How is Naples going to do against the new Skylake Xeon processors?

If Skylake Xeons has only 44 lanes it's really no contest.
Also both core count and clockspeed are in AMD's favor.

File: 1492893633027.png (158KB, 2033x909px) Image search: [Google]

158KB, 2033x909px

Lets just say.. Intel has a lot of work to do.

>>60094205
What do you mean by 44 lanes? I don't know a great deal about server chips compared to desktop.

I've heard Intel has a 28 core, and I think there's a 32 core Xeon, but it might be a custom chip for Google only?

>>60094225
PCIe lanes, from all sources all I'm seeing is 44 free PCIe lanes per CPU.
Naples has 128, and around 116 after putting the basic stuff like IB/SAS/SATA/USB/IPMI on the board

>>60094154
It'll be much worse; way more PCI-E lanes even on a single-chip system, "free" 10GbE from the chip, more cores at higher clocks and relatively low TDP across the range, and it'll likely be much cheaper too.
AMD always intended for Ryzen to be a highly-scalable chip intended for server-level expansion and this is where it'll get to stretch its legs.

>>60094154
Every single technical spec is in AMD's favor:

Clocks
Perf/watt
Core count
I/O
Memory bandwidth
DIMM count


Nontechnical aspects like price and support are up in the air though, I've seen plenty of frustrated Xeon customers in the last few years as Intel has been acting more and more like Apple

This shit might be good for 4xSLI with all of those lanes.

>>60094236
Oh right. So does that mean much slower bandwidth to the rest of the system?

>>60094264
Dude Radeon Instinct lmao.

>>60094154
Naples will skullfuck Shitlake Xeons just like K8 Opterons killed Itanium.

>>60094265
That's one way of looking at it, but it generally means far less connectivity, so your ability to put in extra stuff like more SAS devices, more 10GbE+ NICs, more direct attached GPUs, more FPGAs is severely hampered by lack of lanes.

>>60094154
Naples has so much fucking I/O there are actually motherboards with a PCIe x32 slot, that means you can put a fucking riser card and plug in TWO FULLY CONNECTED x16 GPUs on a single riser, all pulling connectivity from the CPU.

That's impossible for a Xeon, if you're going for most compute per node, you can actually put in SEVEN fully attached GPUs in a Naples system without even looking at the chipset or PLX chip.
It's fucking insane and outside of some very specific mainframe and academic CPUs I've never seen anything like that.

If it only ended at that Xeons would somehow survive, but they also lose in DPC, bandwidth, clocks and core count on top of that.

>>60094335
>custom motherboards with 8 PCIE slots


Get to fucking work Supermicro

>>60094335
Imagine next gen AMD increases from 32 and up to 40 lanes per die, they'll gonna have to make motherboards filled with side mounted PCIe slots, possibly some going on the other side.

This market is way more interesting than the pleb shit the client market gets.

>>60094335
I wonder how much money will Intel spend on bribes.

>>60094399
40 lanes in consumer CPU sounds very good even for an average consumer.

>>60094400
For starters they already lost their biggest customer, Microsoft.

Both from AMD's side and ARM side.

http://www.datacenterdynamics.com/content-tracks/open-data-center/ocp-summit-microsofts-project-olympus-will-support-amds-naples-cpus/97955.fullarticle

File: JUST krzanich.png (1MB, 1002x1480px) Image search: [Google]

1MB, 1002x1480px

>>60094421
>Intel already lost MS
Who's next?

>>60094439
His job.

>>60094441
I fucking wish.

>>60094421
Just google "Microsoft Olympus Naples" into google and look at the pretty motherboard pictures.

File: 1491563265928.jpg (257KB, 800x559px) Image search: [Google]

257KB, 800x559px

>>60094451

File: Qualcomm.png (378KB, 623x349px) Image search: [Google]

378KB, 623x349px

>>60094421
Here's the ARM ones, based on the Qualcomm one.

Too bad they really can't compete on memory amount, but at least these will do real good for fileservers.

>>60094463
THATS A LOTTA DIMMS!

File: microsoft-olympus-qualcomm-centriq-motherboard.jpg (124KB, 857x463px) Image search: [Google]

124KB, 857x463px

>>60094478

>>60094487
32 FUCKING DIMMS ON 2P BOARD.

File: AMD-Naples-Server-Block-Diagram.jpg (166KB, 1117x548px) Image search: [Google]

166KB, 1117x548px

STH managed to snag a block diagram of the mobo layout.

>>60094263
Well if it's anything like that HEDT processors, they'll be half to one third the cost of Intel's with the same performance.

>>60094529
If you go bigger than SMB then price doesn't really matter all that much, but that's still a pretty large part of the market where price does matter.

>>60094487
4U

>>60094529
AMD can throw a hefty price premium on Naples just because it's better than any Xeon in existence.

>>60094559
What's SMB?

>>60094571
Small-Medium business

>>60094564
YOU'RE A BIG SERVER

>>60094529
The server scape doesn't work line the consumer mind does.

So long as it's cheaper to run/cool (Which naples is) then that's already some customers right there.
Then there are those that simply need more cores, so naples again.
Those who want more PCIe lanes. Naples.
etc. etc. etc.

The trouble Intel has, is that they rely on massive, perfect dies to produce their 12/16 core products, whereas AMD only needs 3 good quad core dies, and to staple them together with the Infinity Fabric. Basically guaranteed 12/16 core products.

>>60094585
There are no quad core dies. 16 core HEDT Zen is 2 full Zeppelins and 12 core is one with 2 cores disabled on each die.

The 32 core Naples will be hovering around 2.5-2.9GHz base at 180W TDP, but that's either a B1 or B0 stepping, Naples parts will be B2 so they can either go higher on clocks or lower power.

For comparison the 28 core Intel is at 2.5 base at 205W TDP
Literally no contest.

>>60094596
The simplest way of explaining the Zen architecture is to describe each CCX as a quad core, and simply to say they're glued together.

Which isn't wrong.

>>60094614
It is wrong, Zeppelin is a single fucking die.

But where are crying wojaks and gaming benches?

>>60094652
He tired himself out last night trying to convince us the 7740k is gonna OC to 6.0GHz because it has a disabled iGPU

>>60094659
Poor guy tries his hardest to shitpost and derail any AMD-related thread.

>>60094666
His economical status depends if he's getting paid or not.
I mean I really hope he fucking does.

>>60094683
Or he's just autistic.

>>60094559
Wouldn't everyone just use Xeon Phis with a ridiculous amount of cores if that's the case?

Like why even bother with 32 core processors if price doesn't matter?

>>60094742
A lot of people here are autistic but they aren't that desperate.

>>60094754
Xeon Phis are using duct taped Silvermont cores, they are pure shit for anything besides very specific tasks.

>>60094754
Because Xeon Phis are co-processors, with a limited use case.
Not many things will be better off off-loaded to a large number of hilariously weak cores.

>>60094754
Because PHis use silvermont cores and PHis are accelerators, much worse one than Teslas that can handle only a specific type of workload.

tl;dr you're talking about things you don't understand.

>>60094756
He's alone and he wants attention.

>>60094768
>>60094767
>>60094764
I was asking a genuine question, I don't know much about server processors.

Thanks

>>60094784
This should have been easily found in a search engine.. Oh well, I guess it's better than shitposting.

>>60094791
Fucking anything is better than THAT retard's shitposting.

>>60094768
>>60094764
>>60094767
Remember when Intel promised XPHIs will be PnP and all you'll have to do is modify your headers to use it?

It turns out they were full of shit and PHIs need as much work as Teslas to get working with your code but offer far worse performance and perf/watt

File: IMG_3359.jpg (92KB, 687x683px) Image search: [Google]

92KB, 687x683px

>>60094284
Lul no

Don't fall for the bait.

>>60094791
researching the applications of different server processors? a lot quicker to just ask a question.

File: IMG_3360.gif (2MB, 177x150px) Image search: [Google]

2MB, 177x150px

>>60094603
>32 amd cores

You mean 8 4core ccxs?

>>60094805
Larashit was a gigantic failure and everyone knew that. No amount of Intel's damage control and promises would have changed that.

File: 1493236290964.png (42KB, 653x726px) Image search: [Google]

42KB, 653x726px

>>60094800

File: IMG_3211.jpg (620KB, 2150x1240px) Image search: [Google]

620KB, 2150x1240px

>>60094825
Not as big of a failure as poozen

Oh boy this guy is here.

Someone woke up.

Why, oh god why this retard found this thread. Its about servers pls fuck off.

I wonder what this guy thinks when he looks into the mirror.

>>60094831
>Day one charts
Neck yourself, jew.

File: Panorama.jpg (605KB, 2550x1446px) Image search: [Google]

605KB, 2550x1446px

>>60094808
>>60094831
still posting day 1's kiddo?

Oh look he's samefagging.

>>60094860
>day one

Is you retarded raja? That's a benchmark from April 20th, less than a week ago.

https://youtu.be/D_Yt4vSZKVk

Another decent thread bites the dust. See ya dudes.

I'm out, see ya when this monkey falls asleep.

>>60094891
Don't forget to POO IN LOO

>>60094874
>Only posting RotTR and Farcry the only two games that really show a hit on performance.
>Linus Shilltips.
Discarded

File: 1488679084736.png (84KB, 653x726px) Image search: [Google]

84KB, 653x726px

>>60094874
How much do you get paid? Is it worth applying for a job?

File: IMG_3364.jpg (936KB, 2208x1178px) Image search: [Google]

936KB, 2208x1178px

>>60094901
Check out this Watch Dogs 2 benchmark. Intel 4 core trashes amd poozen 8 core

>>60094923
So pretty much every badly coded game out there? Everyone knows Farcry Primal, Watchdogs 2, Fallout 4 and Rise of the Tomb Raider are really poorly coded and rely heavily on single thread.

File: IMG_3365.jpg (594KB, 2141x1214px) Image search: [Google]

594KB, 2141x1214px

>>60094942
What about GTA V?

>he's talking to himself

I think mommy is concerned

>>60094956
see
>>60094865

>>60094977
>testing cpu performance
>with a 970 that will bottleneck the cpus

I always knew indians were stupid lmao

File: watchdogs[1].png (51KB, 730x603px) Image search: [Google]

51KB, 730x603px

>>60094923
>Still performs nearly as well on your cherry picked game that is heavily Intel biased.
Neck yourself

>>60094989
So why is Intel still beating it in some games? If that was the case they would be exactly even fps. Besides. Steam still says the 970 is the most used GPU out there.

>84 posts
>13 posters

rly mks u thnk

>>60095029
You won't find more than 6 people on this board who know anything about server hardware or the market so it's expected.

File: 1493283204085.jpg (61KB, 500x667px) Image search: [Google]

61KB, 500x667px

>>60094923
>>60094956
OY VEY! I SMELL SHEKELS!

File: Screen Shot 2017-04-27 at 13.26.57-fullpage.png (1MB, 1899x8427px) Image search: [Google]

1MB, 1899x8427px

You think Intel is livid that out of shitload of reviewers only 3 didn't recommend Ryzen?

>>60095097
They're far from pleased, pulling in the launch of Skylake-X, Coffelake and KBL-X an entire quarter should tell you something.

How old is Kabylake? Wasn't it released in January? When was the last time Intel released a new generation after 5 months?

>>60095131
They released something that performs better than poozen before amd even had a chance to release it

>>60095097
>Indonesian, Chinese and Japanese

Lel

>>60094808
Can't you just fuck off? It's obvious that you understand fuckall about servers.

>>60095213
You don't understand fuckall about logic. Do you think if amd gets RAVAGED by intel in the consumer segment they will do any better in the server one?

>>60095097
TOTALLY WRONG

https://www.youtube.com/watch?v=0lmEnA8pzHA
https://www.youtube.com/watch?v=0lmEnA8pzHA
https://www.youtube.com/watch?v=0lmEnA8pzHA
https://www.youtube.com/watch?v=0lmEnA8pzHA

>>60095233
It's obvious that you're a retard if you really think that way.

>>60095213
Didn't your mom ever tell you not to argue with mouthbreathers?

File: IMG_3347.jpg (85KB, 1277x683px) Image search: [Google]

85KB, 1277x683px

>>60095245
Benchmarks speak

>>60095269
Unfortunately so do you.

File: 1493292221665.jpg (2MB, 766x8714px) Image search: [Google]

2MB, 766x8714px

>>60094154
Intel's efforts against the new Opterons will be a hilarious thing to spectate, I hope they try their hands at "talking" with Dell again.

File: ibm-hot-chips-power9-die-block.jpg (254KB, 708x655px) Image search: [Google]

254KB, 708x655px

>>60095321
Just wait when IBM fucks them from the other side as well.
Look at this, an annoying similar core complex layout to Zen, even IBM doesn't believe in the monolithic die nonsense.

>>60095097
Intel is complete garbage at literally everything, news at 11.

>>60095097
Where does it say recommend? It lists what tests were done.

File: IMG_2742.png (1MB, 1228x1502px) Image search: [Google]

1MB, 1228x1502px

>>60095348
JUST WAIT AMD WILL BEAT INTEL

>>60095348
They're going even further than Zen, each 2 cores has its own L3 cache and a shared L2 cache.
This is a pretty nice scalable design, 60MB+ chip wide shared L3 compared is just stupid

File: apustaja_silicia_gel.png (261KB, 1039x559px) Image search: [Google]

261KB, 1039x559px

>>60094808
>Posting a gaming benchmark
>In a thread about servers
Holy shit you've got full retard.

>>60095377
Gayming

>>60095391
That Intlelfag is legitimately clinically retarded.

>>60095414
Or just paid to shitpost and thwart discussion with pure noise.
Thanks, nonexistent mods and jannies

File: unknown (5).png (2KB, 117x78px) Image search: [Google]

2KB, 117x78px

this thread was so good, what happened?? i think intel really sent in a shill squad. is gaming really all intel has? jesus what a shame..

imho dropping support for FMA3/4 for the sake of removing "legacy bullshit". after it use was shilled so much is a fuckhuge mistake.
software packages for simulation are fucking behemoths and support for AVX won't be implemented in 2 weeks

>>60095452
If people would take the appropriate response to blatantly off-topic shitposting and spam instead of feeding him (You)s, maybe we might actually be able to have threads not totally derailed by his trolling.

Friendly reminder to report all shitposters.

>>60095765
lets talk more about naples, then. it looks fucking kickass. i dont even understand server shit but all of amd's numbers are > intel's numbers which is gonna be fucking awesome to watch go down. its like a multi billion dollar multi silicon cock fight, and theyre playing capture the flag where major hardware manufacturers are the flags. if microsoft is dropping intel, whos next? google, facebook, comcast, verizon, blah blah blah they all probably use intel's processors. now that naples is happening they have a REAL good reason to drop intel processors. the entire technology world is gonna flop, because when those bigass companies say "yeah we use amd hardware to power your favorite websites" that leaves a fantastic impression on people. soon you'll see "amd inside" on some of the laptops in your local best buy or whatever. when normies shift to different products the entire world shifts with them, theyre where all the money's at in the consumer market. amd is gonna be making big bucks and we're gonna start seeing shit hit the fan for intel hard. console, soon server, and later desktop market control, what even is there to take next? intel got way too comfortable lol.

>>60095877
It's just good to see genuine innovation and competition again, something to push Intel away from getting fat and happy off a dead end strategic approach that has been holding back x86-64 CPU design in every market segment.

>>60094569

But they won't. They have a pathetic server market share right now, and they will want to make it a no brained. They will probably undercut intel by 15-20% for a chip that performs better and then they will start rocking the shit out of the server market.

>>60094263
There is one way that Intel has an advantage, which is with AVX2, AVX256, AVX512.

But the perf/watt of Ryzen doing AVX2 is still pretty damn close despite doing it at half speed. And AVX256 and AVX512 is hardly used.
Intel is suddenly only selling to a niche market.

>>60095429
Mods don't give a single solitary fuck about this board.

>>60094529
>>60094569
>>60096121
$4000-$5000 would be a perfectly reasonable price. Similar to the 22 core Xeon prices
They compared it to a ~$4500 Xeon.

Like the other Anon mentioned, enterprise doesn't think like the consumer. They look at cost over 3-5 years and make a bottom-line decision (as long as there is no bribery).

>>60096191
Is still lol for pricing to slot in half way between the opterons and XEONs. Whatever happens, we need a silicon price war. That price is only deemed acceptable by anyone because Intel has been gouging the market for years.

>2 0 5
>0
>5

>W A T T
>A
>T
>T

>T D P
>D
>P

>>60096258
Well AMD could sell them at $1000 and still make money. It's literally 4 1700s put together which probably cost ~85$ to make.
Intel can't price theirs that low. Intel would be stupid to attempt to bleed money in a price war as it means that if they do become competitive in enterprise again they would have to introduce their new technology at that price they lowered.

$4000 for 32c/64t would be beyond reasonable as it is.

>>60096326
They put 4 1800Xs together due to lower voltage

File: 12345241332.png (438KB, 1064x1619px) Image search: [Google]

438KB, 1064x1619px

>>60096326
because this

>>60096650
Yeah alright, 1800X's. Maybe even higher binned than 1800X.
But 1800Xs are sold at a huge premium over what they cost to make which is probably well under $85 for the die, chip pcb, IHS, solder, cost to GloFo, etc.
But $4k is reasonable given the R&D costs and to enjoy a profit margin compared to competing chips.

>>60096899
And Nvidia and Intel are both botnets.

>>60096899
ok

>>60096899
Sure is mods giving a shit about cleaning up cancerous offtopic shitposting in here.

File: Intel_xeon_nextGen_specs.jpg (452KB, 900x1397px) Image search: [Google]

452KB, 900x1397px

>>60094154

>Xeon Platinum 8180
>US$ 12K

>>60094439
I've heard jewgle putting naples in their cloud servers too

Naples will not be competitive with skylake-ep in the slightest. 8c summit ridge already struggles to compete with has/broadwell-ep which is 10-20% slower clock for clock than skylake and has serious issues with communication between CCX, now that latency will be 10-50x worse to do die-to-die communication and socket-to-socket communication.

amd really needed to get zen out 6 months sooner to be real successful with it. but hopefully it will claw them back a little bit of the market and keep them afloat until their next design is ready.

>>60095399
So they "recommend" it for "Gaming", "memory testing" and "overclocking"? Are you all retarded here? Those are descriptions of the reviews.

>>60097836
Everyone (especially ML fags who want to attach as much GPU's per rack as possible) will be using Naples.

>>60097904
>ADD MOAR BINGBUS :-----DDDDD
>EBIN:-------D

>>60097904
>Naples will not be competitive with skylake-ep in the slightest
[Citation Needed]

>>60096125
AVX is forever unpopular because the best selling Celeron doesn't support it, duh.

Guy who actually decides what hardware vendor to go with and how deployments are rolled out here. If AMD wants my bidness;
a) Actually compete once real benchmarks are out, not this proto-hype bullshit with no real hard numbers (Ryzen vs top spec Intel chips is eh)
b) Power envelope is actually what they say and they don't chug power like every other AMD system since back to the Sledgehammer days
c) They offer real support to my software devs when shit doesn't work correctly
d) ESXI runs as well as the Intel products, meaing LRO works correctly, SMT works correctly (30% performance loss on Ryzen right now with it disabled) Yes that's right they have to not be pussies and work with Vmware to get their shit dialed in. This is part of something a company like them has to do in the current marketplace.

>>60099926
>SMT works correctly (30% performance loss on Ryzen right now with it disabled)
On what workloads? I've only seen this as an issue when Ryzen had literally just launched.

>>60100251
Have an 1800x with a Gigabyte GA350m board and 64GB of RAM. It pink screens on 6.5 unless I use the workaround.


Workloads are: I hooked this straight into my foreman/puppet install and am using it to test launch speed of VMs. If these are competitive I'll put our dev environment on them to start. Workloads inside of said machines are java based on our product stack.

>>60099926
>b) Power envelope is actually what they say and they don't chug power like every other AMD system since back to the Sledgehammer days
So a 91 7700k drawing 140watts in just normal usage is okay? 185watts doing AVX256, that's okay?
Idling at 45-55c, even with an AIO or big 140w TDP+ cooler, is okay?

Ryzen CPUs actually conform to their TDP ratings reasonbly. You really only need a 95w TDP cooler for the 95w TDP CPUs.

>d) ESXI runs as well as the Intel products, meaing LRO works correctly, SMT works correctly
lol? HT was giving a 45-60% performance loss on Sandy Bridge in Crysis.
SMT usually gives a performance gain. When its a performance loss, it's usually 10% at most. And even then lots of those seemed to have been ironed out already.
And when it works, which is usually? It's over 30% better than HT.

>>60097904
Thank for for supporting Intel in a rational way.

I'm rooting for Naples for a prosumer box. My e3 haswell would have become broadwell ep x99 but it wouldn't have helped with the multigpu pass through enough so I scrapped the build. I'm waiting for skylake-x vs Naples benches now for who can handle 4x x16 GPUs best with breathing room. Right now Naples is winning (and skylake will have problems single cpu) but until release I will continue to wait for actual data.

>lol cpu du jour btfo other guys on suicide watch just wait

>>60100379
Try a different motherboard? Haven't heard good things about Gigglebyte. Their real server stuff should be much more stable than launch tier motherboards anyway.

MODS == GODS

>>60100853
I'm not talking nor interested in desktop workloads like games. I'm talking about server production workloads. Desktop market doesn't matter, only the server one does. AMD needs to not fuck up their server release.

SMT doesn't work on the Ryzen 7 for some reason in ESXI 6.5 or any other version and gives me a PSOD. I'm sure it will eventually be worked out, but the performance still isn't amazing on a core per core basis.

>>60101434
Yeah I already ordered a different one with the x370 chipset instead of the 350 to see if it behaves any different.

>>60102680
Xeons burn doing AVX256 and AVX512 as well.
Servethehome tests showed Ryzen being absurdly more power efficient even at clocks higher than it's really efficient at.

>SMT doesn't work on the Ryzen 7 for some reason in ESXI 6.5 or any other version and gives me a PSOD.
Okay, that's fair. I think that's a good deal why Ryzen 7 launched so much earlier, and even Ryzen 5, before Naples and Snowy Owl.

>>60096125
AVX2 performance is identical if not faster, AVX 256 is half speed and AVX 512 is basically only supported by Xeon Phi's

File: cpu.jpg (301KB, 1600x900px) Image search: [Google]

301KB, 1600x900px

About how much will an entry level Naples system cost?
If I am on socket 2011-V3 would I be better off just buying a 2.0Ghz 20 Core E5 CPU for $600.

Would the low clock speed effect performance too much?

>>60102822
You're simplifying it by calling it half speed, which it's not.

Zen has twice less registers than Skylake for AVX256 ops, it can still do a 256bit OP per cycle, just not two of them at once.
And unless your pipes are fully saturated (which they never are) it's far from a double decrease in performance.

I think there was some tests done on this and Zen only ended up 10% slower than the 6900k over a 5 minute long workload.
I don't even need to tell you how much less power it consumed

>>60094154
I'm too lazy to start a new thread for this so fuck it, it's going in here.

http://www.tweaktown.com/news/57293/intels-new-xeon-rocks-28c-56t-costs-over-12-000/index.html

"Intel's new Xeon rocks 28C/56T, costs over $12,000"

TWELVE
THOUSAND
DOLLARS

File: JUSTnich.png (1MB, 1002x1480px) Image search: [Google]

1MB, 1002x1480px

>>60102894
>TWELVE
>THOUSAND
>DOLLARS
>for less EVERYTHING
JUST.

>>60102894
Patrick Kennedy puts availability in Q3, anandtech even thinks the whole thing was a fuckup and it wasn't supposed to be shown yet, since there's no SKUs on ark.intel and there's only clockspeed and opn codes on the site.

Also, H0 stepping for server parts sounds very suspicious.
Either way you're not gonna see this thing benchmarked until August

>>60102770
So I have used the E5-v4 lineup in a SDN (esxi
+ vyos + pfsense + nginx proxy) and was still able to pull 7gbps over solarflare cards before running out of CPU (Using 2637v4's with pfsense over openvpn ssl tunnels). Honestly if you need more juice you're better off using a dedicated crypto card than trying to hack it with shitty aes-ni and avx acceleration.
I don't have any other workloads that would stress that instruction set other than ssl.

Look, I hope AMD blows the fuck out of Intel because the performance deltas between v1 and v4 xeons on a single core basis have been pretty little. For workloads that require extremely fast single core performance or high frequency (think HFT or above SDN) the only option has been to go with these smaller Intel chips. The problem I have is, their marketing hype has had too many false starts over the past decade. Their products have sucked and I think they know it. I also have a problem with the lack of engineering samples being provided to vendors to build interest in their product. I have to build an expansion at one of the DCs by the end of this calander year and I'd love to use AMD if its competitive, but so far I haven't seen anything compelling to make me hold off on ordering some tried and true e5-v4's.

>>60102894
>$12k for a 28c CPU
I see opportunity for AMD here. They could easily price a fully kitted out Naples system of 64c/128t for that much, maybe 1-2k more for a higher capacity set of ram, and STILL make profit.

This would murder Skylake-E/EP/EX

>>60102992
They don't need to, the people buying these processors don't care about SKU price, they care about TCO

Going into technical differences between naples and Skylake-E* is pointless since Intel ends up looking beyond silly.

>>60102848
>About how much will an entry level Naples system cost?
Naples is not entry level. It's 32core.
You're thinking Snowy Owl for the 16 core <100W TDP. Probably still expensive but the HEDT 12c or 16c and motherboards should be far far cheaper.

>>60102938
True or false, it's still a fuckup.

>>60102848
You're better off waiting for the 12-16 core Zens that are gonna be clocked at like 3.4-3.5 base, for a HCC CPU it'll have plenty of ST performance and absolutely murderous MT performance.

>>60102924
>less EVERYTHING
no even Intel camp but since when did anyone else have a 28C/56T x86-64 processor? And I specifically say x86-64 because Power 8 with 12C/96T is stupid.

>>60102992
Or just a 40 or 48 core.

Problem is that I bet they had limitations with the chip PCB for the connections.
And limited bandwidth of the infinity fabric which starts to use a lot of power when you crank it up to keep pace with 8 channel DDR4.
So na I don't see more than 32core happening until 7nm.

>>60103152
>So na I don't see more than 32core happening until 7nm.
It's already happening you braindead monkey.

>>60103172
>It's already happening you braindead monkey.
Source? Obviously I mean for AMD.

>>60103256
Literally video from Naples presentation. They benched OEM 2P racks: one Intel-based and one with Naples.

File: 1487463821177.png (2MB, 1440x900px) Image search: [Google]

2MB, 1440x900px

>>60103272
Uhh those were 32 core CPUs. 2 of them.
32 is not more than 32.
>braindead monkey

File: 6107bbdb95ceacca71b24da34fba91f4bdf6bd45.jpg (42KB, 620x348px) Image search: [Google]

42KB, 620x348px

>>60094399
>New more compact forms of PCIe have to be invented because THERE'S SO MANY FUCKING LANES

>>60094412
40 free lanes, once you add in the chipset, sata, usb, etc that all use their own lanes you're suddenly down to 24-28 usable lanes.

>>60103152
>And limited bandwidth of the infinity fabric which starts to use a lot of power when you crank it up to keep pace with 8 channel DDR4.

Ryzen's data fabric is a 256b crossbar clocked at ~1.1-1.6 GHz (~35-50 GB/s) on desktop and probably 2.2-3.2 GHz (~70-100 GB/s) on Naples. It simply can't "keep pace with" 8ch DDR4 (up to ~150 GB/s), nor does it need to, since that DRAM bandwidth is for 8 CCXs to share, not what any 1 is reasonably expected to consume.

>So na I don't see more than 32core happening until 7nm.
A single 4c CCX = ~22% of a Zeppelin die. Bolting on 4 more core would still yield a die under 250 mm^2 without a litho upgrade, which is still within reason. And it's not like 3 DDR4 channels per die will ever happen, since a 19" rack mobo can just barely squeeze in 2 sockets and 32 DIMMs, and 1DPC would be a non-starter.

>>60094264
>not 11x Samsung 960 pros

pleb

>>60103509
I sincerely hope they'll clock IF higher for Pinnacle Ridge.

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

3MB, 1651x1475px

>>60094412
>>60103498
Even enterprise Ryzen chipsets will only use 4 PCIe lanes.

That said, it would maybe be smarter (or at least more aggressive) for AMD to put another block of 8 to 16 IO lanes in Zen+/Zen2, since it would allow a 3rd xGMI link for maximally dense 4P (= 128c/256t+ 64 DIMM) systems.

>>60103555
Rather than that they should focus on getting DRAM latency down, it's almost amazing how well Ryzen is doing with generally 15% lower DRAM bandwidth and a good 30%+ higher DRAM latency.
That Agesa update a while ago that lowered dram latency? Straight 2-4% performance increase across the board, and that was only 6ns

>>60103578
Man AMD struck an absolute gold with Zen AND GloFo god IBMs foundry stuff so they will never ever fuck anything up.
>>60103598
I wonder what will may AGESA update bring besides native support for 3000/3200mhz RAM.

>>60103555
Any Zeppelin successor that adopts 256b AVX will be forced to use a 512b IF switch just to prevent data starvation.

Even current-gen Naples and 2-die HEDT will have to use 1:1 DDR4:IF clocking just to allow 4+ channel memory interleaving, so the low clocks for R3/R5/R7 is for power saving alone, not design limitations.

AMD could allow doubled IF clocks tomorrow, but presumably AMD would rather Ryzen look like a slightly unorthodox architecture to tune (and get devs started on optimizations/fixes now) than like an energy pig.

Zen is fucking awesome but now i want Vega.

>>60094154
as with all other amd products id rather wait for review

>>60103509
> It simply can't "keep pace with" 8ch DDR4 (up to ~150 GB/s), nor does it need to, since that DRAM bandwidth is for 8 CCXs to share, not what any 1 is reasonably expected to consume.
But this is what I was saying. 70-100GB/s shared between SIXTEEN CCXs instead of eight wouldn't be enough in many cases.
You would get diminishing returns on the scaling in a number of applications.

>Bolting on 4 more core would still yield a die under 250 mm^2 without a litho upgrade
Umm I never said anything about die size.
I'm talking about the CPU package. The PCB the dies are on and the wiring there.
They'd probably need a second PCB layer to wire them together.

Like you addressed nothing I actually said there. You misread it horribly even if you made decent.. irrelevant.. points.

>>60103701
I frankly hope they don't adopt AVX256. Waste of fucking space on the arch for a consumer CPU. Let Intel have that niche market in servers and pigeonhole themselves into it.
I'd rather just have more fp32 and fp128.

I need AMD to make businesses flood the used market with cheap decommissioned intel parts.

>>60104699
Holy shit I didn't even think of that
cheap used parts would be great

>>60104699
>cheap housefires
Meh, I'll wait for 7nm 32cores to come out so I can grab a cheap Naples.
Xeon E5-1680 v4 is basically just a 6900k, isn't it?
All these models seem fairly useless to me when I could just get an R7 that is going to be way less trouble, still cheap, same clocks, lower power consumption, etc.
Look at The Xeon lineup since Broadwell and later. Garbo stuff.

Or I'll just buy AMD HEDT 12 core since they don't gimp their consumer products like Intel does.
We may even get consumer 2p boards with 64 PCIe lanes or more.

>>60104462
>Umm I never said anything about die size.
>Like you addressed nothing I actually said there. You misread it horribly even if you made decent.. irrelevant.. points.

I think you might be in over your head. I disregarded the "PCB" parts since you clearly have only the roughest idea of what you're talking about:
1) All the components on a CPU die, Zeppelin or otherwise, are wired to each other through the dozen-ish metal layers deposited and etched on top of the base transistors.
2) CPU dies aren't on a Printed Circuit Board, they're on a "substrate", or in the case of MCMs like Naples, a silicon "interposer" on top of the substrate.
3) If you meant that a 48c Naples with 6*Zeppelin would need more layers in the interposer for GMI links, you might actually be almost right, but:
- Zeppelin doesn't even have 5 separate GMI links for a fully connected 6-die mesh
- an extra metal layer on an extremely coarse litho'd interposer is cheap as shit
- the bigger problem is making sure interconnect regions fit inside the reticle limit
- you're blindly ignoring the potential of 12c Zen(2/3) parts that would have none of these issues

My core argument is that:
1) There will probably be 2 more incarnations of Zen before 7nm is ready.
2) There is plenty of opportunity to make 4-die MCMs with 12c each within those generations: adding 50% more cores+L3 and not IO would only increase die size by about 22%, which should still manage to fit on the rather large Naples packages and not actually create much if any more interposer complexity.
3) 12c/2-channel and 48c/8-channel is in the same compute:bandwidth ballpark as upcoming Platinum Skylake Xeons and not somehow grossly deficient.
4) IF doesn't need to clock up much more, since 70-100 GB/s is more than any sane application can sustain with the 4 cores in a CCX.

We may very well not see 48c CPUs before 7nm, but your justification was garbage.

>>60104992
You're the one that's over your head.

>2) CPU dies aren't on a Printed Circuit Board, they're on a "substrate"
Lmao the substrate is literally a PCB. You just used a different word and are too stupid to know they're the same thing.

> or in the case of MCMs like Naples, a silicon "interposer" on top of the substrate.
Those are called FUCKING WIRES on the printed CIRCUIT BOARD. Again, you're just too stupid to know it's the same thing. You're just repeating Naples marketing buzz words.

With more wires on the PCB connecting zeppelin dies, going from 4 to 8, that means more wires. More wires means more layers since they can't overlap touching each other.

Holy fuck, you type a lot for someone stupid.

File: stacking.png (392KB, 1194x1746px) Image search: [Google]

392KB, 1194x1746px

>>60105195
jesus.

literally NOBODY calls even a multi-layer organic substrate a PCB.
and silicon interposers aren't even that, just like ASICs aren't "printed circuit boards" even though fabrication BEOL is printing copper layers.

try not to be such a posturing imbecile, it will get you nowhere in life.

File: vocab for neophytes.png (428KB, 1217x1825px) Image search: [Google]

428KB, 1217x1825px

>>60105195
>>60105718
> (continued)

>>60105718
>it's called a PCB right there
It's umbrella for that whole part of the package.

>>60094236
>CIe lanes, from all sources all I'm seeing is 44 free PCIe lanes per CPU.
>Naples has 128,
Pretty pointless for most applications as 7x 16 PCIE slots are 112 lanes. And even then most server cards are half height, half length and 8x lanes.

>>60094263
>Every single technical spec is in AMD's favor:
Except software support, for instance ESXi wont boot on it. And Ryzen in theory supports ECC even though there isnt a single system which supports it - AMD blames motherboard manufacturers, which is sort of like the jews blaming everyone else for them being kicked out of every country.

>>60094335
>Naples has so much fucking I/O there are actually motherboards with a PCIe x32 slot,
This is fairly common for risers, its been around on Intel boards since forever.

>>60094767
>Because Xeon Phis are co-processors, with a limited use case.
Not since Knights Landing which can boot windows natively.

>>60102894
And it scales to 8 sockets you retard. Naples isnt even trying to compete with any E7 chip, or the E5-4600s. It is competing against Intels Mid/Low-end E5-2600

>>60103498
>chipset, sata, usb,
On Intel chips the chipset is on its own bus, and actually proivdes all of the above.

>>60105785
can you not understand the package ends where it says "package substrate"?

I know you're a little slow, but it shouldn't be that hard to guess that the big circles are solder balls that connect the package to an actual PCB.

>>60105832
You do know what "package" means and why it's called "package substrate", right? Since you asked
>can you not understand the package ends where it says "package substrate"?
it seems like you don't.

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

211KB, 1125x1037px

>>60105830
Naples will have a ton of IO, but it doesn't take too much creativity to think of things that will be done with it.

I can guarantee you that Supermicro will make a 2U with 20 u.2 drives, which would eat up 80 lanes before you even get to the networking.

>>60094505
> tried to repost this pic and got cockblocked

>>60105880
The PCBs in those diagrams should be understood to be motherboards, AIBs, etc...

File: 2028R-NR48N_front.jpg (99KB, 960x720px) Image search: [Google]

99KB, 960x720px

>>60105896
>I can guarantee you that Supermicro will make a 2U with 20 u.2 drives
Where have you been? They already make 2U chassis with 48 U.2 drives. It still doesnt mean it will be used by anyone outside of niche use cases.

>>60105938
Holy shit how many PLX switches does that thing need?
Or do they just do 2 lanes per drive or something?

>>60106018
>Holy shit how many PLX switches does that thing need?
probably a ton


>Or do they just do 2 lanes per drive or something?
i cant even see where all the disks connect to the motherboard

https://www.supermicro.com/products/motherboard/Xeon/C600/X10DSC_.cfm

That card edge connector is on the IO panel side, so i'm assuming its for a NIC in a propritetary form factor

File: x10dsc.jpg (316KB, 2000x951px) Image search: [Google]

316KB, 2000x951px

>>60106057
I can only assume those vertical display port looking connectors near the left CPUare for the U.2 drives.

>>60105922
oh kek. That's a bad diagram then. It should have squiggly edges.
Fair enough.

File: 2028U-TN24R4T__top.jpg (111KB, 960x720px) Image search: [Google]

111KB, 960x720px

>>60105938
>>60106057
>>60106136
I think something like pic related or the 1U, 10 drive 1028U-TN10RT+ would work better for Naples, since they'd still allow a full width mobo for 32 DIMMs.

>>60102822

pretty sure one big new feature of the new skylake xeons is support for AVX-512. As for software support, the people that will buy boxes holding these cpus that need AVX-512 - well they are compiling their own code.

I think Naples is going to be successful - but not purchased by anybody for a workstation replacement. As a server cpu, filling Apple, Amazon, google, etc datacenters AMD will sell as many as they can fab (and how many that will be is an open question, imho)

The Naples cpu is going to be too thermally constrained to allow high clocks, so on top of its opps/tick deficit against intel and lower clocks, any sort of workstation task is not going to work well on Naples.

sooo... why are any of you excited about it?

When is naples being released? I want a 16 or 32 core version for my 3d modelling workstation.

File: Speccy.png (44KB, 660x558px) Image search: [Google]

44KB, 660x558px

>>60108183
>cpus that need AVX-512 - well they are compiling their own code.
Because MMX/SSE doesnt exist? And AVX-512 isnt just basically the next step and trivial to port to?

>>60108183
>sooo... why are any of you excited about it?
I'm not, but its not like there aren't people here who have workstations/servers.

>>60108250
>I'm not, but its not like there aren't people here who have workstations/servers.

My point is that Naples is not going to work well as a workstation cpu. That leaves only use for datacenter (what it was designed for). There is at most a single guy here interested in that use case.

I have been toying with dropping $10-$20k on some future supermicro DP skylake xeon workstation box. (now thinking about waiting for Optane in dimms). Looking at Napes as a possibility instead, I just can't see Naples as being anything remotely interesting for any kind of desktop use.

File: Screen Shot 2017-04-28 at 12.19.38 AM.png (1MB, 3360x2100px) Image search: [Google]

1MB, 3360x2100px

>>60108310
>There is at most a single guy here interested in that use case.
There are plenty of people here who need tons of ram and high core counts, me included. I just dont want a AyyMD chip because ESXi doesnt support it. 32 dimms in a chassis would be nice, but it does me no good if I cant run shit on it, and even if it was on the HCL, it will end up getting dropped a release or two later.

>>60108250
>Because MMX/SSE doesnt exist? And AVX-512 isnt just basically the next step and trivial to port to?

It's not just about the instructions supported - its about the number of execution engines baked in.

I look at where Intel is trying to push the x86 platform, and I see insane levels of innovation. It is not happening in clock speeds - we all know that is mostly tapped out. These vector extensions though seem to have potential to seriously bump per-thread throughput.

High core counts, vector instructions, deep power saving techniques, and the real biggie of persistent memory. These are all YUGE. Yeah AMD is doing some of the same things, but I view Intel's designs as being far more aggressive.

If Intel is right, (and I think they are), aggressively adopting their uarch innovations like avx-512 is important.

>>60108365
Anon, super-wide FP SIMD is antithetical to power savings. Xeons already have separate, lower clocks for handling AVX code so they don't literally housefire.

The primary argument for using Xeons or Knights Landing chips is floating point code with a moderate or high degree of incoherent branching, since SIMT designs like GCN and post-Fermi Nvidia already handle coherent branching code at much higher throughputs and energy efficiencies.

Simulation? Use Intel.
ML, graphics, or pretty much anything else? Use a GPU.

File: 03_big.png (49KB, 849x621px) Image search: [Google]

49KB, 849x621px

Intel is pretty much panicking due to Naples right now.
They couldn't care less if Ryzen takes away a few thousand HEDT sales from them. They do panic because of their server market share though.

Ryzen at lower clock speeds (~3-3.3GHz) has almost twice the performance per watt of any Intel processor.
For server applications. MOAR COARZ and TDP are the undisputed kings, while clock speed isn't. If you look at Intel's Xeon clock speeds and those Naples is supposed to run at, there is literally no difference.
Clock for clock Ryzen is only very few percent behind Intel too, while offering more cores at the same frequency and lower TDP.

Ryzen at 3.3 GHz on all cores, fully loaded uses around 50 Watts. At 3GHz it's in they low 40s. A 6900k at it's stock 3.4 GHz all core turbo uses around 100 Watts.
Let that fucking sink in. You could have twice the cores at comparable clock speeds with comparable per clock performance with the same overall power consumption at around the same price.
This is why Intel is scared, not because Ryzen performs around the same in muh gaymen.

>>60108429
>Anon, super-wide FP SIMD is antithetical to power savings. Xeons already have separate, lower clocks for handling AVX code so they don't literally housefire.

Yes... but nothing you have written is inconsistent with what I wrote. It all comes down to useful work / Joule in terms of throughput. The aggressive power savings that intel and AMD use across their designs just raise the power available for doing useful work.

If you are doing a matrix multiply, sure maybe keep it on a GPU, but:

As you raise the GPU issue:
iirc Intel can now share page tables with their gpu hardware. That means you can share a user space pointer with the gpu, and things mostly just werk. Data can be seamlessly shared between their gpu and cpu hardware. There was a (((Intel))) presentation on openCL where they compared it against a then high end nvidia gpu. The nvidia won the absolute performance tests, but Intel won most of the power tests, because the gpu still had to shuttle data in and out of main memory.

I view the on package DRAM caches in the phi as a feature that is going to be combined with the existing Intel GPU on almost every intel cpu. I am not sure at all that nvidia is going to be left with a lot of breathing room in a few years.

You will be able to move computation COST FREE and at the page level between a GPU, a CPU with serious OoO and deep pipelines, all in the same memory bank with several GB of on package super fast cache.

>>60108673
Dont forget Intel's acquisition of Altera and that eventually we'll see FPGAs on their CPUs as well.

File: happy_wojak.gif (14KB, 633x758px) Image search: [Google]

14KB, 633x758px

>in like 6 years we'll be able to get cheap used Naples processors and dx12 and maybe dx13 by then will mean the moar cores are actually a huge improvement

File: IMG_3372.jpg (165KB, 633x758px) Image search: [Google]

165KB, 633x758px

>>60108744
And still have them destroyed by a 7700k in any game

>>60108750
In older games, yes (from the perspective of several years in the future when used Naples chips are being discarded by businesses). But by then, games will utilize moar cores better because dx12 makes it easier to do so. I expect dx13 to refine that process too, but it's too early to make that kind of prediction with certainty.

>>60108365
>Intel designs
>far more aggresive
>throwing another bingbus and fatter datapaths
Nah. Its a nice way to create housefire and that leaves only one, very small market to Intel.

File: IMG_3366.jpg (995KB, 2159x1216px) Image search: [Google]

995KB, 2159x1216px

>>60108494
>Ryzen at lower clock speeds (~3-3.3GHz) has almost twice the performance per watt of any Intel processor.
>Ryzen performs around the same in muh gaymen.

Is you retarded? Intel 4 core rips though amd's 8 core in any game

HE'S BACK.

>>60108673
>page level

shit. cache line level

File: IMG_2802.png (538KB, 1279x720px) Image search: [Google]

538KB, 1279x720px

>>60108793
Is Far Cry Primal an old game?

>>60108750
>gaymes
do you take it up the ass too?

DONT REPLY TO HIM.

>>60108874
I'm reporting you too, this isn't your safe space

>>60108846
Far cry has mammoth's shit engine

>>60108815
>Intlel shills in charge of reading comprehension

>>60108673
Is the power cost of DRAM to GPU transfers really that significant in real-world workloads, though? One would think that with the majority of the embarrassingly-parallel workloads that people use these days (specifically neural network memery), you can easily do all your operations within GPU memory, then transfer out the network weights at the end.

>>60108980
What about frame timings and stutter?

>>60108997
Intel uses 91w on the 7700k while 1800x uses 95w and the 7700k still beats it by 50% in games.

>>60108800
>>throwing another bingbus and fatter datapaths

You do realize that this is the hard part right? Figuring out how to keep a deep and wide OoO pipeline full of useful work is super hard.

As impressive as the AMD's Zen core is, it is still a simpler, narrower core with less cache bandwidth than Intel's designs. And you see it in the perf/clock. It is a beefier bulldozer - meaning they had enough transistors to make the core be a full core, but it seems to be in the same design thought of more cores and use the GPU for as much as you can.

I have yet to see a serious suggestion anywhere that this design philosophy is going to end up with a net higher performance than Intel in any application.

File: IMG_3219.png (55KB, 802x601px) Image search: [Google]

55KB, 802x601px

>>60109025

>>60109040
You won't. Amd fucks up everytime and then then hire an army of poojets to report you when you destroy their bullshit with actual benchmarks.

>>60109040
90% of the market will never touch housefire-512. Another 8% will use GPU's. Congrats Intel on leaving the smallest part of datacenter market in existence for themselves.

>>60108750
>>60108815
>>60108846
>>60108980

>a thread about a server cpu, a use case where most tasks are heavily multithreaded
>starts talking about single-thread dependent vidya gaymen

The autism in this anon is astounding. No, just because Intel was mentioned doesn't mean you need to start screeching about the 7700k.

>>60104387
wait 3 months after launch then because day 1 reviews always suck.

Mmmm concern shilling sweet I like it.

>>60109069
Triggered I invaded your safe space street shitter?

File: IMG_3347.jpg (159KB, 1248x683px) Image search: [Google]

159KB, 1248x683px

>>60109073
Ryzen still sucks almost 3 months later too

File: excuseme.jpg (155KB, 700x553px) Image search: [Google]

155KB, 700x553px

>>60108815
"Intel quad-core beats Ryzen in any game"
Have fun with those stutters. Also, shitty TIM

>>60109063
>hire an army of poojets
They do it for free. AMD fanbois are some of the most obsessed and mentally unstable retards you can see on the net.

>>60109099
>stutters
Except it's amd that stutters. See>>60109046

>>60109098
The eyes can't see past 200fps

File: 2wukw1t.png (82KB, 694x801px) Image search: [Google]

82KB, 694x801px

>>60109082
>Skylake v Naples thread
>BUT WHAT ABOUT DUH 7700K
>BUT MUH GAYMEN

That's fairly autistic if you ask me.

>Intel shill
>cherry picks all those games that have shitty game engine (Farcry Primal, RotTR, DOTA2, Watchdogs 2 etc)

All using Nvidia GPU's with their shitty DX12 implementation.

File: 1487284046126.jpg (45KB, 1500x1500px) Image search: [Google]

45KB, 1500x1500px

>>60109033
>TDP is power consumption
>all cores on Ryzen are loaded

File: IMG_2802.jpg (169KB, 1248x720px) Image search: [Google]

169KB, 1248x720px

>>60109141
>every game that exists
>"LE CHERRYPICKED BENCHMARK"

Amdrones everyone

>>60108846
What part of (from the perspective of several years in the future) did you not understand?

>>60109007
>you can easily do all your operations within GPU memory

And that is exactly what Nvidia recommends, because once you have to start sucking data back and forth across the pcie bus you loose.

>>60109182
>1500x
>1600x
>1800x

all same performance, well i guess that game is limited to 2 cores if it finds AMD LUL

>>60109197
>JUST WAIT

The future is right now you dumb indian. By 2020 when games will use MOAR COARZ both Intel and amd will have already released newer CPUs that shit on what you have right now.

>intel server CPUs are so bad that it's literally not worth using them over naples even with a bribe

brehs

>>60108815
I hope you die in a fire together with your parents for raising such a retarded shit.

>>60103598

Ryzen reaches the theoretical bandwidth of DDR4 much closer than Intel chips. The latency is something funny going on that Im not sure they pinned down yet. Look at anandtechs forum, the Ryzen strictly technical thread. They got some solid guys there performing tests and trying to discover what the arch can and cant do.

>>60109182
>>60109098
>>60109046
>>60108846
>>60108815
>>60108750
Literally the exact same benchmark spamming retard from yesterday, in the exact same thread, shitting it up for the second fucking time. Mods banned him in another thread but he just keeps coming back. Fuck this, this board is a fucking forsaken shithole.

File: mpv-shot0032.jpg (51KB, 1280x720px) Image search: [Google]

51KB, 1280x720px

Workstation/Rendering folks rejoice, Adobe patch to Premiere Pro nets significant performance for Ryzen, mainly due to the fact the cores were sitting at under 50% before the update.

File: mpv-shot0034.jpg (136KB, 1280x720px) Image search: [Google]

136KB, 1280x720px

>>60112400

https://youtu.be/k9SDgZr-Xt4


The culprit(which is shtty software support and FUCKING NOBODY coding for NUMA)

File: IMG_3379.jpg (79KB, 568x612px) Image search: [Google]

79KB, 568x612px

>>60112170

Oh look, another wojak edit. How unique and original.

Oh, he also samefags.

>>60112170
>>60112726
>>60112816
Mods please ban this shill. Every time I try discussing something he comes up and shits the thread with his cringy memes. I tried browing at day, at dawn, at night and he's always here. He has no life.

Please mods do something.

>>60112726
He got banhammered yesterday, just report and hope for the best.
Remember, don't quote, don't respond, let him rot.

>>60094831
Pls do not post gaymer performance chars about a different set of processors in a thread about server processors shill

File: preload.jpg (100KB, 1424x485px) Image search: [Google]

100KB, 1424x485px

>>60108337
I'm that anon that had to disable SMT to get it to run. I'm sure ESXI will support it in time, just keep an eye out for 6.5u1. Here's a prod cluster that got stood up without anything other than the essentials on it at the moment.

I actually talked to one of my vendors yesterday and they told me that they are not planning to compete with Intel at the very high end and very low end server offerings, but instead targeting the middle market, think dual socket 2650-2680v4's, the sweet spot.

>>60115622
>the very high end and very low end server offerings,

This I find suspicious since they have a very power efficient arch.
Could he have meant the 4P motherboards? That doesn't really take up more than 2-4% of the market.

File: 1492471534882.jpg (514KB, 960x686px) Image search: [Google]

514KB, 960x686px

I can't wait for pooples to fucking burn and fail just as hard and FINALLY force AMDone into bankruptcy

Then we'll finally have Intel lower the price on their chips without AMD overinflating the market with shit

>>60116894
Good bait, monsieur.

>>60116894
Honestly can't tell if trolling or just retarded benchmark spammer back yet again.

The thread is over but be sure to report shitposters for improved board quality.

>>60116974
That faggot was removed for like half a day and yet posting quality improved greatly.

>>60116985
And then he came right back. Like herpes.

>>60117020
We can only report.

File: 1352653283596.png (8KB, 300x300px) Image search: [Google]

8KB, 300x300px

>>60117020
>>60116985
>>60117045
Here's a few words of wisdom this site in its better days left us.

>>60117071
Most people in AMD-related threads are VERY thirsty for bait. And then the thread gets derailed.

>>60094154
Amd showed their hand Intel won't sit still things will be interesting but I wager intel will maintain a lead.

>>60117071
NO I GOTTA TELL HIM HOW WRONG HE IS

>>60117105
>intel will maintain a lead.
How? Their arch is old, inefficient and scales like shit.

>>60117120
Our fabs can fix anything!

t. Intel still thinking it's 2010

>>60117147
Twice is hilarious considering they are behind to both Samsung and GloFo into 7nm.

>>60117159
You forgot TSMC

Now what would be really hilarious if UMC got there as well.

>>60117105
>I wager intel will maintain a lead
>Soure: My ass.

>>60117192
Ah, my bad. Anyway, we're approaching the limits of silicone shrinking. Let the best architecture win!

>>60112400
>mainly due to the fact the cores were sitting at under 50% before the update.
Why did this even happen

>>60117269
Because Adobe writes shit code, and I'm surprised they even updated it so fast

>>60117284
HEDT launch soon. Or something.

>>60094923
>Social Justice Dogs 2: WE WUZ HAKURZ N SHEEEEEEITTTTTTT

>>60094956
>Negroid, cuck, and edgy: the game
>console port

>>60095269
>Team Autism 3

>>60108815
>slavs

>>60108846
>Ape sounds: the game
and it doesnt matter how old a game is if the engine is fucking ancient and in far cry's case it is, still uses a large portion of the original cryengine

>>60109046
>too dumb to play a rts: the game

>>60109182
>posting ape noises: the game for the fucking 7th time

nothing of value lost, all these """"games"""" are shit

>>60116745
He was referring to two socket bread and butter systems. He wasn't specific about if they had any other roadmaps to show.

Literally I called up my vendor and asked what AMD had come by and said. He told me that they came by showed them the slide decks and all that other good shit and that they are targeting that market space. Nothing else on channel availability other than "q3/q4". No performance characteristics or engineering samples. This is the part that got me because the guy I work with works for a fairly large group on the east coast. Usually if you have a product that's gonna wreck someone elses bread and butter, you want to show it off as much as possible.

Also I like never come on this board. Is there any chan style board where who actually work with this shit can talk? All these kiddy HEDT gaymers make my head hurt. Nobody gives a shit about a few more FPS and no games these days make use of top end shit, so nobody cares.

>>60094820
And what is 8 times 4?

>>60117482
Kek, best post in thread.

>>60108750
Here's your (You) faggot