in the year.

>>61590321

Epyc has a different socket.