T H R E E T H O U S A N D

I want AMDs new stuff to fuck Intel up, but are there any real world benchmarks yet? Things like compile time or transcoding, etc.? Idgaf about games

>>60515848
No shit? What is basically 2 1800Xs has double the performance of an 1800X?

>>60515848
Sauce

>>60515989
going from the score that chip likely has boost and HT disabled just like eng samples of the 1800X

>>60515848
D O L L A R S

File: angry-merchant.jpg (76KB, 427x500px) Image search: [Google]

76KB, 427x500px

>>60515848
DELID DIS YOU DUMBFUCK GOY.

>>60516025
>3631
It's a 3.1base with 3.6ghz boost if I remember AMD's sampling key correctly. And it's reporting back at 3.6, so boost enabled. Hard to say if SMT is.

>>60515848
>T H R E E
>T H O U S A N D
Is this the price?

>>60516054
Pretty sure they are going to be much cheaper. They are literally 2 $330 1700 slapped together with more expensive socket to make.

>>60516054
If it was made by Intel, sure. Actually, they charge twice of whatever the cinebench score is, so you'd expect it to cost as much as a high end Xeon and then some.

>2D3101A8UGAF4_36_31_N

>2
So, 2nd gen ES

>D
Desktop.

>310
3.1Ghz base

>1
First revision

>A8
TDP, unknown in this case.

>U
Socket. Also unknown.

>G
16 Cores.

>A
??, usually cache config goes here.

>F4
B-stepping

>_3631_
3.6Ghz boost, 3.1Ghz base clock.

>N
??

File: 1494113566586.jpg (66KB, 766x668px) Image search: [Google]

66KB, 766x668px

>>60515848
Amada approved.

>>60516107
Yeah. $1250 or so seems likely. Whatever the 2nd most expensive intel one costs.

>>60516107
>2 1700s
Do you have inside info or you just pulled that out of your ass?

>>60516158
Is this based on something or are you pulling it out of your ass?
If that's the actual meaning, thanks for the info!

>>60516376
AMD's OPN code has been deciphered for a while

WHERE DID YOU GET THIS THIS ISN'T ON RELIABLE SITES PROVE IT'S REAL OR ITS A FAKE

Clearly fake

>>60516025
SMT is clearly on, once it hits 4ghz it'll be exactly double

File: LL.jpg (75KB, 500x448px) Image search: [Google]

75KB, 500x448px

>>60515848
I'd rather this baby.

Clearly there's no scaling issue with multiple dies and IF seems to work just as intended.

This is still a early ES though, clock will naturally go up once we reach QS stage

>>60516244
That's how Zen do. While Intel is making Monolithic monstrosities, AMD made Zen with scalability in mind. A CCX is a four core module, each with its own cache. These don't share fp units like Bulldozer modules did, so they're closer to Intel cores in that regard. All consumer chips are 2 CCX's glued together with a super fast interconnect called Infinity Fabric. They are made in 4+4, 3+3, and 2+2 configuration due to yields. This is going to be 4x4 with IF linking the modules together.

>>60516561
2TB of memory possible on a single socket.

Unreal

File: SDTnCWZ.jpg (93KB, 1280x720px) Image search: [Google]

93KB, 1280x720px

Just a reminder where this ES fits in.

File: 1493235143127.jpg (20KB, 600x448px) Image search: [Google]

20KB, 600x448px

>>60516741

>>60516741
Jesus Christ

>>60515848
By the time the retail clocks comes in this is gonna score at least 10% higher.

Also we don't really know its all core turbo, it might just be 3.4 or 3.5, which isn't that far off the 1800X's 3.7 all core turbo

>>60516741
Well the xeons are only at 2ghz

>>60516907
Yeah, and that Xeon has 80% more cores and hasn't even come out yet, it's that $12000 Skylake-EP Xeon.

Also, as I said we don't know the turbo clocks on the skylake, their base clocks are always low but turbo clocks on all cores should be at least 300-500MHz higher at its 205W TDP

>>60516907
And much more cores.
In shit like this, cores matter more than core clock.

>>60516907
That's two Xeons, wouldn't it be quite pathetic if two Xeons with almost twice the cores and using twice the power score the same as AMD's consumer engineering sample?

File: 1495358050273.png (495KB, 1070x601px) Image search: [Google]

495KB, 1070x601px

>>60516741
>>60515848
NOOOOOOO IT WASN'T SUPPOSE TO BE LIKE THIS!!!

>>60517024
Don't worry bro, Skylake-X will bring Intel back..It'll be only 30% slower and won't support ECC but you buy Intel Quality

>>60516957
What are you trying to say? 16 amd cores is matching the performance of 28 intel cores in a multithreaded benchmark.

>>60515848
I want to fuck the shit out of Amada and take responsibility by giving her babies.

File: 34680984.jpg (13KB, 424x335px) Image search: [Google]

13KB, 424x335px

>>60516741

>>60515848
>2D3101A8UGAF4_3631_N

2 - 2nd gen ES
D - Desktop
310 - Base clockspeed
1 - Revision
A8 - TDP(180?)
U - Socket (SP3r2)
G - core count(16)
A - cache config(unknown)
F4 - stepping (still B1, retail units are B2)
3631 - Turbo/Base clock


Here's the 8 core 3.6/4.0 we currently have
ZD3601BAM88F4_40/36_Y

>>60517065
is it going to use less power/run cooler tho

>>60517213
It has a way lower tdp than 2 of those intel chips combined

>not getting a overclockable 32 core at 4.1GHz
>not strapping a industrial fan or six to cool it

What's the matter too scared of 7000+ score?

You niggers told me two CPUs glued together doesn't work, that infinity fabric is too high latency.
Why is the scaling almost linear?

>>60517365
This infinity fabric meme gives me a boner.

>>60517365
Because in reality IF actually works.

>>60517365
m-muh monolithic 1.7% yields , muh $10000

>>60516212
where's the fang?

>>60517365
>Why is the scaling almost linear?
Because we don't know the limits of AMD's InfinityFabric. So we can only guess and be pleasantly surprised when it works out better than estimated.

>>60517431
What?

>>60517447
What?

Threadripper indeed.

>>60515848
Given the results, AMDs upcoming starship processor with 48-cores on 7nm and a refined arch I wouldn't be surprised it could hit over 10k on cinebench

>>60516107
at 660 bucks it would be really tempting. if they try to pull an intel and jew more shekels out of it, my guess they'll price it at 750-800 to make it "high end" and push it away from the 1800x

at 600 would be a shoa for intel.

>>60517496
$600 is too low. I'd say $999 for top model.

>>60517511
>get the cheapest lowest TDP one
>overclock it
>$200 saved

>>60517550
That's how it works with AMD.

>>60517496
I'd be thinking the highest model would be around $1000, it'd sure as fuck demolish anything Intel has at that price. And anything near considering what we know.
I'd expect threadripper range to start at $750 with maybe a 12 core.

>>60517550
>AMD is confused why 1700 is outselling X models 3 to 1

If they care about margins they won't do that or price them closer to each other

>>60517365
Ryzen inter-CCX latency is a meme started by beta PCPer niggers with a dubious closed-source benchmark and some click-bait amateur level analysis.

IF bandwidth is more of a real concern, but it's highly likely that Threadraper clocks its crossbars at 2x the rate of desktop Ryzen to make 4+ch interleaving viable.

>>60517629
THIS. Fucking casuals.

>>60517474
Maybe with IPC improvements, since I doubt Starships will be more than like ~2.5GHz all-core.

>>60517496
They simply have to be higher priced due to lower volume and supporting the HEDT platform. They may come with premium support service and such as well. And the substrate may cost more to handle the infinity fabric between them directly.

It's going to cost more than just 2 1600s for the 12 core (which would be less than a 1800X, you retards), and more than just 2 1700s for the 16 core.

I would guess $600-$700 for a 8 core that is quad channel and such, $850-$100 for the 12 core, and $1250-$1750 for the 16 core.

>>60517629
The test they did was clearly flawed when the cross-CCX latency was slower than DRAM.

>>60515848
It's over. Soon the only Intel product worth buying will be the 7700K.

>>60515848
sauce me senpai

>>60518041
>It's worth buying a chip on a dead socket
k

File: amadaexcite.png (38KB, 422x537px) Image search: [Google]

38KB, 422x537px

>>60517423
Right here.

>>60515848
Magnificent. Might even get one for the future, seeing as intel i9 is headlining 12 cores optimisation is inevitable.

>>60515848
Fake news

File: 1482805549856.jpg (38KB, 500x332px) Image search: [Google]

38KB, 500x332px

Thank you based Jim Keller, certified shit wrecker.

>>60518228
I mean already, pretty much anything that scales to 8 threads scales well to 16 or 32 threads.

>>60518345
JIM KELLER
JEM KELLER
JEW KELLER
JEW KILLER
>it's that easy

>>60515848
Why the X5650 is still rocking? I have a mobo where i can put that CPU

File: 35120723.gif (3MB, 319x239px) Image search: [Google]

3MB, 319x239px

>>60518345
>Jim Keller

>>60518423
That's two X5650s.

>>60516647

There is also 4+0
>see R5-1400

>offer 3000 dollars of intel performance for 1000 dollars

based jim keller

>>60516741
Holy fuck the second shoah has truly begun.

>>60518423
>2 6 cores are much weaker than an 8 core
huh?

>>60518439
No,its 6 cores and 12 threads a single cpu or i am wrong?

>>60518509
It clearly states 12C/24T which means 2S board.

>>60518506
Still suitable for gaming
>>60518530
Ok, just noticed sorry

>>60515848

1713 /16 *3.6 /4 *32 = 3083.4

can't believe everyone is jumping on this fake train.

>>60517550

Ahhhh, the good old days of Athlon64 X2 ....

Intel really fucked up with the "k" cpu meme. It isn't sustainable anymore.

>>60515848
Fake and Gay
Scaling like that is literally impossible, even with theoretically perfect interconnects between CPUs

Stop drinking the koolaid, /g/

File: amd-zen-core-scaling.jpg (33KB, 597x430px) Image search: [Google]

33KB, 597x430px

>>60518599
F-fake!

>>60518599
>he's basing AMD's core scaling off of Intel's bandwidth starved ringbus design

Lmao

>>60515848
>>60516741
Don't worry, I'm sure it will lose to the 7700K in 720 x 1280 benchmarks, so Intel isn't dead after all.

>>60518724
So would $10000 Xeons, Intel just literally BTFOd itself

>>60516107
These are binned higher than the 1800X before gluing them together. Besides, they know Intel will charge 1700+ for their HEDT. AMD can charge 1200-1400 and still be low enough to grab sales.
>>60516741
V2 is Ivybridge with 1866mhz ram and lower core clock.

>>60515848
And not a single OEM will actually buy any, rendering it's use in a professional environment useless as fuck.

>>60518743
And V3 is Haswell with DDR4

>>60518750
OEMs rarely dabble with HEDT CPUs at all, even Intels.
The largest market for these is companies buying workstation gear for their employees, but I'm sure Intel will be glad to sell them 12 core Xeons that are far slower at a higher price.

>>60516741
Holy fuck Jim what are you doing, you were supposed to bring balance to the semiconductor industry, not destroy everyone.

https://videocardz.com/69717/threadrippers-fake-score

>AYYMDPOORFAGS FAKING SCORES BECAUSE THEY'RE THAT PATHETIC

>>60518599
More Binbus!!

>>60518853
Damage control already? That's fast.

File: coppermine-house-fire-saskatoon-fire-department.jpg (22KB, 720x405px) Image search: [Google]

22KB, 720x405px

>>60518853
Kek. I wonder what Intel has in stock for us.

Hopefully not another house fire.

>>60518853
Oh yeah it's fake alright, because the real one will score higher since it'll be clocked higher.

>>60518779
24C at 2.5ghz with 1866mhz ram scores 3040 ( lower ipc as well )
vs
16C at 3;1ghz with 2400mhz ram scores 3083 ( BW-E ipc )
looks close enough

>>60518979
Broadwell and Haswell IPC is the same.

You also don't know all turbo clocks for either chip.

File: YLozVQk.png (447KB, 1885x956px) Image search: [Google]

447KB, 1885x956px

Does Intel even believe their own shit?

>>60516741
These are chips designed to run in a 1U which have very stringent cooling requirements.

>>60519070
This is bizarre to read.
They seem jelly af

File: Intel-Working-on-7nm-and-5-nm-Manufacturing-Technologies-3.jpg (312KB, 1688x795px) Image search: [Google]

312KB, 1688x795px

When did Intel go bad?

>>60518996
But what's that got to do with comparing that V2 graph vs the new 16C X399 graph?

That graph has 24C Ivybridge at 3040
and the OP's graph is 3083.

>>60519202
When they started fighting laws of physics and doing diversity shit instead of hiring quality engineers and designing a new fucking arch.

File: 1430460027689.png (450KB, 812x720px) Image search: [Google]

450KB, 812x720px

>>60518853
You're right, should be around 3500.

>>60519104
>Intel's flagship xeon gets matched by a processor with a little more than half of the core count
and just a reminder that the new xeon flagship is going to run at 1.4GHz and 205W tdp

File: Untitleddd.png (47KB, 661x357px) Image search: [Google]

47KB, 661x357px

Daily Reminder:

File: ....jpg (22KB, 684x453px) Image search: [Google]

22KB, 684x453px

>>60515848

are those winzip benchmark results?

>>60519311
>I don't know one of the widest used cpu benchmarks today

Man, FUCK ASEAN countries. You can get Intel stuff at msrp and yet AMD is 30-40% more.

>>60519215
And it has a 28 core Haswell at 3150.
Clearly that DDR doesn't do shit in this test as it's purely compute bound and cinnebench doesn't even touch the memory.

>>60515848
F A K E
A
K
E

>>60519513
Just wait for Skylake-X!

>>60519513
>skylake-x 12 core is probably going to score at most 2400~2500cb and be priced at $1500+, while only having 44 pci lanes
kek, intel is finished

File: the-16-core-version-of-amd-ryzen-will-have-the-frequency-formula-3-1-3-6-ghz.png (304KB, 603x600px) Image search: [Google]

304KB, 603x600px

>>60519569
Just like Ryzen only exposes 28 of 32 possible PCIe lanes (x16 slot + x4 slot + x4 NVMe + x4 to chipset), ThreadRaper is rumored to only expose 48 including the 4 to the chipset, so it's not like AMD will dominate in IO by very much at all.

>>60519678
Or they could just expose all the lanes like with Naples.

>>60519735
no anon, HEDT will have a chipset, so it has to give away some lanes, Naples won't have a chipset

>>60519799
The chipset only requires 4 lanes, that's still 60 lanes free.

>>60519799
Naples has to have a chipset if it wants IPMI, infiniband and SAS

but how well does it gaem?

File: AMD-X399-Chipset.jpg (211KB, 1125x1037px) Image search: [Google]

211KB, 1125x1037px

>>60519735
it's a matter of what the socket spec allots for IO lanes, and there's not much of anything floating around for the HEDT platform's socket, and very little for even AM4 at the moment.

there's a sane limit to what you can do on ATX, since at some point using every last possible PCIe lane will bump the PCB layer count up if you have traces running in inconvenient directions.

>>60519799
Epyc may or may not have a chipset at this point.
On-Zeppelin controllers are only PCIe, SATA, and 10GbE, plus USB 3.0.
USB 3.1 specifically needs a chipset, and SATA and 1GbE ports might be less wasteful there than using the 12Gb IO lanes on the CPUs.

>>60519678
Those 4 missing IO lanes are to the 4 cpu attached usb3 ports. It's all there.

Breadripper is probably going to have the same 12 lanes for nvme/chipset/usb and expose 52 to motherboards to do whatever with. Thats if the image you posted is indeed a fake or actually real. Another anon said it was a modified asus skylakex mobo diagram,.

File: 1489855732519.jpg (52KB, 568x612px) Image search: [Google]

52KB, 568x612px

INTEL IS BANKRUPT

NOW WE WAIT™

What can you possibly do with 128 lanes?

>>60519934
everything

>>60519914
>Literally no competition against Ryzen

>>60519934
Anything.

>>60519934
things

>>60519934
Compute:
6 GPUs

Storage:
countless SATA/SAS/SCSI/NVME drives

Balanced:
GPU + FPGA + mezzanine cards

ohshitniggerwhatareyoudoing:
A bunch of capture cards with risers

I find Naples ability to handle all those interrupts smoothly if you fill up all those lanes much more impressive than the sheer amount of them.

File: this kills the SL-X.png (3MB, 1651x1475px) Image search: [Google]

3MB, 1651x1475px

>>60519911
No, the 4 onboard USB ports have their own dedicated controller and PHYs since they're only 5 Gb = 3.0 = 3.1r2 = 3.1a = (pick you own kike lying name).

Actual 3.1r2 = 10 Gb ports require a chipset.

The 32 muxed <=12 Gbps lanes are for PCIe, 10GbE, SATA, and inter-socket Infinity Fabric bridging (eXternal Global Memory Interconnect) only.

>>60520183
Could that SRAM in (?) be some kind of cache for the infinity fabric?

WE DID IT R/AMD

R/INTEL PERMANENTLY BTFO'D

What will AMD do when Intel has to hire Jim Keller to compete, which causes a return to the Bulldozer days when AMD was four generations behind?

>>60520379
if jewtel hires keller then their cpus will become like the 9590 housefire

So how about that 7740K?

You guys want to pay more for what is essentially a de-lidded 7700K on an even more expensive platform but can't take advantage of any the things that make that platform worthwhile?

>>60520230
>that SRAM

It could be for any number of things, and normies probably won't know anything substantial about Zeppelin's un-core until the presentation at HotChips in August.

It's on the order of 1MB for each cluster and doesn't seem to have a bunch of supplementary control logic, which limits the possibilities though.

Could be dedicated per-CCX snoop filter storage (fairly likely), fast lookup tables and frame buffers for the 10GbE controllers (also pretty likely), buffer space for the AES-128 DDR4 encryption stuff (awfully big for that), or something else entirely.

I would not expect it to be any sort of GMI/xGMI link buffering or caching though, since that would just add latency and it would be smarter to throttle sends in the CCX L3s or the IOMMU.

>>60520419
some dumb rich kiddies will ask their parents to buy it for them while claiming that the 7700k is obsolete now

>>60520379
Too jewish to hire good ol' Jim. They're have to be pretty desperate to do that.

>>60520453
if they hire keller then intel will become a true meme

>>60520379
Why would they hire a talented engineer when they can instead hire 500 college flunkies and niggers.

File: oy vay.jpg (38KB, 394x512px) Image search: [Google]

38KB, 394x512px

>>60520379
> rabbis inviting Heinrich Himmler to the bar mitzvah

File: rip intel.png (84KB, 653x726px) Image search: [Google]

84KB, 653x726px

>>60520475
>flunkies and niggers

>>60516957
Dual socket, retard

>>60515945
This is cinebench, a rendering benchmark that should show simular results to how this cpu will compare to others in most multi threaded workloads.

File: intel-diversity-fullbleed.jpg (196KB, 1600x917px) Image search: [Google]

196KB, 1600x917px

>>60520475
They still need to reach their diversity goal.

>>60520531
> Intel being staffed by 12% dindus and 16% beaner, and cutting asians down to 5%

this will surely end well.

>>60520587
The ironic thing is: AMD is a much more diverse company than Intel. Their CEO is a Taiwanese women and the leader of their GPU department is an Indian. But they were hired because they were competent, not because of "diversity quotas."

>>60520793
Raja is yet to prove his competence with Vega.

>>60520379
Jew Killer only works for people that give him full control of whatever project they want him to do. Unless Intel is literally about to go bankrupt that wont happen

>>60520877
Is being the lead architect of R300 and Evergreen somehow not good enough for you?

>>60521055
It's good enough but after bottleneckfrst that was Fiji I have doubts about anything AMD.

File: threadripper-fake-2[1].png (56KB, 794x368px) Image search: [Google]

56KB, 794x368px

>>60520326
WAIT. IT DOESN'T STOP HERE.

File: Joe Biden Smiling.jpg (11KB, 320x236px) Image search: [Google]

11KB, 320x236px

>>60521262
AHAHAHAHAHAHAHAHAHAHAHAHAHAHAH

INTEL SHILLS PERMA-BTFO'D

>>60521098
GCN was what got developed in his absence, and even Fiji and Polaris were pretty much planned out by the time he reappeared.

Hell, I'd be more worried that Vega has little to do with him and that Navi will be his first new opportunity for design contributions.

>>60521262
9001 when?

>>60521420
I think he has more direct input on Vega.

>>60521496
GCN's honestly not that bad of a SIMT design, but AMD has been fucking coasting on it for far too long.

If Vega is really just GCN upgraded to Pascal-level (binning rasterization and non-shit DCC) plus 2xfp16 and HBM, it won't really have gotten itself ahead of Nvidia at all.

AMD need to be actually innovative again sometime this decade.

>>60521580
>AMD has been fucking coasting on it for far too long
NVIDIA still fucks G80 cold and rotten corpse though. Designing new arch from the ground up takes time and money. A fuckton of R&D money. Which AMD currently has almost none.

>>60521715
>NVIDIA still fucks G80 cold and rotten corpse though

howso?
Tesla->Fermi was a pretty clean break in most important regards.

>>60521890
And then they stripped most of Fermi's features moving to Kepler. Fermi was weird. And hot.

>>60521890
Fermi was just a Tesla re-tuned for compute and with the hot-clocks hard-set to a 2:1 ratio.

It ran hot because it was a new arch and a new node simultaneously, and they had to take what they could get, what with their hilariously bad 1.7% yields.

Then like what >>60521946 said, they stripped many of the features Fermi had out in the name of cutting Fermi's power consumption.

>>60521946
>>60522295
My bad, I mis-remembered Fermi as the Nvidia's first SIMT, not Tesla.

Jesus, I don't think I ever realize that Nvidia beat AMD to SIMT designs by 6 years, even if their first attempts got stomped in gaming by VLIW.

>>60522392
Pretty sure that by the time Tesla reached EOL nvidia had taken the lead, not sure though. The benchmarking i did recently with a Radeon HD 4870 and a GTX 260 core 216 had the core 216 in the lead.

File: Intel-Broadwell-EP-Xeon-E5-2698-V4_Cinebench.jpg (201KB, 950x1057px) Image search: [Google]

201KB, 950x1057px

>>60515848
T H R E E
T H O U S A ℕ D

>>60522773
DOLLARS

>>60522873
AND

>>60522886
free shipping?

File: --.jpg (79KB, 1024x1024px) Image search: [Google]

79KB, 1024x1024px

>>60522773
>>60515848

(4980 / 2.1)*3.6 / 80 x 32 = 3415

A M D P C P / I P C B T F O

File: mycomputer.png (6KB, 364x129px) Image search: [Google]

6KB, 364x129px

>>60522967

I ' M S C R E A M I N G

EVEN THE OPTIMISTIC AMD FAKES GET BTFO BY A MODEST HYPOTHETICAL 3.6GHZ E5-2968 WHEN MINE REACHES 4GHz ON 40 CORES FUCKINGINGGGGLMAOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO

>>60523086
>linux

>>60515848
proof or it didn't happen

>>60523129
16C BW-E ipc matching a 24C ivy isn't really that great when it has significantly faster ram and 1T timing ( often enough to offset 200mhz of memory OC ) and cpu clock.

isn't intel releasing some beyond fucked 28 core 56 thread Xeon Platinum thing?

>>60523825
yeah, highest plat xeon is still expected to be 28c, still with 6 channels DDR4 and only 1.375 MB L3/core but now with 1MB L2 apparently.

should be entertaining to see what kind of yields they have on these while AMD is gluing together cherry picked 200mm^2 Zeppelin dies.

>>60524383
>should be entertaining to see what kind of yields they have on these while AMD is gluing together cherry picked 200mm^2 Zeppelin dies.

http://www.guru3d.com/news-story/amd-ryzen-14nm-wafer-yields-pass-80-threadripper-cpus-on-track.html

Maybe the exact opposite of AMD's yields, kek.

>>60524473
at least 20% is better than 1.7

>>60516561
Where's the fahking SATA?

>>60525208
>onboard SATA
>server board
kek

>>60525222
I guess you can add some SATA RAID cards?

>>60516680
2TB in one socket is impressive, but it would make zero sense to not just by a 2P system and use 32*64GB DIMMs.

Even if the mobo was $500 more and a second processor cost you $5k, you'd still come out ahead.

Shit, even 1TB builds would probably come out ahead with 32*32GB instead of 16*64GB.

File: 1362661022441.jpg (22KB, 306x276px) Image search: [Google]

22KB, 306x276px

>AMD actually winning
This chaos timeline is just too much

>>60520793
Diversity through having a friendly work place environment rather than giving money to scammers and having a cuck CEO.

>>60524383
>>60524473
Wasted yields aren't a huge deal as long you get at least 10-20% of them.

A wafer likely costs around $300-$500 for Zen dies. It's probably 200-400 per wafer.
You're talking about a cost of GloFo in the $0.66-$2.5 range per die outside of R&D and other overhead. That's just the material and yield capacity cost.

AMD sells their GPU dies to card partners for $25-$80 or so. Like the 290, a huge die, was $80, that's on the high end. Most of that cost is AMD's profit to recoup R&D.

So really the difference of 50% vs 100% yields is maybe 1% profit change for Intel given that they own their own Fabs and charge absurd amounts for these chips.
AMD is still only expecting about 65%+ profit margin on Snowy Owl and Naples while Intel gets ~75% on their Xeons.

>>60525921
fabs charge close to $5k for a 300mm wafer, and that's after putting millions into fabricating the masks.

Raja may or may not have been around for the initial groundwork of Vega. He may very well have come up to a year late on the project. And architects can only do so much with certain talent. if it's anything like software development I wouldn't get my hopes up too much.

>>60516741

That is the $9000 one, isn't it?

>/g/ keeps falling for it

File: 1490313247424s.png (76KB, 240x244px) Image search: [Google]

76KB, 240x244px

>>60516741

>>60516561
I think I'm getting old. What are those 4 black things in the bottom right corner? Is that some new connection?

>>60526194
are they possibly SFF-8087/mini-SAS?

could also be the header for u.2 NVMe drive cables, though I thought those generally had dual connector rows.

>>60515848
have we reached limits of Instruction Level Parallelism ?

>>60523086
wat

>>60522773
>HEDT/workstation CPU is nearly matching 2 socket server configs
LITERALLY CANNOT MAKE THIS SHIT UP

File: 12b34737e912062970dabb6d46d147c1_csi-yeaaaahhhh-meme-csi_256-256.png (64KB, 256x256px) Image search: [Google]

64KB, 256x256px

You could say that AMD discussion in here...


...is a threadripper.

File: cGIay9e[1].png (3KB, 280x272px) Image search: [Google]

3KB, 280x272px

>>60526434

>>60516741

NO NO INTEL HAS BETTER PER CORE PERFORMANCE THIS IS RIGGED

File: keller amd great again.png (418KB, 627x627px) Image search: [Google]

418KB, 627x627px

>>60518345

>>60516741
AHAHAAHAAHHAHAAHHAHAHAHAAHAAHHAHAAHHAH

>>60526292
>are they possibly SFF-8087/mini-SAS?
Looks pretty similar, would be reasonable to have on a server board too. Thanks.

N-NO

INTEL BEATS AMD IN 640X480 GAMING BY GETTING 145 FPS, 5 MORE FPS THAN AMD

AYYMD BTFO

>>60525377
But that just means you can put 4TB in a 2P system!

That's crazy.
And that's pretty much what any serious cloud provider like Amazon and Microsoft need.

>>60526876
If you asked cloud providers they'd want 8TB per socket if it was possible.

It's currently not with only 256GB DIMMs.
And I doubt AMD is crazy enough to go for 12 channel in Zen2/3

>>60519735
lewd

>>60516037
And it would still be good value. Intel doesn't even have a 16C chip that clocks that high.

File: 28e.jpg (58KB, 680x646px) Image search: [Google]

58KB, 680x646px

https://videocardz.com/69717/threadrippers-fake-score

Its fake you dumbshits.

>>60517024
What's the page that makes up the background in that image?

>>60526050
Ya. I was saying GloFo's cost for the wafer itself. I think I was clear about that.

$5k for 200-400 dies is still $12.50-$25 each

File: hot-2.gif (1000KB, 500x281px) Image search: [Google]

1000KB, 500x281px

>>60516741
Now this really puts 2-socket Naples into perspective, doesn't it?

>>60527097
>it's fake
You're right! The reality is that these are from an engineering sample, not final clocks, and it's likely to score at least 3200 instead.

>>60517629
That being said, though, Cinebench is an embarrassing parallel workload with extremely little coherence conflicts. It stands to reason that it would scale much better than workloads with more inter-thread communication.

>>60527275
Most workloads that scale up to a shitload of cores are similar.

>>60520183
Interesting picture, where did you find it? Are those actually just the PHYs, as opposed to the memory controller logic, taking so much space?

I can't believe no one mentioned the kernel compilation numbers they showed, the dual socket naples finished in 15 seconds while the single socket naples finished in 33 seconds.

Do you see something wrong here? Yeah, it's that the dual socket system is over 100%+ faster, which should be impossible if they're using the same chips, it could be possible if IF was sheer magic and it scales some 115% with 100% hardware, yeah right.

Anyway, we're probably looking at a lower clockspeed Naples with lower TDP for denser environments like blades or 1U

>>60527340
>32core at 135W TDP
I don't care about the clockspeed, even if it was 2GHZ please give me

>>60515848
>X5650 is still this relevant
what are some good motherboards for LGA1366? fuck having all this DDR4, NVME, Optane, etc.

>>60516741
BUT HOW MANY FPS DOES IT GET IN A 20 YEAR OLD GAME AT 640x480?

>>60527359
Even at that clockspeed it would be faster than all of Intel's Xeons besides their 205W 28 core.

And probably cost a quarter of it

>>60527411
Even if it's not faster is has huge advantages in I/O and memory capacity/bandwidth
There's no clear cut winner, the only ones looking at pure compute performance nowadays are only universities and scientific institutes, and those are slowly moving over to GPUs

File: Capture.jpg (96KB, 1096x864px) Image search: [Google]

96KB, 1096x864px

>>60527047


>clock speed =/= speed

>>60527457
I was just looking at that and the 12 core Skylake-X is barely 15% faster in multithreaded than the 10 core Broadwell-E

I'm far from impressed if they've managed to clock it close to 4GHz for all cores.

>>60527466

its clocked at 2hz and is the fastest of any CPU

>>60518853
>>60527097
Based WhyCry

>>60515848
can it run crysis at full res on a 4K TV

>>60527481
Lol

>>60527481
Its single core turbo is 4.5, which is inline with its 10% higher ST performance over the 6950X, and its all core turbo is probably somewhere around 3.5GHz, which again is roughly inline with its MT performance.

There's also a 10 core Skylake-X there, but it has no turbo and loses in all metrics to the 6950X

>>60519202
at 32nm.

prove me wrong.

>>60515848
Meanwhile

>WAIT FOR VEGA

>>60526982
>And I doubt AMD is crazy enough to go for 12 channel in Zen2/3
Someone screencap this shit

>>60527569
Well.. I'd personally be really glad if that happens, but where would you fit all the DIMMs?

>>60515848
Why are they comparing it to two ancient xeons

http://ark.intel.com/products/47922/Intel-Xeon-Processor-X5650-12M-Cache-2_66-GHz-6_40-GTs-Intel-QPI

Here is a list of more xeon cpus performance in cinebench.

https://docs.google.com/spreadsheets/d/1sxzGshuqVtFe_2zgRhN3gXCraR7d8p-NazJ6z0nsGGc/edit#gid=0

If these results are true thats pretty damn good.

>>60515989
The fact that is scales so perfectly with core count is the impressive thing though. This is $5k xeon territory being torn up by a hopefully <$1700 part.

>>60518371
its even easier than that
>JIM KELLER
swap the I and the E
>JEM KILLER
Flip the M
>JEW KILLER

THE PROPHECY IS COMPLETE

Intel's bread and butter aren't the ultra expensive 22+ core Xeons, over 50% of the market are the 12 and 16 core Xeons, those are the most popular 2S solutions.

This would effectively shit on that, Naples doubly so.

>>60527578
Surely we don't need ALL those PCIe slots, right? :^)

>>60527578
I have an idea
IDEA
3D RACKS.

>>60527578
On the other side.
You already got M2 mounts on the backside, why not DIMMs?

>>60515848
Naples/EPYC confirmed for 6k cinebench score

>>60527650
Fuck nigga, this is a workstation/HEDT part.
Naples STARTS at 16 cores, and goes all the way up to 32 core parts.
Imagine that, companies replacing their 24/32 core Xeon 2U servers, with 64 core 2U Naples
Lots of happy customers right there.

>>60527578
In your ass

>>60527685
Actually what's gonna happen is a lot of 2S setups being replaced by 1S, for anyone not needing compute performance but I/O one socket Naples is godly.
Still, a 32 core one socket Naples should do even better in compute than two 12 core Xeons

>>60527711
Not to mention that right now AMD seems to match Intel power consumption and TDP while maintaining twice the cores, so going from a 2S 12+12 Xeon set up, to a 32c 1S naples should also be better on cooling and power bill.

>>60527798
Basically, Skylake-EP is DOA.

>>60527826
Dunno about that yet, but Intel seems hellbent on offering lower core count for 1S solutions, those 28 cores are for 4S and higher.

And 1S/2S are over 90% of the market.
What is Intel thinking?

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

119KB, 1000x1000px

>>60527875
>What is Intel thinking?
>implying Intel with their current CEO ever thinks
Nah.

>>60527875
>What is Intel thinking?
>How do we diversify our employment roster?
>We have too many intelligent engineers, FIRE THEM!, HIRE THE NIGGERS!
>And the CEO?
>ESPECIALLY THE CEO, GET A KIKE, THAT'D DIVERSIFY THIS SHIT REAL NICE

>>60527875
>>What is Intel thinking?
Jews.
They handed over the 1S market to AMD.
They're trying to put up a fight on 2S with Xeon Gold, but you need Platinum Xeons to compete with Naples, so anyone buying Xeons for 2S will be purely from brand recognition, which doesn't go well in datacenter as it goes for gaming faggots.
4S+ AMD isn't even competing, I guess Intel can be satisfied with that 7% market.

>>60527898
>4S+ AMD isn't even competing
Well, with 2S, there's already a stupid amount of DIMM slots. Where the fuck would they all go on a 4S?

>>60527898
>>60527875
Margins are more important to Intel shareholders than marketshare, they have to keep their historic 60% of shareholders will go nuclear.

>>60527913
>Where the fuck would they all go on a 4S?
Higher compute density? That's extremely niche though and why not to opt for POWER then?

>>60527917
>We only hold 30% of the server space now, but hey! we make 60% on the xeons we sell!

I wonder if Ryzen Mobile would be successful or Intel would bribe OEMs again.

>>60527934
If ASUS is coming out with a ROG Ryzen lapop you can be sure Intel already failed to do that with their biggest laptop partner.

>>60527924
>That's extremely niche though and why not to opt for POWER then?
IIRC purchasing a reasonably sized Power8 server and installation runs nearly $80k does it not?

>>60527953
But i want RR-based DELLs. And DELL loves kike cock.

>>60527954
Who the fuck cares about costs when you're maximizing compute density anyway?

>>60527976
Depends on the end goal.
Cost is always a factor so some end.

>>60527963
XPS 13 with raven ridge would be godly.

Imagine a 4 core 15W APU with 700 Vega shaders.
Dunno what clocks it's gonna be, can't really base it off Ryzen since Ryzen is 2 CCX and has a lot of server shit like GMI on it, and 32(28) lanes.
But I'm guessing somewhere around 2.3/3.8, with a 2.6 all core turbo.

>>60527984
to some end*
ftfm

>>60527986
The all core turbo would be higher if the iGPU is idling.

>>60527986
>But I'm guessing somewhere around 2.3/3.8, with a 2.6 all core turbo.
Probably higher. Full Zeppelin on ~2.2 has some ~35w cTDP.

>>60527986
>>60528005
Would someone with a Ryzen chip and Win10 mind opening up your RyzenMaster and disabling a few cores, dropping clocks. May be that we can get some solid numbers to run on.

>>60528026
Stilt did some testing. Zen is extremely efficient on <3.3ghz clocks.

File: 1493967449274.jpg (510KB, 1315x794px) Image search: [Google]

510KB, 1315x794px

>>60515848
delet

>>60528033
Mind dropping a link? Would be interested.

>>60527986
If they can stick 700 shaders on 15W chips hats off to AMD, even more hilarious would be if they use the desktop APUs for mobile, just lower clocked.
In that case we're looking at 768 shaders.

Personally, shaders are better in perf/watt than clockspeed, so 768 shaders at 800MHz would use less power than 512 shaders at 1100MHz.
Just use more die space.
But if AMD is recycling them from the desktop parts, then that doesn't matter and more shaders are always welcome.

>>60528041
https://forums.anandtech.com/threads/ryzen-strictly-technical.2500572/

>>60518462
it's 2+2.

>>60528050
They would also be easier to cool, which is kinda important to mobile.

Hopefully AMD doesn't go full jew really uses a 200mm^2 die for mobile parts, that would make short work of Intel's 100mm^2 mobile chips

>>60528078
>They would also be easier to cool, which is kinda important to mobile.
Trouble being. None of the OEMs seem to care enough about AMD products to put the effort in to properly engineer a chassis with proper cooling.
May be different now. But that's been the tune sang for the past few years in regards to AMD laptops.

>>60528078
>Hopefully AMD doesn't go full jew
When was the last time AMD did go full jew? Athlon64FX? But these were recycled Opterons.
>>60528088
>Trouble being. None of the OEMs seem to care enough about AMD products to put the effort in to properly engineer a chassis with proper cooling
That was before Ryzen Mobile.

Oh boy really cheap xeons, motherboards and ddr3 ecc ram are gonna land on ebay soon. This gon be gud

>>60528050
>>60528078
AMD should really capitalize here, they're more comfortable with cTDP and laptops should really unlock more TDP when plugged in the AC but fall down to lower when not plugged in.

>>60528096
>That was before Ryzen Mobile
And this is still before Ryzen mobile.
We don't know for sure yet.
We do know that Zen is EXTREMELY fucking efficient, so it should work out OK unless OEMs are feeling especially jewish.

>>60528088
That was Bulldozer derivatives though, which weren't good in anything, much less power consumption.

Zen is surprisingly better in perf/watt than Intel, especially at under 3GHz clocks.

>>60528050
Is it possible with current technology to strip away CCXs that don't work with a laser or something and replace it with something else on the die?

>>60516741
certified

File: certified shit wrecker.jpg (13KB, 381x286px) Image search: [Google]

13KB, 381x286px

>>60528114
No, APUs are another die.
>>60528113
>Zen is surprisingly better in perf/watt than Intel
>surprisingly
>SURPRISIGNLY
>ever doubting shitwrecker

>>60528114
That's just inefficient, it's cheaper to make a new photomask with 1 CCX with a GPU glued in.
It's easier now than before because they're glued with IF instead of custom onion buses.

>>60525921
80% yield
:^)

>>60528114
Lasering cores away has been a thing for some time.
A CCX is a different construct on the die. So it should be possible to sever it. How this impacts the base function of the CPU we don't know. How reliant upon eachother are these CCXs, we don't know.
Should AMD decide to use single CCXs in mobile, and replace that die area with an iGPU then it would be a more "conventional" CPU design, and should be fairly well accepted into the marketplace as something easy to implement and adapt to.

>>60528112
>And this is still before Ryzen mobile.
>We don't know for sure yet.
ANUS already teases new ROG laptop with Ryzen. That certainly means fucking SOMETHING is happening.

>>60528126
>AMD's previous CPU<->GPU interconnects are called onion and garlic

AMD's bad naming schemes are historical.
Really surprising they came out with cool names like K8/Hammer and Phenom

>>60528148
Link?

>>60527879
the hector of amd.

File: mark.jpg (1MB, 1505x4363px) Image search: [Google]

1MB, 1505x4363px

>>60528125
If noble prize wasn't run by jew, he would get one

>>60518041

Why the K variant though? Intel has forbidden you from overclocking it anyway :^)

>inb4 DELID THIS

>>60528163
>Hector Ruiz
>Ruinz

>>60528162
https://www.youtube.com/watch?v=-P87BBSHAsg

File: lisasu.jpg (7KB, 125x132px) Image search: [Google]

7KB, 125x132px

>>60524383

>>60528172
There's a 30% chance there's a 8 core Ryzen in there with a discrete GPU.

>>60527647

>>60528276
Nah. I wonder if there is going to be discrete mobile Vega chips.

So Vega in the new APU, is it confirm yet

>>60528379
Yes, new APU uses Vega for iGPU.

>>60528379
I want a new video game system with this

>>60528410
You know sony is going to fucking call them. Have a Zen/Vega PS to derail Scorpio

>>60528425
No, AMD has no time for custom hardware this year.

>>60528450
Scorpio rampup is happening in this quarter actually.

>>60528425
not for another 4 years, but they will start calling right now to start development.

Holy shit PAJEET ON SUICIDE WATCH
READY THE HALDOL

>>60515848
INTEL LITERALLY ON SUICIDE WATCH LMAO KIKES BLOWN THE FUCK OUT
BUY AMD NOW!!!!!!!!!!

>>60528493
It still uses Bobcat and Polaris. We're talking Zen and Vega.

>>60528511
There's Vega IP in both PS4 Pro and Scorpio.

>>60528527
And Xenos had unified shaders before HD2000 and even G80. So what?

>>60528527
What's with fucking consoles and ALWAYS showing up before massively new shit comes and if they waited a year could have 50%+ faster consoles?

>>60518743
>These are binned higher than the 1800X before gluing them together
Getting real sick of you retards pulling this phrase out of your ass

>>60528610
It's mostly true though.

>>60528036
delid

>>60528583
Unloading the old shit for cents on the dollar so the consoles can be cheap enough.

>>60519070
>>60519159
So much for embracing competition..