Naples AMD

>>59282479
How can jewtel ever recover

File: mpv-shot0010.jpg (90KB, 1280x720px) Image search: [Google]

90KB, 1280x720px

>>59282479
Benchmark against top of the line

Intel® Xeon® Processor E5-2699A v4
2255M Cache, 2.40 GHz, 22 cores, 44 threads.

https://ocpussummit2017.sched.com/event/9TZ3/amd-amd-naples-the-future-is-open
https://ocpussummit2017.sched.com/event/9TZl/amd-amd-naples-open-innovation-for-the-software-defined-era

File: Forrest_Naples_Embargoed-Until-3_7_17-17-1440x810.jpg (65KB, 1440x810px) Image search: [Google]

65KB, 1440x810px

Amazing

>>59282487
>55M Cache, 2.40 GHz, 22 cores, 44 threads.

Fixed*

File: Forrest_Naples_Embargoed-Until-3_7_17-18-1440x810.jpg (65KB, 1440x810px) Image search: [Google]

65KB, 1440x810px

>>59282508

http://ark.intel.com/products/93790/Intel-Xeon-Processor-E7-8890-v4-60M-Cache-2_20-GHz

AYYMD as usual did not bench against the best

Rypoo Naploo is a fraud as usual

These arent games

File: Forrest_Naples_Embargoed-Until-3_7_17-19-1440x810.jpg (63KB, 1440x810px) Image search: [Google]

63KB, 1440x810px

>>59282526

>>59282535
come now we don't have to fight over this one, we'd have to just check q3 AMD finances to see who is more clever you or all the people responsible for datacenter hardware

This made me want to visit Naples. Real world things

>>59282554
Lol

But yeah it seems AMD has support for more memory and channels per socket while having more I/O AND more cores, if their TDP stays within that range then it's a win on all fronts.

File: 1488892667673.jpg (179KB, 1280x720px) Image search: [Google]

179KB, 1280x720px

>>59282479
>dem misaligned DIMM slots
Triggered

>>59282570
depends on clocks, zen can keep it 6w per core at 2.8 I think or somewhere near those numbers

>>59282602
someone didn't use a grid while modeling it

File: Screenshot from 05-03-2017 23:38:48.png (62KB, 957x532px) Image search: [Google]

62KB, 957x532px

>>59282613
6.25W on 3.3GHz per core running AVX without cherrypicking dies like they would for naples.

35W TDP on 8 cores with 1.9GHz

It's looking pretty damn good for AMD.

File: 018.gif (2MB, 580x433px) Image search: [Google]

2MB, 580x433px

No game benchmark, into trash it goes

File: 1488576947538.png (64KB, 960x540px) Image search: [Google]

64KB, 960x540px

>>59282632
>>59282632
Performance scaling at TDP

>>59282641
>7700k still has more fps

AMPOO BTFO

File: 14767188-Rejection-Stock-Photo-disgusted-woman-face.jpg (171KB, 1300x997px) Image search: [Google]

171KB, 1300x997px

>being so much of a consumerist cuck that you have to root for "your side" and bash their competition in the enterprise market even though you'll never own enterprise hardware

File: 1487656747556.jpg (16KB, 500x322px) Image search: [Google]

16KB, 500x322px

>>59282479
Yeah but how fast is it in GTA V @ 1080p?

games

>>59282670
>>>/v/

File: Screenshot from 04-03-2017 03:47:54.png (54KB, 1489x889px) Image search: [Google]

54KB, 1489x889px

>>59282632
Some voltage scaling.

Thank fuck someone actually bothered going into low-level testing instead of just fucking benchmarks.

>>59282711
>>59282653
How do I read these graphs, WTF does fmax and vmin mean

>>59282668
>Intel still has microsttuter

INTEL BTFO

seriously will we ever agree that ryzen and kaby two absolutely different CPU for different needs, and you can choose either for benefits one of them provides?

File: 1405229211592.gif (2MB, 308x310px) Image search: [Google]

2MB, 308x310px

>>59282641
laughed way to hard at this

>>59282479
I hope the 16 core Opterons don't cost more than $800, doesn't even have to be the highest clocked one, just needs to have a lot of memory and I/O

And hopefully 1P motherboards with IPMI don't cost more than a kidney.
Oh who am I fucking kidding I'm probably looking at $600 for mobos alone.

https://ark.intel.com/products/96899/Intel-Xeon-Processor-E5-2699A-v4-55M-Cache-2_40-GHz

>Max Memory Size (dependent on memory type) 1.54 TB

>Intel's comparable offering? Twenty-two cores and 44 threads, eight memory channels, and a maximum of only 384 GB RAM.

AYYMD intentionally crippled Intel again, what a scummy company

>>59282632
Those 2 core complexes instead of 1 monolithic one does WONDERS for low power pstates, effectively makes everything easier to gate and power consumption goes down.

File: 1488258781492.png (269KB, 376x449px) Image search: [Google]

269KB, 376x449px

Ars already called out AMD BS benchmarks:

Since Naples has twice the memory channels and therefore twice the bandwidth, of course Naples will out perform the Intel setup with a MEMORY CONSTRAINED workload.

Since Intel has literally twice as good AVX 256+ performance, a pure computationally constrained workload similar to their example should heavily favor Intel.

File: 1488454105571.png (284KB, 650x581px) Image search: [Google]

284KB, 650x581px

>>59282508
>intel
>only 22x2 cores in 2017

>>59282785
Both are valid, Intel doesn't have physically the memory and AMD doesn't have the AVX512 throughput, they're comparing what's there.

File: 1488420066835.jpg (42KB, 898x886px) Image search: [Google]

42KB, 898x886px

>>59282479
Lets see that stock drop now bois

>>59282641

>>59282763
They do offer quad channel memory per die... or 8 channel per socket.

So right there that's something more expensive than current Ryzen chips.

Plus they have WAY more I/O. Like 32 PCIe lanes per die, 64 lanes per socket... compared to only like 20 or 24 on Ryzen.

>>59282763
16snowy owl chips are rumoured to be embedded parts to compete with xeond. Think low power, clocks and expensive motherboards because of 10gbe.

>>59282798
I think they released a 24 one IIRC.

>>59282748
>seriously will we ever agree that ryzen and kaby two absolutely different CPU for different needs, and you can choose either for benefits one of them provides?

Did you forget what site you're on?

>>59282812
There are no quad channel dies.
There is only the singular Zeppelin die.
Naples is a 4 die MCM. It has up to 128 PCI-E lanes per socket.

>>59282820
BGA? I can deal with that, might make the setup cheaper if it's prebuilt.
Most of these large boards are.

>>59282479
>Naples
apparently Ryzen wasn't a bad enough name I guess

https://twitter.com/amdserver?lang=en

>>59282632
>>59282653
holy fuck that's some efficient shit right there
once the APUs come out I'm building myself a really efficient ryzen HTPC

>>59282727
fmax = max frequency
vmin = min voltage

for AMD's sake I hope they have their firmware and software support sorted when they launch this fucking thing this time. Data Centers have a lot less patience for broken buggy pieces of shit than your average amdrone.

File: 1488207915702.jpg (125KB, 960x639px) Image search: [Google]

125KB, 960x639px

>>59282554
>insufficient memory to load

>>59282727
voltage and frequency.

What are you doing in this thread, on this board.

>>59282772
you idiot the process crashed because the cpu couldnt feed fast enough the program it resulted on a halt->crash

there is a reason why those kinds of workloads are passing through the gpu on intel systems

>>59282883
Why do you think consumer 8 cores were released first(besides binning reasons), all these 32 core chips are made out of 8 core dies, fixing firmware and software on them fixes them for the Naples ones too ignoring edge cases.

Everyone currently owning a R7 Ryzen is doing validation testing for AMD's server platform.

File: 1486270615924.jpg (45KB, 352x351px) Image search: [Google]

45KB, 352x351px

>>59282840
Ah, I was thinking they were gluing two Ryzen7 dies together on a substrate.

You know what I meant though.

They're offering twice as much memory channels and over twice as much I/O per every MCM pair.

File: 1488664232472.png (84KB, 653x726px) Image search: [Google]

84KB, 653x726px

>>59282554
NOOOOOOOOOOOOOOO DELET THIS

BUT CAN IT REACH 300 FPS CAP IN DOTA?!

>>59282898
You're the idiot, it's about memory capacity, not speed

Intel's chip would certainly have sufficient memory to load that benchmark but AMD intentially equip it with 384GB only

Learn to READ, moron

File: 1487883723328.gif (1MB, 235x232px) Image search: [Google]

1MB, 235x232px

>>59282554

>>59282632
The 32core won't be 3.3Ghz. That'd be like 180-200w TDP.

My guess is something like 1.9Ghz base, 2.6Ghz turbo for the 32core.
And maybe 3.3Ghz turbo for 16 core.

>>59282929
AMD didn't equip anything, these are prebuilt OEM 2p racks and no small number of them top out at 384GB of memory for the current Intel.

>>59282943
The top Naples SKU has a 180w TDP.
They have a significant clock advantage over Broadwell Xeons.

>>59282727
Max frequency minimum voltage

Cmon dude

>>59282554
Man that must be a massive amount of memory usage.

File: 622bd01406f7538af882cf5f38969691be0186eb9ae95f7978d506ce28e8560a.jpg (36KB, 657x527px) Image search: [Google]

36KB, 657x527px

what is the gaming performance like

File: 1473711192024.jpg (80KB, 766x960px) Image search: [Google]

80KB, 766x960px

>>59282913
Shit, actually I'm wrong.

AMD has 8 MCM per socket... 32c/64t

Still won't cost less than $1000 bucks...

That's literally like 4 Ryzen 7 CPUs on a single socket.

>>59282973
this
run it at 480p pls

>>59282883
ryzen is beta test for servers
drop in the ocean compared to money they will make on data centers

>>59282943
>>59282958
lol oops. I retract what I said.
Yeah the benchmarks line up with it possibly being 3.3ghz all-core? On 32 cores?

Wow, that's crazy. Way way way beyond my expectations and any other expectations I've seen.

>>59282692
>criticize blatant consumerism and shilling
>HURRR U R /V/

>>59282479
It runs games 5fps slower than a 7700k, so it's trash.

File: 1488750861088.jpg (619KB, 3999x2250px) Image search: [Google]

619KB, 3999x2250px

Jesus

File: 1486329456451.png (60KB, 400x400px) Image search: [Google]

60KB, 400x400px

>>59282823
>intel
>only 24 cores in 2017

File: AMD-Radeon-Instinct-In-A-400-TFLOP-Server.jpg (271KB, 1600x900px) Image search: [Google]

271KB, 1600x900px

>>59283030

>>59283012
I'll likely have a base clock somewhere in the range of 2.6ghz inside that 180w envelope, unless their hand picked enterprise dies are radically better than what we've seen in consumer Ryzen. Its possible for them to have better clocking parts, but I don't think we'll see that much of a leap. A 2.7ghz base clock would be tremendous.

That seismic analysis must use some obscene amount of memory, anyone knows exactly how much?

>>59283080
>Finally, AMD offered a workload that quadrupled the size of the data set. The Naples machine completed the task in 54 seconds, while the Intel system crashed with inadequate memory. This proved convincingly that, yes, 2TB of RAM is more than 1.5TB of RAM.

1.5TB

>>59283065
I don't know man, the 1700 has a configurable TDP at 35W and it runs 8 cores at 1.9GHz.
Server dies are always gonna be more cherrypicked on a tweaked process for higher efficiency, so I wouldn't be surprised to see server parts at that frequency at 2.1GHz.

8 cores at 2.1GHz, 35W

I don't really know what to say when you scale that up.

>>59283080
see
>>59282487

>>59283102
If you follow TheStilt's Cinebench scaling then at 45w they're hitting 2.5 or 2.6ghz.
Four 45w Zeppelin dies would come out to 180w.

>>59283139
>>59283102
2.0 base, 3.3 turbo, with a max of 2.5-2.6 all-core turbo sounds very likely to me.

File: . (81).jpg (22KB, 480x360px) Image search: [Google]

22KB, 480x360px

>>59282554

LMAO

File: 1487657738786.jpg (16KB, 480x360px) Image search: [Google]

16KB, 480x360px

Will Linus update his shill boxes?

>>59283087
Jesus Christ

>>59283139
see
>>59282653
1000 score at 35W, 1150 at 45W

15% increase for 29% more power, efficiency loss.
Zeppelin is most efficient around 2GHz.

>>59283175
Only if they give him one, this shit must be expensive as hell

Intel is over. No gamer faggots, compilers, Microsoft feet-dragging will shit up this launch.

>>59283172
>>59282943
What is turbo, exactly? My only experience with turbo clocks is turning my unlocked chip's multiplier up to the turbo clocks or higher. Windows seems perfectly able to downclock everything during idle, even when the multiplier is overclocked, so I'm not clear on my manufacturers even make the distinction between "base clock" and "boost clock" when boost clock seems to kick in essentially whenever CPU usage is high, which is no different than when you don't have this turbo feature... right?

Thread theme
https://www.youtube.com/watch?v=cLoytewvn0g

>>59283058
>4U rack
>almost 300 lanes with full x16 lane support for 8 GPUs and every SAS/SATA/NVMe/NIC setup under the sky
>6600 watts Platinum with 3 PSU redundancy


MY
FUCKING
DICK

Also only $70000 starting

>>59283234
Turbo overclock certain cores and downclocks others while still respecting the combined TDP of the chip. It mainly improves performance in single threaded tasks on multicore chips. The TDP does not account for all cores running at the turbo frequency and the voltage might not be sufficient to run all cores at turbo frequency at least not in a stable manner.

>>59283234
Max Turbo is traditionally used for one core, there's an all turbo that boosts all cores by a few hundred MHz at best if it doesn't go over the TDP in certain conditions(non-SIMD load)

This all-turbo is very rarely marketed outside of datacenter spreadsheets.

>>59283234
Turbo is a temporary overclocked done by the chip itself when more single core performance is needed.

The turbo frequency is almost never what all cores can operate at, and the "all-core turbo" is always somewhere in between.

>>59282479
>128 PCIe Lane

PLX switch vendors on suicide watch.

>>59283058

couldn't they have atleast made a cooler design?

>>59283282
>>59283283
>>59283287
So if I have some random locked i5 CPU with a base clock of 3 GHz and a turbo of 4 GHz, and CPU-Z reports 4 GHz while I'm running Prime95, what is actually happening? Is just one core OC'd to 4? All 4?

>>59282479
It's a complete royal flush.

More memory, more bandwidth, more cores, lower power package power, lower platform power.

Is this the return of k8?
Zen seems to be called K17 internally, it's literally K8x2 and then some.

>>59283058

This is so sick

>>59283316
One.

>>59283351
Is there a way to view the clock speed of each core individually? And why does it stay at 4 GHz throughout the duration? I assume that depends on cooling and temps, correct?

File: 1c347159-fbda-499b-925e-3b99a2969e9f.png (119KB, 1269x628px) Image search: [Google]

119KB, 1269x628px

>>59283316

File: ahegao.jpg (53KB, 960x564px) Image search: [Google]

53KB, 960x564px

>>59282479
16 channel memory? Not quad channel? Not octochannel?

What the fuck?

>>59283331
its K12, K8 and barely half more, sounds about right

>>59283383
hwinfo or openhardwaremonitor IIRC, all-core-turbo is SKU dependent.
see
>>59283387

For consumer i5-i7 I haven't found something similar.

I hope they have a 16 core 3.6ghz for workstations. If there's a 1P board it will be ideal.

File: IMG_0546.jpg (68KB, 412x627px) Image search: [Google]

68KB, 412x627px

>>59282479
Fucking hurry up and do it, EC2 prices need to come down

>>59283420
>>59283383
To be clear, I can simply slap a big cooler on an unlocked chip and as long as it provides enough cooling I'll be able to run all four cores at whatever turbo is capable of, right?

File: mpv-shot0011.jpg (170KB, 1280x720px) Image search: [Google]

170KB, 1280x720px

>>59283415
http://valid.x86.fr/cache/screenshot/vhfv93.png

It's k17, K12 is the ARM one.

>>59283415
K12 is AMD's internal name for their ARM core version of Zen oddly enough while K15 was Bulldozer. Zen itself is probably K16 but I'm not sure.

>Zeppelin die has 32 PCI-E lanes
>some are flexible, but 16 are always reserved for GPU
>Naples is 4 Zeppelin dies
>1 socket has so much PCI-E bandwidth that it could serve 4 GPUs that all max Titan X Pascal look like a mid range card, and they wouldn't even come close to eating up IO bandwidth
>8 channel memory with 2400mhz ECC, 153.6GB/s system memory bandwidth per socket
>2 DIMMs per channel
>64GB DIMMs
>1TB per socket if not more supported
>more AES and SHA hashing performance than anything else available
>higher SSL throughput than anything available
>higher perf/watt than Xeons
>likely to still cost less too

The only question is how much market share is AMD going to gain this year

>>59283461
No, it's TDP driven, not temperature driven.
While ICs at lower temperatures do use slightly less power, it's not a noticably difference.

>>59283471
question is how many yachts am I buying with this stock about the skyrocket

>>59283058
This kills the PC

>>59283407
8 channel, but 1 DIMM per core, 2 per channel.

>>59283471
The only question is if GloFo will manage to meet wafer demand for these things and will Samsung's backup fab come into play.

>>59283508
>1 DIMM for every 2 cores.

>>59283483
>No, it's TDP driven, not temperature driven.
so the reason we can do this with unlocked chips boils down to the ability to control voltage?

>>59283494
I bought a boat a while ago. Its not worth it. Fuel is a constant cost, if you live somewhere cold you have to worry about storage and winterizing, every so many engine hours you need a rebuild, you deal with fixing gel coat, minor cosmetic damage from docks. Its a hassle. The water is fucking over rated anyway. If I wanted the surface under my feet to shake and bounce I'd buy a house near a fault line.

>>59282929

Thats the maximum memory that the Intel rack can support. Dont you see the idea? AMD supports more memory per die, hence more per rack.

>>59282479
If only AMD's memory controller supported higher clocked ECC LPRDIMMs DIMMs this would be total obliteration, in that sense they're currently equal to Intel

>>59282632
HOLY FUCKING SHIT THAT EFFICIENCY.

Can someone calculate how a mobile APU would do? Do remember that those "15W" mobile Intel ULVs can pull over 30W at times.

>>59283331
I see Zen as K7. Zen+ will probably be the K8 in this sequence. I wonder if they'll name it K18 though.

>>59283640
2 cores at over 3.3base and 4.0 turbo with a Vega GPU.
4 cores at 3.0 base and 3.7/3.8 single core turbo.

Really depends how much that GPU goes into equation as said TDP especially for mobile parts has literally nothing to do with power consumption, while on servers it's pretty close to the power usage.

>>59283595

You got a source on that claim? Naples had 2400 ram in those demos.

>>59283640
Aside from uncore power draw you could just cut those power figures in half to see what a quad core CPU would pull at a given power level.
Only caveat is that APUs throttle CPU clocks to give more power to the IGP, so something with a 2ghz base clock may not always run at 2ghz depending on the workload. It depends on what power budget AMD would give to each component.

>>59282479
>tfw fell for the 1024gb ram meme

>>59283595
Octa-channel memory though... the throughput is insanely high. For any equivalent memory speed AMD will have a 33% advantage compared to Intel's six channels.

>>59282554
When will RAMlets ever learn

File: 1487888155171.png (162KB, 633x900px) Image search: [Google]

162KB, 633x900px

A-A-AYYMD PAJEET LIES, DELETE THIS THREAD ALREADY

File: Screenshot from 07-03-2017 15:46:50.png (174KB, 1778x968px) Image search: [Google]

174KB, 1778x968px

>>59283737
2400ECC as well.

But, not all Broadwells support it, so that's another win for AMD since the Zeppelin die supports 2400 across the board.

File: Superman Kek.webm (655KB, 640x266px) Image search: [Google]

655KB, 640x266px

>>59282554
>insufficient memory to load

>>59283797
DELID THIS

>>59282554
Fucking RAMlets

>>59283570

The only two great moments you have when you own a boat is when you buy it and when you sell it.

File: 1487073588681.jpg (219KB, 667x1000px) Image search: [Google]

219KB, 667x1000px

BUT AMD GETS BEAT BY INTEL 7700 IN GAMES SO NONE OF THIS MATTERS
DELETE THIS RIGHT NOW

>>59283807

Thanks, also the good folks at Anandtech found that Zen comes closer to the theoretical bandwith of RAM than Intel. Zen gets more througput from the same ram theoretically.

2016: MOAR CORES
2017: MOAR DIMMS
Wtf amd

>>59283595
There's no point in making support for higher clocked 64/128GB DIMMs when kingston, samsung, hynix, micron etcetc don't bother making higher clocked RDIMMs, no doubt due to power issues.

>>59283896
2017 onwards : MORE EFFICIENCY
WTF I LOVE AMD MORE

>>59283896
These are both good things

anyone got that SSL benchmark?
will datacenters eat that up?

>>59283058
Now this is something to drool over.

I could probably serve a Xen/ESXi environment to half my block with this.
I'll just need a bunch of fast switches and NICs.. which fuck me cost more than a car.

>>59283896
>2017: MOAR DIMMS
More like pcie wet dream.

>>59283936
VPN/VPS and certificate servers certainly will.

>>59283918
Unless you actually like to play video games pajeet

>>59283896
Naples has more of everything.

>>59283889
Speaking of anandtech, have those folks finished the proposed gaming benchmark yet?

>>59282806
Silly boy, avx512 is only on phi co-processor

File: ASN_IBM_SOI_FinFET.gif (118KB, 960x478px) Image search: [Google]

118KB, 960x478px

>AMD has even higher IPC parts with radically higher clocks coming down the line made on IBM's 7nm SOI FinFET process

Wew

>>59283989
what was proposed?

>>59282479
how much FPS in Crysis will I get with this?

>>59283781
It appears to be 8 dual channel controllers.

>>59284022
at 640x480? probably 15 with drops to 11

Now the question is.. what are the clocks for the 32 part..

Intel's 22 ones top out at 2.2GHz IIRC, there might be newer binned models.

>>59284018
anandtech's review of ryzen was only cpu shit.
You know benchmarks that favor ryzen.

They said they would do a gaming benchmark review later
Or you know, stuff that ryzen doesn't excel in.

>showcase the good
>wait until people forget about it to show off the bad

>>59284036
they made them work in x8 bandwidth?

Has anybody noticed that the Intel shills have literally just vaporized? Any left at all? So sad!

>>59284022
In 300x200? Not much.

>>59284047
Intel's top core count ones are 2.2 base, 3.4 turbo.

I'm guessing 2.0 base and 3.3 turbo with around a 2.4-2.7 all-core turbo for the Ryzen parts.

File: 97360685.jpg (49KB, 455x427px) Image search: [Google]

49KB, 455x427px

>>59282479
>AMD OWNS SERVERS
>AMD GETS LOTSA MONEY TO PUT INTO THEIR TECH GAME
>WE GET AMAZING COMPETITIVE PRODUCTS AS A RESULT
fuck yeah.

>>59283781
>Intel's six channels.
Only on Skylake-EP, which comes a quarter or two latter than Naples.

Broadwell-EP/EX is only quad channel.

>>59282479
How much for a 2P SuperMicro board, a couple lower cost Naples CPUs and about 128gb DDR4 ECC ram?

File: tyrone laughing.gif (3MB, 264x242px) Image search: [Google]

3MB, 264x242px

>>59282554

>>59282602
>>59282623

In the video it moves down, OP captured before it was in place.

>>59284018
http://www.anandtech.com/show/11170/the-amd-zen-and-ryzen-7-review-a-deep-dive-on-1800x-1700x-and-1700/23
Shit, the gaming benchmark is clearly not "proposed" but "upcoming".

>>59283896
MOAR EFFICIENCY
MOAR PERFORMANCE
MOAR MARKETSHARE
MOAR OY VEY ANUDDA SHOAH

>>59282479
Still worse than i7 7700 in games. Ill pass.

Can't wait for the 640 x 1136 benchmarks using sodimms from Iranian imams that show intel absolutely crushing amd

>>59283058

Inb4 this is the new meme home computer that everyone wants. If I had the cash to blow I'd get this, install Windows XP and play Runescape on it. Why can't you get a mortgage for a meme? I mean I can afford the deposit and repayments so why not :(

>>59284095
good, first not rushed review

>soc

Motherboard manufacturers on suicide watch.

About time one of these faggots put these literal cheap ass "gaymen" companies out of business.

>>59284087
I don't think you're looking at this under $4000

But can it run Crysis?

>>59284138
>say you're starting a hosting company
>and you also do cloud processing
>sell instances and processing time to local weather station, university labs, memecoin miners, etc
>pay back loan
>earn passive income
>play runscape

File: ________________.jpg (23KB, 355x354px) Image search: [Google]

23KB, 355x354px

>>59284060

File: 1469650137081.jpg (12KB, 426x382px) Image search: [Google]

12KB, 426x382px

>>59282865

My Naples are Ryzen from all this delicious AMD news.

128 threads? It would be ideal for 4k H.265 encoding servers.

>>59284191
Doesn't Intel do a bit better on H.265 and AMD on H.264?

>>59284191
Unlike x264.. x265 isn't very well threaded yet, so no, not really.

>>59284060

They cant shill vs this. Naples is better on every level except avx, but most of the server market doesnt even use it.

>>59283058

Up to 144GB network bandwidth?
144GB?
G B?

>>59284230
HPC does use it, around 50% or higher.
Good thing HPC is under 20% of the datacenter/server market.

>>59284239
Good fucking eye, I thought these things platteaud currently at 50GbE

>>59284246
Why in the fuck wouldn't they just use GPUs instead of relying on shitty GPU extensions?

>>59284214
It's not that slow on x265 thanks to fast int and 4 SSE units.

http://www.hardware.fr/articles/956-13/encodage-video-x264-x265.html

>>59284270
If it was cost effective or doable they would do it.
SIMD isn't very good at branching.

>>59284214
Well they probably did against Bulldozer. I'm not sure it's still the same

>>59284060
Na. they're still making those 7700K beats ryzen in gaming threads. There has been like 4 new threads since i woke up today.

>>59282479
lol

>>59284186

Have a (You), my man.

>>59284270
Ask Nvidia, I'm sure they'd tell you.

>>59282487
>>59282508
>>59282526
>>59282554
>>59282602
>>59282632
Oh are we building hype from leaked benchmarks again?
I'm sure this went really well the last time.

File: light.dotup.org422459.jpg (62KB, 600x661px) Image search: [Google]

62KB, 600x661px

>>59284280

>>59284309
Ask them what?
They'd love to sell you a rack full of their compute GPUs.

>>59284320
Looks pretty fine to me, and you can see it's not all that well threaded because of the 6950X scores so close to the 6900k

>>59284320
Hm so slightly lower IPC for x265, but likely better perf/watt.

But AMD stock isn't skyrocketing on this news?

Naples is the ideal choice for h264 encoding, h264, DNS, VPN, cloud compiling, and a very very very long list of common servers. What gives?

>>59284320
Now how about a non meme encoding like VP9

>>59284333
You can split the video and run multiple x265 processes, then.

>>59282681
Asking the real questions

>>59284349
I don't know anyone is encoding in VP9 other than Jewgle.

>>59284148
That much? I'd figure about $400 for the board, $700 * 2 for the CPUs and about $1200 for 128gb DDR4 ECC ram.

~$3000

>>59284319
One of those is from a chip that was already released. We already know Ryzen is a monster in server workloads, it's one of its strongest areas

>>59284387
>other than youtube
>the largest video hosting site on the internet

Right.

>>59284320

>not far off matching performance
>95w versus 140w

Jesus fucking Christ. This whole Zen release has blown Intel the fuck out of all their Lakes. I thought efficiency was Intel's thing? How the hell have AMD done this?

>>59282681
more like
>@800x600

>>59284392
Just an estimate, I don't think you'll be using 1GbE NICs or switches for these things.

File: [HorribleSubs] Hibike! Euphonium S2 - 01 [720p].mkv_snapshot_32.03_[2016.10.06_06.53.21].jpg (164KB, 1280x720px) Image search: [Google]

164KB, 1280x720px

>>59284320
>images/s
x265 is currently not worth it anyway.
use x264 10bit.

>>59284402
>How the hell have AMD done this?
The 2x CCX design contributes a lot to this, it's easier to keep thermally and easier to gate off.
Plus, the integrated southbridge really helps.

Also nobody mentioned (but Wendell of course) this but AMD does have a integrated voltage controller like Intel did with Haswell and Broadwell, removed Skylake and later, but it's currently disabled and the motherboard does voltage control.

This will no doubt be enabled in some latter revision and for mobile and servers, which will only increase perf/watt once more, especially.

>>59284387
And I don't know anyone who is doing hevc given two different patent pools and a retarded license fee that has profit leeching stipends.

File: firefox_2017-03-07_10-26-19.png (14KB, 600x147px) Image search: [Google]

14KB, 600x147px

>>59284402
Being on Samsung's LPP helps a lot.

We're looking at 1.5-2x perf/watt over Xeons on almost everything except AVX.

lets go bois

>>59284392

In your dreams, family.

>>59284420
I was thinking about building a mega esxi home lab server out of it.

>>59284449
x264 10bit is great on PC, but it lacks ASIC support unlike H.265 Main 10.

>>59284471
Eh, you can bulldoze it through software with 8 cores without issue.

>>59284468
Well I mean I based the motherboard and ram prices off of what it would currently cost to buy a lower end 2P 2011-v3 supermicro board and 128gb ddr4 ecc ram. If you say "nuh uh it's not that cheap" then I'll fucking link you.

You think 2* $700 for Naples 2P compatible CPUs isn't high enough?

Is there a chance that AMD has an answer to AVX, or simple doesn't need it?

>>59284500

Yeah I was aiming my comment at the 2P estimate, but I haven't seen the full SKU lineup so who knows

>>59284494
It causes a battery issue on mobile. I don't know why they didn't develop hardware decoder for H.264 Hi10.

>>59284530
Because only Daiz uses it.

>>59284453
>Wendell
This seems to be the go-to guy for finding out about architecture besides RWT when you're interested in more than mugaym

>>59282670

AMD is on everyone side right now. Even if you only buy intel their prices are dropping because of AMD offering an actual viable option even if they are not top of the line.

>>59284453
The server part will also probably use an integrated voltage regulator instead of relying on the motherboard which will further improve efficiency another 10-20% I believe.

For everyone saying AVX is important, and AMD cant compete with it, AVX isnt as important as you think, AMD can compete.

Not all problems/work loads can be vectorized easily, many not at all. It often isnt worth the man hours to vectorize a problem thats hard to vectorize. You often have a mix of scalar and vectorized problem sets, so the theoretical gain of AVX is less in the real world, where on average AVX is a ~10% increase of perfomance on workloads that can use AVX to begin with.
A 10% gap can be made up by AMD with Naples superior bandwidth, efficiency, and likely price.

>>59284527
If I use retardo estimates,

$300 1700x / 8 cores = $37.5 per core

$700 / 37.5 = 18 and two thirds of a core

Maybe we'll get ~$700 12 or 16c / 24t or 32t Naples?

>>59284567
All these AVX guys don't really know what they're talking about, not all workloads saturate two 256bit pipes, in fact most don't at all.

AMD's only issue is that it can't do AVX2 at 2x256, it can do 2x128bit without issue.

>>59284578
I doubt there will be 12c ones. Maybe not even 16c unless they just make them more geared toward the "home server" market.

It makes more sense, in this market, for them to just disable the cores or CCX that are actually busted and sell a 30c, 28c, etc, instead of disabling 1 core on every single CCX and only selling it as a 24core, or disabling 2 on each and selling it as a 16core.

>>59284722
16 core ones are confirmed as Snowy Owl platform, but anything lower than that is not, but it would make no sense, not everyone is running huge 1000 core+ clusters, SMB's don't need that.

>>59284722
What's your estimate on how much the full 32c/64t Naples part will cost?

>>59284764
24 core Broadwell's go up to $8900 so take a guess.
If the Naples is faster, more efficient, has more I/O total memory and bandiwdth, there's no fucking way it would be much cheaper.
The people who buy these monsters don't care about hardware cost, they care about TCO

>>59284600

You bet the hardware guys at every HPC/server farm/cloud etc are going to want a test system to play with.

I expect release in mid May to early June. They are defintely ironing out as many problems as they can in the next few weeks before they tape out final silicon.

>>59284812
>large companies are using gaymers to validate firmware, microcode and software

Works every time.

File: 1448108916364.jpg (38KB, 408x439px) Image search: [Google]

38KB, 408x439px

>>59284800
But I wanted a 64c/128t homelab, anon

play DOOM vulcan with this many core

>>59284829

Genius isnt it? I even think the R7s will turn out alright in regards to gaymen in the next couple months once MS and mobo makers get their shit together, never mind once game devs get a clue.

>>59284892
Too bad, wait for used ones in a few years.

>>59284898
you laugh, but people do that
https://www.youtube.com/watch?v=kYH7eVp6r6I&t

>>59284892
>2P 32 Naples setup with adequate memory
>under $15k optimistically

In your dreams.

>>59284904
The R5s are gonna completely miss all of these problems when they enter the market. It will be like a second launch of Ryzen

>>59284764
At least $5000.

File: 09560575.jpg (36KB, 362x367px) Image search: [Google]

36KB, 362x367px

>>59284402
Pretty retarded how some cherry picked game benchamrks at 720p destroyed Ryzen's image. The processor is amazing.

>Xeon-d competitor with similar I/O at lower price and much higher clockspeed

Jesus Christ When.

>>59284951
Oh well. It got stock to dip so I could buy in more at $13.

Do the Broadwell Xeons even support 4 sockets? I haven't seen that on the wiki, maybe only the older Haswell's do?

>>59284924
that core utilization, it better then porn

>>59282929
you seriously dont know shit eh when an app runs out of memory it results on a memory leak thus crashing

when an app results on having x4 the data to play with and the memory cant feed fast enough it gets parked and thus crashed because the app will always send MORE data than the memory can handle

https://www.youtube.com/watch?v=bIbwuLdHbMg

>Wendell talks about stuff while everyone else is confused, as expected of youtube reviewers who can only read marketing sheets

File: 1484109919337.jpg (77KB, 640x638px) Image search: [Google]

77KB, 640x638px

>>59284924
One day, everyone is going to be laughing at everyone who said 4cores will always be adequate for gaming, won't they? Including the people who once said that themselves as if they weren't one of those dumb fucks.

>>59285019

I'd watch this if I could cut Brian out and not give him a view. He's a meanie.

>>59285001
Stuttering is triggering me, though.

>>59285019
I remember that moment. It actually infuriates me.

Those other fucks aren't qualified to be reviewing hardware.

>>59284972
Broadwell-EP/X is only dual socket.

>>59285019
He's in a completely different league, he's a real hardware enthusiast.

>talks about IOMMU grouping
>everyone else looks durf

>>59285010
>App

Get the fuck outta here

>>59285084
sorry next time i will type it correctly

a bunch of zeros and ones brought together in such a way that can maximize efficiency

>>59282798
Intel phi/knights landing 72 cores... derp. http://ark.intel.com/products/95831/Intel-Xeon-Phi-Processor-7290F-16GB-1_50-GHz-72-core

>>59285019
How could you even stand listening to these other fucks?

>>59285019
I bet 3 of those faggots haven't booted from PXE once in their life and others think these people are qualified to review shit.

>>59285136
>Silvermont cores

Dude.

>>59285141
by being australian

>>59285141
I hold a high tolerance for incompetence and bullshit that doesn't directly harm me.

>>59285136
It's really 18 cores and Xeon Phi is shit.

>>59282479
Kek. ARM not only got kicked off the server space by Xeon Ds even AMD brutally murders them with Naples.

>>59285188
ARM ain't out for high performance though, ARM will do great where performance isn't needed but I/O and memory are.
So.. dense servers, microservers and data hosting without too much concurrent connections.

>>59285136
and amd is bringing the EHP your point is?

>>59285019
Is talking about memory timings and mhz the most these 3 can do? Pretty sad if that's the extent of their knowledge, I'm 35 minutes in.

>>59285228
>Is talking about memory timings and mhz the most these 3 can do
They can do it in an Australian accent.

>>59285019
didnt knew about this channel..
damn lol one guy explains everything and the rest are just
lol ok
yeah i agree
hmm yeah
LOL

>>59285266
https://level1techs.com/

When you compare a mainstream baby like Linus to Wendell, Linus looks like he just got rid of his diapers.

>>59282785
Kek what kind of multicore workloads are not memory bound? Literally all of them are.

>>59285287
linus is a good speaker and nothing much he has some knowledge i cant deny that but this guy is just top notch

every time he talks about amd he smiles i bet he wants to carry jim (kevin sorbo) keller kids

>>59284186
>shilling a discord here

>>59282785
calling out amd bs

amd builts something better and compares it with the better 2p intel can offer

literally they are going to compete on the same level and yet its bs to showcase that they are going to lay waste

Point is. All you cucks consistently compare grapes to wine. AMD puts out new tech and cucks flock comparing it to Intel's old tech. Ryzen is Shit, compared to my Sandy bridge 3930K @ 5 Ghz on xspc 240 water kit. All you GayMD cucks need to pull your head out your ass. It's been 5+ years and AMD can't double the performance of the 3930k. SHAME ON YOU ALL FOR SUPPORTING GAYMD AND SHAME ON GAYMD FOR THEIR SHIT CPU AND HYPE.

>>59283508
It's:
> 2 sockets
> 4 Zeppelin dies per socket in an MCM
> 2 DDR4 channels per Zeppelin die
> 2 DIMMs per DDR4 channel

i.e., 8 channels and up to 16 DIMMs per socket, 16 channels and up to 32 DIMMs in a 2S server.

>>59282785
Is this really AMD's fault when the only product close to their new product still has less channels?

>>59284924
Interdasting any other game that use that many core

>>59285385
when naples comes out shills will say that hpc doesnt matter anymore

CAUSE THEY CANT GAME

>>59284270
Because GPUs can't have 512GB of RAM. What's the point using a GPU with 512GB/s memory bandwidth when you have to constantly send data beyond 12GB over PCIe at a rate of 8GB/s?

>>59285019
>this guy has to explain to Ed, Edd and Eddy how the CCX(and what it is)works, how load balancing works, and how Windows' scheduler is shit at the moment since it missreads caches and shuffles threads-willy nilly

Jesus Christ this is painful to see, who are these 3 monkeys and how do they not know any of this? This was literally known on day one.

Give me names of their sites/channels so I can blacklist them.

>>59285287
Linus is still better than those other 3 youtuber fucktards that Wendell was talking to.

>>59284319
>Implying the hype wasn't real.

File: 1473291918535.png (123KB, 785x757px) Image search: [Google]

123KB, 785x757px

>>59282554
>this is what happens if you didn't fell for the 16GiB RAM meme
>mfw I fell for the 16GiB RAM meme
who's laughing now ramlets???

>>59284319
Basically everything AMD told us about Ryzen was true /g/ expectations were just set way too high. Ryzen has already met all of my expectations plus I think there's still a lot of work that can be done to improve it even further.

File: Intel.jpg (34KB, 534x345px) Image search: [Google]

34KB, 534x345px

>>59283058

How does it compete tho? (Real question, not sure how this things compare)

>>59285438
Perhaps, but he's still limited to general knowledge, and would probably choke when asked what's the difference between 8 and 4 way cache associativity and difference between home and early snoop modes.

>>59285512
when the mapping is done on windows (its on the fast ring now on content creators update) you will see quite a lot of perfomance uprizing

>>59285568
just imagine that those firepros there have 2 nvme dedicated on the cards that are completely independent of the system up to 1tb

it will be obviously a beast

>>59285568
it competes directly with teslas
but tesla has to rely on intel and all bells and whistles from separate vendors
amd is all in house

q3-4 financial report will be fun to see from all three companies

>>59285512
>/g/ expectations were just set way too high

You mean /v/ gaymer expectations. DO NOT EVER talk about gaymes on /g/!!!

>>59282554
>https://ark.intel.com/products/96899/Intel-Xeon-Processor-E5-2699A-v4-55M-Cache-2_40-GHz
> Max Memory Size (dependent on memory type) 1.54 TB

Wonder why they bother to equip amd system with 512 gb of ram and only bother to have 384 on intel one, despite that you can have up to 3 tb of memory on intel side using something like X10DRi-LN4+ for example.

>>59285568
It's mostly used to build supercomputers because they require you to specifically programm for them.

File: intel.png (21KB, 653x174px) Image search: [Google]

21KB, 653x174px

No worries guise, everything is under control

>>59285666
its a rack you idiot a 2p server its a system intel sells with 384 gb

>>59282479
>Chipset & CPU on the same PCB -- SoC.
that's not what SoC means...

>>59285666
Because $/GB is highly super-linear above 16 GB/DIMM.

32 GB DIMMs aren't too far outside the realm of reason, but 64 is extravagent and 128 is essentially non-existent at this point in time.

You could do similar tests with 768 and 1024 GB builds and get similar results, so nothing really changes.

It's also worth noting that Xeons need fully buffered DIMMs when using 3 per channel (LRDRIMMs won't cut it), so there are small performance penalties when trying to max out slots with lower capacity DIMMs as well.

>>59285736
Yeah that's clearly a SOM.

>>59282479
This thing is an absolute nightmare for NUMA tuning

With that said, I'm sure once figured out it would be way faster

>>59282554
MAKE SCIENCE GREAT AGAIN

Can it run couchdb?

>>59285687
AHAHAAH

>>59285434
It would have been better for the 3 stooges to not even be there, this way they just end up looking like idiots.

>>59282785
NOPE, try out looking for Quad Channel against Dual Channel, The only place quad channel shows superiority is on FutureMark, everywhere else the results are negligible.

>>59285687
>intel can't let process shrinks carry their lazy asses anymore
>glofo on the cusp of 7nm once EUV stops being vaporware

Holy shit, what a time to be alive. Intel might finally be forced to innovate in processor architecture.

>inb4 itanic pt. 2 electric boogaloo

>>59285576
>hey $youtubereviewer what's a virtually indexed, physically tagged cache?

>aaaa..
>urrgh...
>eeeh..
>uuug..
>Hey, did you play GTA5 recently!?

File: IMG_0270.jpg (24KB, 302x326px) Image search: [Google]

24KB, 302x326px

>>59282554
>insufficient memory to load

Intel can't fit as much memory, not to mention I/O, holy shit the I/O is insane in a 2P rack.
This is hands down AMD's win even without the core count, power draw or memory bandwidth.

>>59285687
>B-B.B.-but muh 480p benchmarks!

File: Screenshot from 2016-07-14 22-27-20.png (409KB, 621x635px) Image search: [Google]

409KB, 621x635px

>>59285687
>tick>tock>optimization>optimization>PANIC

http://www.pcworld.com/article/3168319/components-processors/intel-demotes-pcs-giving-datacenter-chips-first-crack-at-new-technologies.html


On fire, ass is literally already 20cm down in the magma.

>>59285687
It's alright, just add a longer pipeline

>>59285859
>>glofo on the cusp of 7nm once EUV stops being vaporware

So like, by 2030, by the time IBM and Intel are alraedy mass producing Graphene and InGaAs chips?

>>59285982
>tick>tock>optimization>optimization>PANIC

You mean

>tick>tock>node shrink>node shrink>node shrink>PANIC

>>59285696
> its a system intel sells with 384 gb
> intel sells
Point to concrete model, if you so enlighten.

>>59285746
Nothing against price per gb in realm of 64 or greater sticks, just asking why they bother to equip their sample with 512 and cut quarter of that in intel system.

File: SEASONED2.jpg (46KB, 287x404px) Image search: [Google]

46KB, 287x404px

>>59282554
>insufficient memory to load

>>59285880
The issue is that companies don't pay these guys to know low-level IC knowledge, they pay them to sell to normalfaggots.
Companies know that the ones who know these things don't need youtube reviewers and are competent enough to find stuff themselves, plus they're able to come to a conclusion by themselves, so marketing to them by definition is worthless because general marketing for the masses will never change a smart mind.

>>59284239
1 Terabit network bandwidth is nothing special in network industry, especially network provider such as AT&T, Verizon, etc.

>>59283058
God damn, This Radeon Instinct is a little beast on the papers and they put 16 of them in it? All that power in a single server drower, will a room filled with it finally find the cure for cancer or will it be used to spy on china?

>>59283175
If AMD gives to him, cause he didnt pay shit on his shill cabinnet.

>>59283058
Take the PSU number and divide it by the GPUs, you get 410W per GPU, without including CPUs, DIMMs, SATA/SAS/NVME drives/NICs

Then remember that PSU's run best around 55% efficiency, and buying a 6000W PSU rack for 5500W of hardware is dumb.

tl;dr VEGA IS EFFICIENT AS FUCK

>>59283494
It will probably double.

>>59286185
>tl;dr VEGA IS EFFICIENT AS FUCK
Considering how Zen is efficient I wouldn't be surprised if some secret sauce from the CPU division made it into RTG

Looks like Intel will have to cancel their sabbath

>>59285311
Every fucking esx cluster I've ever worked on has always been memory constrained.

>>59286185
Vega is 220W

>>59285587
Windows has two issues to fix, SMT topology and scheduling issues related to the CCX.
Both are across the board performance improvements.

Motherboards dude still have buggy BIOSes still, and that's more performance.


What a crazy launch.

>>59286237
Source? For 12.5TFLOP fp32 that seems even lower than the 250W Titans

File: 1433978183190.jpg (138KB, 980x1040px) Image search: [Google]

138KB, 980x1040px

>>59282554
Fucking RAMlets, when will they learn?

>>59286185
>VEGA IS EFFICIENT AS FUCK
Eh basically slightly more efficient than Maxwell/Pascal.

As much as Intel is dropping the ball, Intel isn't fucking around. I don't understand how Nvidia's market cap is so high, but whatever.

>>59284344
Look at the stock price next week.

Right now investors are either went full sell after Ryzen's huge success or went full wait to the lowest stock price and go buy.

File: 1354222592741.gif (2MB, 320x240px) Image search: [Google]

2MB, 320x240px

>>59282554

>>59286266
Nvidia sells expensive as fuck shit.

Wow, just one 2p server can replace the 6 host esx cluster I had at my last job (which served like 300-400 vms).

>>59282772
Most 2011-v3 server boards have, from what I see, 12 dimms per socket. Naples has 16. Since dimm cost at the highest capacity goes into infinity, the Naples will naturally get more memory per socket per dollar spent.

>>59286217
A lot of the efficiency gains in Vega are reverse engineered from Maxwell. Kek. Though a lot are all AMD, like its superior cache and hardware memory controller and schedulers.

I don't think anything with Zen is really benefiting it. They're on a different manufacturing process.

>>59285982
>>59286025
Hm well going enterprise first for optimizations make sense.
It also means Skylake-X will come sooner. Meaning HEDT 8-12 cores. But how will they be priced? If Intel tries to price a 12 core at $2000, AMD will just bring 16c/32t quad channel Ryzen to desktop at $1400.

I think Intel didn't think AMD would compete so well with servers, but they already knew even with the most optimistic expectations for them and worst for AMD that AMD would still compete in PC sales with their APUs that were coming. Even with Zen was a super fuck up and barely got a 40% IPC increase and 3ghz speeds, they'd still dominate the laptop market just thanks to being on the 14nm process.

But now Intel has nothing going for them except the G4560 and 7700k.
Two processors out of 100+ SKUs.
And in 2019 they're mega fucked when Zen+ comes with 30% performance improvements and a drop to 7.5nm while Intel will be on ~9.3nm?

>>59284319
what? you expecting gaymen benchmark again?

>>59282479
coming soon on /g/:

>not having 512GB of RAM
what are you poor?

>>59286257
I'm pretty sure the measurement for both it and the Titan you're seeing is fp16.

I'm not that guy and I don't have a source, but 220W sounds right. The difference is that 8GB of HBM is way, way lower power than 12GB of GDDR5X. I'm not sure how much exactly, but it's a significant amount.

>>59286434
It supports 2TB of RAM, anon. Anything less is peasantry.

Yeah, let's hype X's amd thing that is coming out in the future and how it will totally kill competition and all that crap that, you know, you guys say literally EVERY SINGLE FUCKING TIME AND IT HAS LITERALLY NEVER BEEN TRUE FUCKING EVER?

This has got to be stockholm syndrome now.

>>59286535
Sorry, but you're alreadly proven wrong
https://www.servethehome.com/amd-ryzen-7-1700x-linux-benchmarks/

File: 1487258141504.png (973KB, 801x1500px) Image search: [Google]

973KB, 801x1500px

>>59285998
never get really tried of this.

>>59284319
>>59285512
>amd says broadwell ipc
>everyone really expects haslel tier
>gets skylake tier
>for some reason it's bad 'cuz isn't better than kaby lake in a single-threaded bench
lmao

>>59286580
This reminds me of my early days. Intel were all over the TV and Newspaper ads. The first time I head about AMD was when I saw a newspaper with a store listing. They were selling PC with AMD Athlon that costs much much less than the Intel counterpart. I had never ever heard of AMD before this, there were no TV ads, no Newspaper ads, no science magazine talking about it still, they managed to be noted by people. I'm not from USA btw so this is why in the early 90's, 8yo without much access to information wouldn't know anything.

>>59282623
Even Intel server boards with 16 DIMM slots have this problem, it has to do with the PCB trace placement

>>59286631
And they keep saying it is worse than Ivy Bridge. Even worse, before it, they said that if it was 10% behind broadwell for less it was totally fine. Now We get a low cost, same Broadwell IPC, better SMT, lower power usage, cheaper build overall and people shit at it. I really don't get it.

>>59286683
I did my first built in 2002. all of the retailers was selling only intel p3s and celerons. I had to go to another city to buy my duron 950 mhz. I loved that chip.
I'm not from the us either by the way.

>>59286683
>>59286845
I was 13 when I made my first computer in about 2002.

Intel was in all the pre-built computers. Gateway, HP, Comp...something. Whatever, those old brands.
I had Pentiums in computers that my parents got me. Before I built my own computer, I just thought running like complete dogshit was just something that computers were supposed to do. I swear, this Pentium 3 one went faster if I repeatedly kicked it. And my dad beat me for kicking it. :(

When it came to build my own, I had read a little and it seemed these benchmarks showed that the AMD was better for cheaper. That despite the lower Ghz, it outperformed Intel.
Like a Tbird 1.33Ghz was I think.. $130? And a Pentium 1.5Ghz $180? But the Tbird 1.33 was faster. MUCH faster in some games, by 25%+.
And the guy in the store said that yeah, I should probably get the Tbird. Motherboard was cheaper too.
I knew nothing about the GPUs. I just knew some games needed at least 128mb. The 9600pro had 256mb.

Man I lucked out... all the shilling was telling me buy Intel goy, but no this PC was amazing.
And then I start meeting people online in games. They're all like, yeah, AMD is so much better for the more. Overclock so good, so much better performance even at lower clocks.

But no one else knew that. Just enthusiasts that built their own PCs. It really was a scam. It's crazy that Intel only had to pay AMD a billion dollars for that anti-competitive suit and not 10 billion.

>>59286631
The IPC is Skylake tier or close, but the clock speed is so low we can't take full advantage of it. Still, it doesn't deserve the garbage it's had thrown its way. It's like everyone secretly wanted it to fail

>>59286046
Because *well Xeons have 4 DDR channels per socket with up to 3 DIMMs per channel, so 1 unit/2 socket/24 DIMM slot is by far the most common server build.

The standard Naples build will be 1U/2S/32 DIMM, which gets you a third more capacity.

It's very common for server farm operators to spend supstantially more on memory than on all other components combined in a machine, so getting more capacity without having to get raped on high density DDR is actually easily one of the top value propositions of Naples.

File: 15397820_10154079266158244_1724267271_o.jpg (420KB, 1080x1920px) Image search: [Google]

420KB, 1080x1920px

>>59282479

>Amd trying to innovate anything
>Still has this bug

Yeah I'm sure it'll be great op

>>59286482
Gp102 is 25tflop fp16, like Vega

>>59287646
I stopped getting that bug years ago with my 7970.

>>59287664
I'm presuming that Vega will do int4 way more efficiently and economically, then, to be competitive in the server market.