AMD Ryzen 16 Core Threadripper CPUs & “Whitehaven” P

File: 1494298475605.jpg (62KB, 525x525px) Image search: [Google]

62KB, 525x525px

>>60356849
>4094 Pin Socket
Holy shit. Actual pins used for I/O, maybe intel could pick up something from this.
Cough 28 PCIE lanes Cough

>>60356849

didnt bulldozer and piledriver and scud x2 fentynal all try the MEGA TASKING THREAD-RIPPER shit before?

howd that work out?

>>60356849
DELET

>>60356849
>4094 pins
Oooh boy.

It's pretty impressive how these clickbait sites report 1 year old news.

For fucks sakes who didn't know these things are launching on 4000+ pin LGA?

>>60356849
HOL UP BOYS LET ME TURN ON MY THREAD RIPPER AND KICK SOME SHIT UP A NOTCH. GOTTA RIP THESE BLU RAYS WHILE TRANSCODING SOME RUSSIAN LOLITA UPSKIRT FOOTAGE AS I PLAY WITCHERINO

>launching literal server CPUs into consumer


Lmao, most 2U racks are either 12 or 16 core Xeons currently, they're legit launching far higher clocked chips as workstation solution to us plebeians.
Whoever does work with this will probably cream himself thanking god he won't have to pay $3000+ for slower Xeons

Overclocking 16 cores to 4.1GHz! I can already smell my house burning.

>>60356849
how many gw will the power use be?

>>60357258
Around ~200W. It's basically an overclocked server CPU.

good thing that i didnt buy a new cpu yet.

>>60357182
The AM4 Taichi board actually has enough room to accommodate a theoretical 24+ core Zen at 4.1GHz, that leaves it some breathing room for the mosfets to not blow up.


AM4 can't handle that much load nor does it have enough pins, I hope there's some really insane SP3r2 motherboards out.

>>60357258
Says up to 180W, so probably 130-180W depending on clocks and SKU if they use 3 SKUs like they did with Ryzen

>>60357296
I want Supermicro to DELIVER 2P boards.

>>60356849
Fixed link http://wccftech.com/amd-ryzen-16-core-threadripper-whitehaven-4094-socket/

>>60357328
2P is only SR3 :L

Meaning you'll need Opterons

>>60357296
not that bad as long as thats not the idle usage.

>>60356849

>currytech

>>60357349
Aw fug. Anyway 16/32 at THESE clocks is probably good enough.

>>60357356
Why would it be idle usage?
Modern chips use literally nothing at idle

>>60356849

Imagine mining bitcoins with that.

>>60357356
Zen is gated so aggresively it barely uses power while idling.

>>60357401
asics will still rule

>>60357401
Just a waste of resources, both GPUs and ASICs are faster at that task

>>60357373
No it's not enough, we need 32 core 3.0GHz ones too!
2 sockets? Why would I need two sockets when I can get one chip with the same amount of cores?

>>60357555
Because 32 core Naples will be sold through OEM channels and the chance of you getting one compared to companies that have them ordered while AMD is still buying wafers is about the same chance as Intel releasing Netburst again.

File: 2f7.jpg (28KB, 601x508px) Image search: [Google]

28KB, 601x508px

>>60356849
>tfw only 8 cores

>>60357599
Just wait for companies to dump them on ebay in 2-3 years when they're rebuilding their racks.

>>60357599
Oh well, one can only dream. I wonder how high margins on Naples will be? AMD can allow themselves to throw a hefty price premium, Intel simply can't compete against that.

>>60356900
Core i9-7920X will have like 44.

>>60357686
Wow, $1000 one will have more lanes, unbelievable.

>>60357752
Uhh.. These 16 cores are gonna be 64 lanes.

>>60357752
?? are you stupid

why are we getting so many cores now? we will have 60 by december

>>60357793
Realistically somewhere over 54 depending on how much M2 slots, usb and sata they put in.

>>60357838
Welcome to HEDT. If you read the article, you'd see AMD could potentially put Naples 32 core CPUs on there as well. Which would be truly bonkers.

>>60357838
There's always a demand for more computational power.

>>60357863
I don't think usual mobo makers will want to deal with pain in the ass which is wiring 8 memory channels.

>>60356849
You got my attention anon.
Just give me 40+ pcie lanes and I'm on it.

>>60357913
see
>>60357861

>>60357913
It's some 50+ lanes for you, fuckboy.

>>60357913
Well, you got 60 of them, but pray tell me why do you need that much?

>>60357938
Crossfire/SLI + m2.
Plus at least 64gb ram and a hell lot of vms

>>60356849
>Threadripper

Oh man, I'm gonna rip through some major threads with this
It's gonna be like dubs city
I wonder why it's not an even 4096 pins though

>>60358046
Why AM4 is 1331 pins instead of something like 1333? Dunno.

>>60357938
Even without GPU you can fill those up with capture cards, SDR cards, FPGAs, PCie SSDs, and RAID cards.

Those who need those aren't small in number.

Fuck me it's looks like the best workstation CPU since forever.

>>60358085
lol from what world do you hail from traveler?

>>60358085
Realistically you'll have one or two of those and 2 or more GPUs for virtualization.

x265 and avxenc(AVi encoder) would demand 16 core or more for 4k videos.

>>60358172
It would probably rip through 4k high bitrate content with software decoding like piss too.

The datacenter version of this is practically made for 1U and blade server deployments.

>>60358389
Zen was a server architecture from the start.

I imagine there are going to be a fair number of X390 chipset uATX boards for use in microserver setups.

I would do it, have a threadripper and dual Vega and run it as a personal server and render box.

>>60358389
Zen was made with servers in mind.

>>60358455
>>60358525
>this backtracking after getting destroyed in the consumer market by Intel AGAIN
mmm smells like faildozer all over again boys

>>60358510
>have a threadripper and dual Vega


Jesus Christ, listen to yourself

>it's 2019
>AMD put 4x 400mm^2 CPUs on a chip
>6000 pins LGA

>>60358570
(You) tried.

Honestly I'm more interested in that 16 core mega APU that's probably coming out next year, which will most likely come with 4S support for maximum compute density per node

File: fkugoyim.jpg (49KB, 440x515px) Image search: [Google]

49KB, 440x515px

>>60358774
>the year heterogenoeus compute stopped being a meme
Wait a sec-

>>60358774
>4 iGPUs + 6 dGPUs
>on a single motherboard some way or another

Oy vey Seasonic/Superflower better get working on those 2000W+ CPUs

>>60358856
PSUs*

>>60358856
And with HBM2 cache (i want super-APUs to have HBM2 cache, amd pls deliver). Hilarious.

>>60358884
Cost won't be a issue for these things so there's no reason to not have a HBM2 cache.

>>60358936
Hahaha the ultimate shoah is yet to come. Poor Intel.

>>60356900
>Cough 28 PCIE lanes Cough
You do realize that i7's and Xeons have had 40 lanes per chip for a long time?

>>60359859
THREADBUMPER 2.0 WITH HYPERSHILL-X 2.3 TECHNOLOGY ENCODED

>>60358978
Intel gimped the 6-core and 8-core Skylake-X processors down to 28 PCI-e lanes from 40.

With Broadwell-E, if you needed more PCI-e lanes you could get a 6850k for around $500. With Skylake-X you need to go to the 10-core chip to get 40+ lanes. That's a $1000 chip if we're lucky. Shitsux.

>>60358083
am4+ should have 1337 pins

File: odpKUGD.png (57KB, 625x626px) Image search: [Google]

57KB, 625x626px

>>60357824
No. Are you?

File: 1490310525069.png (2MB, 1920x1080px) Image search: [Google]

2MB, 1920x1080px

>>60356849
House fire: the processor

File: 1489122339460.png (495KB, 1070x601px) Image search: [Google]

495KB, 1070x601px

>>60357296
I CAN'T TAKE THIS ANYMORE

>>60357116
I told myself to never get AMD again but this shit would realize my dream of a linux host and windows vm with gpu passthrough gaming. My 6 core intel just doesn't cut it. More and more games actually use up to 6 cores and the Intel one is clocked very slow.

>>60357116
>Whoever does work with this will probably cream himself thanking god he won't have to pay $3000+ for slower Xeons
Oy vey

So, what you are saying is AMD are killing AM4 already, a platform that they promised to support for 5 years.

What motherfuckers. Never buy AMD.

>>60363591
>being this mad intel is complete garbage

>>60357599
>is about the same chance as Intel releasing Netburst again
With all of Intel's current crazy decisions recently, that seems pretty probable.

Imagine this with AES-NI and cryptonight.

>>60363692
Their latest CPUs are looking pretty netburst DESU.

>>60363591
I give this bait -10 out of 10.

>>60363692
They don't even need longer bibelines anymore.

>>60356849
Please delet dis

>>60361358
the only choice.

>>60358978
>muh xeons
meanwhile Naples will have 128 lanes

>>60367201
Yes but you'll never get one.

>>60363591
holy shit this is the most retarded post ever

>>60367216
WOOOOOW REKT

>>60356849
>every ryzen cpu is multiplier unlocked
>cpu comes out of the box with 3.9 ghz
>oc's to 4.0-4.1 ghz max
Wow what an amazing selling point!

Too late amd street shitters. Intel is releasing a 12 core 4.5ghz i9

>>60367237
yeah, going from 3.0 GHz to 3.8 GHz.

>>60367324
300W or 400W?

>>60367413
666W

>>60367324
>4.5 GHz turboboost
>On the 10 core version
>Less cores, less PCIe
>100 MHz more base clock on 12 core

Wow it's fucking nothing. And it's still a housefire. https://videocardz.com/69457/specifications-of-intels-core-x-i9-and-i7-series-supposedly-leaked

>>60367465
http://wccftech.com/intel-skylake-x-core-i9-7920x-7900x-7820x-7800x-x299-leaked/

12 core 4.5 ghz boost without OC, 44pcie lanes. Amd is fucked

>>60367486
The turboboost on the 12 core hasn't even been revealed yet.
Nobody cares about turbobutt on a HEDT platform that is all about more cores, where AMD has already won.
AMD has 52+ PCIe lanes. And it won't have some absurd $2000 price tag, either.

>>60367324
>4.5ghz
Single core turbo. The 10 core has a base clock of 3.3ghz vs. 3.6ghz for the highest binned Ryzens.

>>60357838

Because we have hit a wall with everything else. Single core performance hasn't budged forwards for a while now.
Only thing they can now improve is moar cores.
It's actually strange that this didn't become a trend sooner, but that's probably because Intel was the king and didn't feel too threatened, so the whole field stagnated.
At least the future content will be making more and more use of cores.
About damn time.

>>60362897
DELID THIS

>>60367660
The leaked picture says Turbo 2.0 is max on all cores though, that's 4.3 for the 8 and 10 core CPUs.

>>60367209
64 will be enough with the 16 core.

The 32 core will be achievable in a few years from ebay

>>60368005
I don't think you're going to hit 4.3 GHz turboboost on all cores. Turboboost scales based on CPU usage and temperature, so if you've got all cores going 100%, it's prob. not going any higher than 3.7 GHz.

>>60367958
>>60362897
Samefag

>>60367486
Comparing l3 cache with amd is a joke. Intel is really coupling 12mb l3 to a 8 core CPU? God I'm with and on this

>>60368437
hello i am not an informed person
what does having more cache does? what workloads does it benefit?

>>60368294
Kill yourself retard.

>>60369453
http://www.extremetech.com/extreme/188776-how-l1-and-l2-cpu-caches-work-and-why-theyre-an-essential-part-of-modern-chips
https://www.extremetech.com/computing/55662-top-tip-difference-between-l2-and-l3-cache

File: bbc.png (41KB, 200x219px) Image search: [Google]

41KB, 200x219px

>tfw no 256 core 5.0Ghz Turboblast 6.0 CPU with 420 PCI lanes and ultra threading technology to double the logical core count wth octaple memory channels for my gigabit Ethernet ram to be torrented
>tfw will never be a gigatasker™

Its simply not fair broooos

>>60369453
L3 is a victim cache, ie it holds what is evicted from l2. Furthermore l3 is shared among all cores and is essential to coordinate inter core communication.
The latency between l3 in ccx cores is one of the performance reasons of slowdown in zen.

>>60370044
Yet it has no problem fighting off a equal core Broadwell with more cache while being far more power efficient.

>>60370279
It's a problem only when you have very busy interthread dependence. Zen have more cache but slower interconnection.
If they manage to release the hedt with better l3 interconnect it will probably top Intel in most scenarios.
What is really nice is that amd is pushing the envelope, forcing Intel to move forward and even then, and will sell a 16 core behemoth.

>>60370707
AMD can make some real progress with Zen in the upcoming years.
Intel can only up the clocks and TDP and hope for the best.

>>60370985
We really don't know.
One thing is sure Intel has been completely absent on desktop, it only target is to keep reducing power on mobile and offering mediocre chips.
Intel even stated now it's a service company.
Even an Intel fanboy needs to admit most of recent milestones on desktop are due to amd, first real multicore, first 64, pushing the cores envelope. Intel committed an error with bulldozer, while Intel done several like the phone chips but Intel has bottom less coffers.

>>60363229

They still need to get IOMMU support done, but I think Vega and infinity fabric will have a surprise for us about that similar to how they have their multiuser GPU servers.

So imagine being able to feed GPU cores across multiple VMs on the fly and instantly as needed. So if you're just compiling linux on one machine with 1/4 your CPU. Encoding or rendering on another 1/4, and half your cores playing the newest game while your dual Vega GPU splits it's time rendering/encoding, displaying the Linux kernel compiling, and gaming on the other half of your dual Vega.

File: 1492801783942.jpg (134KB, 954x605px) Image search: [Google]

134KB, 954x605px

>>60371960
Yeah, Zeppelin's architecture allows raw physical address space memory access (IOMMU-bypassing) over the IO lines muxed with PCIe.

It would be really surprising if Vega didn't do the same, such that untunneled Infinity Fabric traffic can be passed over a PCIe slot to a Ryzen/Naples CPU, or switched through a CPU between PCIe boards.

>>60372368
Speak english rajeesh

>>60370985
>Intel can only up the clocks and TDP and hope for the best.
This can't work forever, Intel's process (and any future process they have in the pipeline) favors efficiency, and they are already squeezing everything clocks-wise from their current process
AMD can pick GloFo and TSMC, or anyone else, but GloFo has the ridiculously high clocking process IBM uses for their POWER chips
They could easily wreck Intel in pure performance, but it's not worth it when the only market for that it's cost-no-object servers and enthusiast gamers
Intel's seriously fucked if they don't get some miracle, like getting their FPGA acquisitions to deliver something revolutionary within a year, the only way they can remain somewhat competitive it's by upping ICC gimping and revving up those bribes and discounts
>>60372444
>Pajeet grasps at straws

>>60372463
>blablablalblabla AMD is da best!11

t. Streetshitting pooer Aka YOU

>>60363591
2 rupees for you rajesh. Thank you come again

>>60372482
Poor self-hating Pajeet, don't worry, Raja wants to build a Pixar-like powerhouse in India
Eventually him and a few nationalist Pajeets will bring you proper hardware

>>60367237
Yes better buy Intel, OC it to 5ghz and have same performance.

>>60372444
Ryzen's 32 main IO lines can be used for any combination of PCIe, Ethernet, SATA, or Infinity Fabric (coherent memory interconnect for multi-CPU systems) connections, so there's nothing stopping PCIe cards that speak native Infinity Fabric.

In theory, you could even make an add-in card for a 2nd processor + memory, and although inter-socket bandwidth would be a bit low, it would be properly coherent.

>>60372444
Only English you know is muh games, you fucking cuck

>>60372444
>>60372482
>>60372530
>>60372577
Oh good IMG_xxxx is here to shit up the thread.

>MOAR COARS didn't work
>time for MOAR PINS

>>60372880
No, but what are you doing on technology board if tech talk is like an alien language to you? Go play some games before sleep. You have school tomorrow

>>60358085
I can't wait until we go back to the days of 80's IBM PC's and every computer has a ton of cards to control everything.

>>60372980
You mean Intel PCs, Intel invented x86

So if they're using the sameish socket, how long is it until someone puts a 32/24 core nipples chip in the HEDT motherboard. Sure it's going to run at quad channel memory and only 64 pcielanes out, but it'd hardly be a downgrade 3 years in the future when they all hit ebay.

>>60372680
>In theory, you could even make an add-in card for a 2nd processor + memory, and although inter-socket bandwidth would be a bit low, it would be properly coherent.
You're a mad man. Do we know how far it would be able to go though. Does each socket need a direct link to every other socket, or can the coherency work jumping through chips with a latency hit.

>>60373131
IBM invented the gas chamber :)

>>60373131
IBM was the first company to call it a "PC," with the PC 5150. It's why when talking about 80's PC's people say IBM PC's and compatibles.

>>60372680
Wait, does that mean Vega will talk to Ryzen CPUs using infinity fabric over PCIe? What would be the performance implications if so?

>>60373213
They're also the reason AMD's in the x86 game

>>60373250
Unknown because there hasn't been something like this before....

>>60373263
AMD isn't in any game, it's bleeding money and will go bankrupt soon

>>60361466
>still less power consumption than an i5

>>60373159
coherent memory traffic is routed in each chip's central IF crossbar.

there is probably a fundamental hop limit, though the protocol is still private, so nobody knows what it would be.

somebody could probably make a processor/memory card with a cabled/external PCIe port in back, and you could have 4 daisy-chained NUMA nodes in 2 separate chassis.

It would probably perform like dogshit having things strung out so far on such narrow channels, but it could at least function in theory.

>>60373309
They're in both the CPU and GPU game. How well they're doing is debatable, but this is not.

>>60373250
It means that it's highly likely, though still not fully guaranteed, that Vega can talk IF over a PCIe slot.

Performance implications are harder to guess, but it probably mostly means lower latency for remote memory reads and the ability to do APU-like combined calculations without as much added programming complexity.

Best case for gaymen is that the "High Bandwidth Cache Controller" performs more efficiently and that fps are higher when the card needs to fetch remote assets.

>>60373309
Kek! How long has this meme been going on for? 3 years?

>>60373421
Since 2007

Every year AMD is bankrupt, for real this time.

>>60373470
No senpai, it's been going on since 1987, possibly before
People were shitting horribly on AMD during the K5 days, and everyone said they were alive in the 80's thanks to IBM
When Tom's released their video of the burning XP people went again at it, they said the same when Tom's praised Presscott, or when the first Pentium was launched and AMD made the Am586

File: 1477092251969.png (258KB, 506x588px) Image search: [Google]

258KB, 506x588px

>>60368074
>in a few years
J U S T W A I T
U
S
T
W
A
I
T

>>60373788
You're welcome to shell out $5000+ now for it, dumpster child.

>>60373788
Ah, the mating call of the noble AMDfag.

>WHITEhaven
If you do buy this, you are a race traitor.

File: YuExZ.jpg (183KB, 800x1000px) Image search: [Google]

183KB, 800x1000px

I can't believe how well this image stood the test of time

>>60373907
Kikes on other subhumans on suicide watch.

File: 1476374643770.jpg (290KB, 1024x683px) Image search: [Google]

290KB, 1024x683px

>>60373911
Yeah, go look at i9's

>>60373969
lol good one

>>60357884
>I don't think usual mobo makers will want to deal with pain in the ass which is wiring 8 memory channels.
They could simply support 1 DIMM per channel.

I'd be more interested in having a 2p board with 2 12 cores, though.
One where I could overclock one to 4GHz and underclock the other to 2.2GHz.

>>60357793
>>60357861
It's likely 32 PCIe on the CPU itself. Motherboard chipset could add 8-20 more or so.
64 lanes is when you have TWO of them. Just like 128 lanes is with TWO Naples.

>>60374063
One Naples is 128 lanes, 32 per die.

File: Screenshot_20170513_152413.png (2MB, 1920x1200px) Image search: [Google]

2MB, 1920x1200px

>>60361466
>0.22v
>housefire
acks dey

>>60374100
Silicon doesn't even function at .22V

Sooooo what does this mean in general? Doesn't CPU use more than cores?

>>60374153
has anyone really been far even as decided to use even go want to do look more like

>>60374165
Shut up Blaine

>>60356952
Pretty good for everything but games. Why do you ask?

>>60374124
What is it then? Why is lm-sensors reporting wrong?
The heatsink is cold to the touch.

>>60374238
It's being misread, threshold voltage of silicon is slightly over .5V

https://en.wikipedia.org/wiki/Threshold_voltage

>>60374290
How do I make it read correctly?

>>60374473
Modify lm_sensor source yourself, but good luck, not even Windows programs read Ryzen's voltages correctly yet

>>60374063
>One where I could overclock one to 4GHz and underclock the other to 2.2GHz.
You could just use different profiles on Ryzen master
>>60374473
You install Windows and Ryzen Master :^)

>>60374554
I can't figure out this qemu and kvm stuff.

I really wish I had a cute trap gf who is a linux pro. and i could sit between her legs and have her play with my hair and giggle while she walks me through it...

File: homu~.png (55KB, 150x224px) Image search: [Google]

55KB, 150x224px

>>60356849
It's over, Intel is finished!

>>60374554
>You could just use different profiles on Ryzen master
Na. It's so stuff that needs fast threads can run on one CPU and background stuff on the other.

Like the bigLITTLE mobile CPUs.

>>60374063
>>60374084
Every Zeppelin die has 32 general purpose IO lines that can be used for PCIe, 10GbE (max 4), SATA (max 6 or so), or inter-socket xGMI (external Global Memory Interconnect = Infinity Fabric bridge) lanes.

2P Naples uses 64 of 128 lines on each socket for the interconnect, so you have the same 128 lane PCIe lane limit as 1P.

>>60374850
Huh.
So there is more PCIe lanes per die than Ryzen?
Ryzen I thought was 20 PCIe lanes per die, but 4 is reserved.
So with a X370 board that adds 8 to the remaining 16, you get 24 total.
Then if you put 2 together, you get 32. 4 and you get 64. 64x2 is that 128.

>>60374970
Don't ever bother counting what the chipset adds when factoring in total IO bandwidth. The chipset uses 4 lanes (similar to intel) out of the 32 available to a zepplin die. But it's also going to use 4 lanes on the 128 lane naples chips.

On ryzen am4 it's something like:
16x graphics (8/8x on x370)
4x general use
4x chipset (this add 4-8 more pcie2 and various usb and sata)
4x usb3
4x NVME (or 2xnvme + 2sata)

All of that direct connected to the cpu. In theory it's enough to not need a chipset for true soc, which I'm hoping what x300 itx boards are going to do. Leaks have shown ryzen mobile having a similar configuration.

But naples is set to be far more configurable as the extra dies simply add 32 lanes each without anything being reserved. (Apart from 2socket where half of the 256IO lanes are used for interconnect, so 1s and 2s will have the same external IO)

>>60374970
Ryzen is 32 PCIe lanes, with 8 going to the chipset. There aren't 24 lanes going to slots since mobo makers are opting to wire misc shit like 1GbE controllers directly to the sockets.

>>60375195
Isn't 4x for the chipset?
Makes me think like a lot of PCIe lanes are disabled on Ryzen

https://www.servethehome.com/second-amd-naples-server-pictures-platform-block-diagram-112-pcie-lanes/

112 after mobo takes its share of lanes, infiniband, IBMI, SAS, M2/U2, etc

I've read somewhere that Naples can service 12 or more 10GbE and they're capable of 40GbE bridging

That's insane

File: 1493827103174.png (716KB, 713x785px) Image search: [Google]

716KB, 713x785px

>>60375519
>Naples can service 12 or more

>>60367237
>every 7700K cpu is multiplier unlocked
>cpu comes out the box with terrible TIM under the heatspreader
>doesn't OC at all since Intel said not to do it
What an amazing selling point!

>>60375519
>10GbE
Too bad there's not a single AM4 motherboard with that even if the CPU is muxed to do 4 of them and if there was they'd change $200 extra as if like a fucking 10G NIC was new technology or something, fucking clowns.

>>60375539
>>60375519
>12 10GbE links on a very fast CPU that can have boatloads of ram.
NSA must be creaming themselves, imagine the SPI you do on one of them!

>>60375519
40+GbE bonding is completely unconfirmed and not really even rumored.
But yeah, in principle 2P Naples could drive 32 10GbE interfaces (up to 4 per Zeppelin die), even if that will never be seen.

>>60375986
Wow, you could theoretically use Naples as a switch for an entire fucking building.

>>60376039
That's a pretty expensive switch, though those 20+ port Broadcomm low latency switches aren't much cheaper

>>60376039
A 1U 30+ port 10GbE swich is much cheaper and lower power, so nobody will ever make a mobo that can do this.

I recently bought a pair of used 40Gb Mellanox swiches for only $1200 each, which probably won't cover a third of a single 32c Naples chip.

>>60376222
But can your Mellanox rip through a mySAP Business Suite like it's putty as well?

>>60375723
And here I am being pissed off at my boss for thinking IMACs make good machines for our editors when they have no upgrade path to 10gbe outside of stupid expensive TB2 boxes. He even had cat6e wired through our office.

>>60376374
What's the min guaranteed power level for a USB3.1 port? 10GBASE-T needs maybe 5W, but there's just barely enough bw for a 10GbE port.

>>60376374
>>60376582

>>60376374
Businesses don't really upgrade their client machines. They replace them. Their upgrade path tend to be whole new machines with a whole new warranty and support agreement.

>>60376582
>>60376374
Don't forget that 5GBASE-T or whatever is coming out for exactly this category of nigger-tier problems.

>>60363229
WHAT? I VM game on a i3 7100, that's only a dual core with HT! Just pass and pin all the cores and threads, it'll run no problem, Linux can manage with just a little bit of CPU time unless you really need to heavily multitask on both OS at the same time. Come on, this has been the most misinformed post I've read this day. And I have legit benchmarked this low end VM setup of mine and it reaches the median of its kind in 3dmark.

File: kizuna_ai__iykwim__by_thela28-db0ax94.jpg (121KB, 1280x720px) Image search: [Google]

121KB, 1280x720px

>>60367324
>$1500

>>60376639
We're small and I'm pretty much the whole IT department. All these 5k imacs are brand new and really expensive, I don't think they're going anywhere for a while but we're moving into 4k content production and they just can't be upgraded to handle the extra load outside of using external SSDs, but everything still has to go back to the servers at some point anyway. Costs money and time regardless.

>>60376582
>>60376645
I'm certainly keeping my eyes on this stuff. But i read it's pretty much the 10bge protocols but using less frequencies/bandwidth. So I'm expecting controller to still be expensive for hardware (initially before it ramps), it's mainly to save costs on already installed cat5e.

This is getting offtopic though, Threadripper is certainly on the list as far as the next hedt I need is concerned.

File: 1491166649444.jpg (66KB, 568x612px) Image search: [Google]

66KB, 568x612px

>RYZEN WILL CHANGE PROCESSORS
>benches come out, all looking good
>actual benches come out
>pic related
>bit later
>THREADRIPPER AND WHITEHAVEN WILL DESTROY INTEL
>still gets beaten by a 2600K @ 4.4Ghz

Why even bother?

>>60377220
You're doing a shit job here self hating poojet

File: 1494519493351.png (2MB, 769x8779px) Image search: [Google]

2MB, 769x8779px

>>60377220

File: 320180.jpg (91KB, 800x1002px) Image search: [Google]

91KB, 800x1002px

>Threadripper

>le 16 cores
I don't give a fuck
Give me a alternative to the pentium g4560, better and cheaper fucking already

>>60377336
R3/Raven Ridge is coming out later this year.

>>60377465
How late

>>60356849
Is it actually a Ryzen or a server part (Opteron or a successor to Opteron)?

>>60377483
2H 2017

>>60377539
It probably will be branded under Ryzen, but every current Ryzen chip it's an Opteron reject
It's on a different socket, on the server socket, not on AM1
It's closer to the Athlon FX-64 of yore, possibly gonna be a server platform rebranded for muh extreme gaems

>>60377539
It's still Ryzen but requires a new socket/mobo. Every HEDT platform is like this.

>>60377483
Probably very, since it's built off Vega which is barely done validating as of now.

>>60377539
This seems to be a consumer edition of Opteron on a one socket desktop motherboard.

>>60377684
Vega has already been on silicon since December, they're quite past validation
It's probably going to see a paper launch on Computex at the end of this month

>>60356849
Holy shit, this is actually going to take a steamy shit all over HEDT. Intel has been caught with their pants down.

>>60377832
JUST BUY THE 7740K OKAY

>>60377902
I bet it still has toothpaste under the hood.

>>60377973
No. People have tried replacing the TIM with toothpaste, and it actually works better for 2-3 months.

>>60356849
> 4ch HEDT chips needing 4000+ pins

what did AMD mean by this?

>>60379438
>what did AMD mean by this?
32 PCIe lanes, 8 10GbE NIC's with 40GbE bridging, along other nifty things

whitehaven? White? Haha fuck that's racist lmao

>>60379480
each IO lane, whether used for PCIe, 10GbE, or whatever, realistically requires only 6-8 pins (2 directions * (2 wires for differential signal + 1 or 2 for ground)), or ~400-500 pins for 64 lanes.

4 channel HEDT really shouldn't need much more than 2k total pins, unless they're trying to have socket compatibility with Naples for some reason.

File: 8adcba34fe39ab4052d2f35751911a2796203c5d5afceb85ccaf0f46734d1aeb.png (359KB, 536x468px) Image search: [Google]

359KB, 536x468px

>>60357116
Pic related has a 128GB computer with DUAL XENONS (32 cores). he paid only around a 1000 for his machine about 3-4 years ago. Not bad.

I think he said he was disappointed that his machine was still so good despite being so old.

>>60379554
>4 channel HEDT really shouldn't need much more than 2k total pins, unless they're trying to have socket compatibility with Naples for some reason.
They're rebranding the server platform senpai
>>60379674
>Pic related has a 128GB computer with DUAL XENONS (32 cores)
He doesn't, his machine is 1P, it's a 12c/24t Haswell-EP Xeon at 3.6 Ghz, and he paid way more than 1K
He usually calls threads cores, so he calls 24c/48t Xeon's 48 core machines
On one of his recent streams he showed where he bought it, Northwest Falcon iirc, and told how they fucked him over with defective RAM and that he wouldn't recommend them
>I think he said he was disappointed that his machine was still so good despite being so old.
Terry wants a whole fucking 300mm wafer as his CPU with immersion cooling, something like a few hundred cores running at 10 Ghz iirc
Also he upgraded to 256GB of RAM in February

>>60379751
>Terry wants a whole fucking 300mm wafer as his CPU with immersion cooling, something like a few hundred cores running at 10 Ghz iirc
Sounds pretty cool, get on it.

>>60379751
>Also he upgraded to 256GB of RAM in February

wtf is he doing with his time (besides templeos) that can reasonably use that much memory?

>>60379674
Xenons are triple core processors

>>60379805
Stuff in TempleOS like physics (mostly optics iirc) simulations that can balloon into eating a shitton of RAM
Nowadays he seems to just browse two or three tabs in Firefox, he once showed htop and he was using like 1.5GB of RAM
So yeah, that RAM sits unused 99% of the time, I believe his SSD isn't smaller than his RAM

>>60356952
As long as you weren't gaming they were great, there's still a few communities out there buying cheap bulldozers now because they're cheap rendering rigs.

>>60356849
i bought a new pc at the wrong time
i should have waited more years and 12 cores would be the norm

Terry Davis has a 12c/24t xeon with 128gb ram, not a 24c/48t xeon with 256gb ram

>>60379855
>I believe his SSD is smaller than his RAM
fix'd
>>60379886
He has 256GB of RAM nowadays, he upgraded this year under orders from God

File: 1494703919865m.jpg (85KB, 767x1024px) Image search: [Google]

85KB, 767x1024px

>>60356849
>BULLDOZER
>PILE DRIVER
>THREADRIPPER

Question, how much will I get a performance hit if I were to use my old 2400mhz sticks on a say 1700?

>>60380124
It will be barely noticeable compared to 2666 and maybe 3% compared to 3200. You'll be fine.

>>60380290
Thanks familam.

>>60380124
Shut the 1700 is so fast it curb stomps on literally the slowest ram you can pair. Ryzen is idiot proof, literally.

File: lOFGE.jpg (180KB, 944x538px) Image search: [Google]

180KB, 944x538px

>>60380124
Pretty big.
Faster RAM and overclocking can drop you from 110ns memory latency to almost 60ns, along with making the cache and CCX faster.

Combined they can be over a 40% performance improvement in gaymes compared to 2400MHz RAM and the 1700 at stock speeds.
But most applications won't benefit from faster RAM. especially not faster than 2666. And you can possibly overclock your RAM to like 2800.

>>60375539
Sharp is not for lewd.

>>60380550
Will Ryzen even boot on anything other than 2667mhz? I can't even find any 2667mhz anywhere, there's only 2666.

>>60380550
Keep talking im almost ryzening

>>60381742
wat?
2666 and 2667 are the same thing.
Your motherboard will try various clocks until it boots.

>>60381791
I'm going with Bioware this time, which I heard is finicky as fuck and has very poor QC.

File: [HorribleSubs] Ange Vierge - 09 [720p].mkv_snapshot_06.58_[2016.11.04_01.07.14].jpg (39KB, 393x391px) Image search: [Google]

39KB, 393x391px

Should I go for Ryzen now or is it better to wait for more/better motherboards and possibly 15% better performance a year later? Isn't Ryzen 2 also supposed to use a smaller process?

>>60375178
B350 can run 8x8 for gpus as well as long as you use amd gpus.

>>60377220
>2600k
I see what you tried doing there.

>>60382375
>Should I go for Ryzen now or is it better to wait for more/better motherboards and possibly 15% better performance a year later?
No one has announced a better mobo yet, we're only waiting for mITX
>Isn't Ryzen 2 also supposed to use a smaller process?
No, GloFo's promising their 7nm process on 2017H2, and AMD has parts on 7nm at 2018 in their latest roadmaps, they have implied that Zen+ will still be 14nm
It would theoretically arrive around march next year
tbqh If you buy now you can upgrade next year anyways, but it might not be worth upgrading if you have anything Haswell or newer

>>60382375
The sooner you start MEGATASKING the more time you'll save. BUY NOW

>>60382375
Na I wouldn't wait a year for the potential 15% performance improvement.

You buy it now, then in 2-3 years you upgrade to the 7nm or 7nm+. The performance is already really great.

>Isn't Ryzen 2 also supposed to use a smaller process?
No. Zen3 is, in 2019.
Zen MIGHT use LPU instead of LPP which could be 5-10% higher clocks, but there is no confirmation of that.

>>60382375
If you need to upgrade now, ryzen is good.
If you have a haswell or newer i5 or nearly any i7 and you're mainly doing games you should wait until at least this time next year and see if we have any better news.

File: threadlets.png (842KB, 880x499px) Image search: [Google]

842KB, 880x499px

>>60382611
Any i5 isn't good enough. Even in some older games.
4 threads just doesn't cut it. Not even for gaymes.

>>60382646
>1280x720 low
when will this bulldozer era meme die

>>60382646
It's fine. It's not ideal but the average user will not notice any difference going from an i5 to ryzen.
I would recommend a ryzen over building a new intel rig, but if you've already got an intel rig it's fine and it'll be fine for long enough to wait out 7nm or whatever.
Intel's still got the edge in a lot of games too, even AMD admits it. you have to hardcore cherrypick to make a newer i5 to R5 look like anything more significant than a sidegrade for gaming.

>>60382665
>720p low benchmarks are bad when they show the 1600X over 50% ahead of 4c/4t i5 garbage.
>720p low benchmarks only matter for 10 year old games and not ones from a few years ago.

>>60382707
For people just on 60hz, yeah. A haswell or newer i5 will generally do 60fps+.

Though in some newer games, even well optimized ones like BF1, i5s will sometimes dip below 60.
Even a 2600k overclocked to 4.5Ghz is better than a 5GHz overclocked 7600k in games now days. The clock locked 4.2GHz i7-7700 is better as well by a larger margin. i5s are really garbage, even if they're "good enough". They only thing they have going for them is that they were better than Bulldozer.

>Intel's still got the edge in a lot of games too
On average, the minimums are better with Ryzen. Thus it tends to do better in 1440p, 4k, whenever you're GPU bound which almost everyone tends to be. Averages are pretty meaningless.

>>60358584
>tfw trying to straighten the pin on a 6000 pin CPU

>>60382709
They are bad in every case and don't represent future performance and much less current performance, this shitty Intel forced meme shouldn't be endorsed anywhere

File: bf1_1920_1080_min.png (17KB, 500x330px) Image search: [Google]

17KB, 500x330px

>>60382787
>On average, the minimums are better with Ryzen.
On average yes, but the actual numbers are all over the place, the minimums in some games like Styx, Shadow warrior 2 and some RTS are way better on ryzen which is good, it shows the future potential of the platform, but focussing on an average drawn from these values obscures the truth that in nearly every other title on the market right now, ryzen performs roughly on par or worse than i5s, even in minimums.
You mentioned battlefield 1, it gives better minimums to intel CPUs.
That would be cherry picking. I only mention it since you did specifically call that game out.

>>60382646
Which game?

>>60373788
I WANNA FILL UP MUH 256GB OF RAM WITH TENS OF THOUSANDS OF INSTANCES OF SHIT I WROTE MYSELF

>>60382913
Metro Last Light.

>>60373621
nigger ther was no groups like 4chinz it was just big nerd gatherings where everyone dressed like a steriotypical nerd.

AMD didn't reassate with anyone it was a unknown brand then. Just shut up young fag.

Get your head out of your ass.

>>60383080
>Metro Last Light.

You mean Metro Last Light redux

That game has serious tessalation and lighting
Lots of partical effects doimate that game.

Have a look at how the remade thief lighting if that type of effects was implemented.

fuck consoles they are holding up graphics and compute power.

at least servers can go full bang with out having to wait for others to be at par with it.

You can't talk about games in the same way as severs bro it doesn't work like that.

>>60382889
But MUH gpu bottleneck

>>60384733
May as well remove the graphics card and run in software if you want pure cpu tests.

>>60385247
That's not fair to Intel

>>60373184
wtf? i love IBM now!

>>60379987
>News today
>AMD got 2 scoops of the Servermarket
>Intel only get 1 scoop

>>60382391
It's not in the feature lists and all the 350 mobos I looked at only have one physical and electrical x16 slot. Any others are only 4x electrical.

If it's possible it's silly as that's pretty much the only thing 370 has over 350, 2 sata and 2 pciev2 aren't going to make much difference in most cases. 370 should have had a 48lane plx switch for proper multi gpu crap.

File: 1428286733368.png (463KB, 1070x601px) Image search: [Google]

463KB, 1070x601px

>>60356849
FUCK FUCK FUCK FUCK I JUST GOT A 1700X

File: Screenshot (5).png (166KB, 535x536px) Image search: [Google]

166KB, 535x536px

>>60381742
The big bios release last month made booting with fast memory easy. It just didn't do it automagically.

Just set my settings to what the xmp profile stated manually and set to 1.3v and away it went without a dropping a beat. I'll play around with timings and voltage when I get time in the future though. Mobo is an asrock a350m pro4 as it' hidden in the ss.

>>60356849
should have made it 4096 pins so they could market it as "4K"

>implying WCCFTech is a reputable source

>>60387104
It's not a farfetched thing. I don't think this hedt socket is going to be a big seller, so if they reuse the server socket it keeps manufacturing costs down. Motherboards are only going to need half of the traces compared to a server board so there's more gains from that as well. Just hoping a full 32core chip can work in it for science.

>>60386745
As far as i read, this chip is going to be land grid with pin on mobo

>>60357296
The Taichi isn't even their 1300w vrm design.

>>60387239
If they don't completely JUST the pricing (like making the 16 core more than $800-$850) this is going to be an amazing seller for people who want a lot of power but don't need the 2P capability of Naples in the future or boatloads of ram

>>60356900

A lot of those pins will go unused in this platform. Only half the ram channels on the package are going to be used in these HEDT chips. Naples with all four cpu dies would wire up all the memory connections in the package.

>>60357863

what would be the point. HEDT needs high clock rates. Ryzen takes ~90 watts. These things with 2x dies are going to take 180W, and should deliver similar clock rates to Ryzen.

With Naples you are going to be thermally limited by the package (article insinuates this with "200W and more"). You can't get great clocks, but it will work great as a server cpu, as it was designed to be.

>>60373131

He really means IBM. They designed everything about the original PC. The ISA bus, they wrote the original bios, etc. They just purchased the chip from Intel, but they forced Intel to license the design to AMD so that IBM would have a second source to buy from. Shortly after the release of the original IBM PC, Compaq reverse engineered the bios and made a clone. And thus the pc was born.

>>60389379
everyone goes insane over high clocks but it literally doesn't matter if you can optimize your IPC. If you double your IPC, 2.5GHz is just as fast as 5GHz, and Ryzen has fantastic IPC.

>>60389531

You are being a fanboy. Ryzen IPC is close but a little behind Intel's best. Intel's clocks are a little higher. For single thread performance I see nothing that suggests Ryzen or derivatives is going to win that.

Now throughput and certainly throughput/$ - fuck AMD is going to kill it.

I regularly am limited by single threaded software, and I often have idle cores, even on my old 4C8T box, so I don't see AMD as an automatic win for my next workstation, and when I jump it will probably be a 5 figure purchase or close to it.

>>60386144

If you didn't have the money to trash just for the premium binning of 1800x or the intelligence to realize getting a 1700 and OCing it is better, you don't need these workstation CPUs.

>>60389755
>ryzen is either almost as fast or much faster than kaby lake despite kaby lake cpus running at 5GHz when a 1700 is below 4 GHz
>"Ryzen IPC is close but a little behind intel"

>>60390704

I stick with what I wrote, Ryzen IPC is not at Intel's level:

http://wccftech.com/amd-ryzen-4-core-benchmarks-intel-core-i7-7700k/

Close but no cigar.

>>60391089
timings aren't even the same

1800X 16-15-15-35 1T
7700K 16-16-16-35 2T
Even with inferior timings it had better results?

>>60391371

I did not notice that, but yes, that is how I read it.

I don't think these changes give Ryzen too much of an advantage in those tests, but in the end I read Intel as winning.

That's not to shit too much on AMD. Ryzen is an amazing CPU. It is unbelievable that this tiny company can punch with the heavy hitters. It is going to put pricing pressure on Intel. 12/10. Will I buy some AMD boxes this year? probably.

>>60391371
in apps like winrar, 7700K gets 43% boost from HT whereas the 1800X only gets 20%

in cinebench R15, it has stronger HT vs the 7700K

On the 7700K you could even run 4133 17-17-17-37 1T vs 3200-3600 on the 1800X