Register a SA Forums Account here!
JOINING THE SA FORUMS WILL REMOVE THIS BIG AD, THE ANNOYING UNDERLINED ADS, AND STUPID INTERSTITIAL ADS!!!

You can: log in, read the tech support FAQ, or request your lost password. This dumb message (and those ads) will appear on every screen until you register! Get rid of this crap by registering your own SA Forums Account and joining roughly 150,000 Goons, for the one-time price of $9.95! We charge money because it costs us money per month for bills, and since we don't believe in showing ads to our users, we try to make the money back through forum registrations.
 
  • Post
  • Reply
Combat Pretzel
Jun 23, 2004

No, seriously... what kurds?!

Zen+ is scheduled for when? Beginning 2018? The B2 stepping ain't it, right?

Adbot
ADBOT LOVES YOU

Twerk from Home
Jan 17, 2009

This avatar brought to you by the 'save our dead gay forums' foundation.


Does anybody know when the Epyc NDA lifts today?

Risky Bisquick
Jan 18, 2008

PLEASE LET ME WRITE YOUR VICTIM IMPACT STATEMENT SO I CAN FURTHER DEMONSTRATE THE CALAMITY THAT IS OUR JUSTICE SYSTEM.



Buglord

3pm ct is the launch, Lisa has 30 minutes to wow us with launch partners otherwise its AH/PM trading

nvm I was right/wrong

https://twitter.com/AMD/status/877167059448233984

Risky Bisquick fucked around with this message at 15:33 on Jun 20, 2017

eames
May 9, 2009



Combat Pretzel posted:

Zen+ is scheduled for when? Beginning 2018? The B2 stepping ain't it, right?

There's no roadmap but rumors suggest 12 months after the Ryzen launch, so Q1/2018 although I feel like that would be quite ambitious.

B2 stepping is most likely going to be a minor bugfix/errata revision and won't be advertised separately.

lDDQD
Apr 16, 2006


eames posted:

There's no roadmap but rumors suggest 12 months after the Ryzen launch, so Q1/2018 although I feel like that would be quite ambitious.

B2 stepping is most likely going to be a minor bugfix/errata revision and won't be advertised separately.

There's no B2 stepping, it's just techpowerup rumor mongering again.

BangersInMyKnickers
Nov 3, 2004

I have a thing for courageous dongles



Paul MaudDib posted:

Consider the following: Threadripper mATX boards with triple PCIe 3.0x16 slots, into which you can attach NVMe SSDs on those quad sled things. Then you do soft-RAID with maybe like a 3-wide RAID10 or something. Or just one single 24 TB dataset. While I'm wishing for a pony how about also having the same 8 SATA + 1 NVMe M.2 of the Asus P10S-M WS, so that you can attach 8 commodity HDDs for secondary storage and a final NVMe M.2 drive with a buttload of L2ARC cache.

Or you could connect an InfiniBand card in one of the PCIe slots and get remote DMA going and poo poo. No, let's go with having that engineered into the board too.

Asrack you better not disappoint, I have faith in your crazy design shenanigans!

I think you'd be better of with a U.2 interface to do that work and get the drives in a hot swappable form factor

rex rabidorum vires
Mar 26, 2007



Old system died last week. Motherboard gave up on life I think. Took a trip to Microcenter and came home with a 1600, AsRock AB350M Pro4, and EVGA 3200. I picked the 1600 because I needed the fan to get my system up quickly and once my Cryorig bracket shows up it will be replaced. Unfortunately MC didn't have any of the better B350 boards so it was either a cheaper B350 or spend the extra for a X370 that I honestly don't need. The EVGA is Hynix B die (I think) so in theory should be able to run it at a reasonable speed. Plugged everything in when I got home and booted right up only to find none of the USB plugs worked because I didn't update to Win 10. A PS/2 keyboard solved that issue and the driver installation went fine. Updated to Win 10 and everything was functioning as it should.

The ram was a bitch and would not run above 2133 no matter what I did. Granted it isn't on the QVL list so chalk that up to me being dumb and not taking my own/others advice while also being in a rush. So checked ryzmem and found that Corsair LPX seemed to run pretty well and was also on the QVL list. After some toying I was at least able to get it to run at 2993 with loosened timings. Checked stability and it was fine. Moved on to OC'ing the chip and was able to get to 3.6 across all cores with stock voltage. Ended up at 3.8ghz with 1.325v which was stable after a hour of Prime95 running and the temps are well under control. Played with ram one last time and was able to get it to go to 3200 and factory timings. Pretty happy with that.

eames
May 9, 2009



Server launch happening right now at http://www.anandtech.com/show/11562

I don't think there's a livestream.

e: NDA lifted; overview at http://www.anandtech.com/show/11551...d-epyc-analysis

eames fucked around with this message at 20:02 on Jun 20, 2017

Risky Bisquick
Jan 18, 2008

PLEASE LET ME WRITE YOUR VICTIM IMPACT STATEMENT SO I CAN FURTHER DEMONSTRATE THE CALAMITY THAT IS OUR JUSTICE SYSTEM.



Buglord

Dell/HP/SuperMicro confirmed OEMs so far, no cloud customers yet. If they announce even one

Twerk from Home
Jan 17, 2009

This avatar brought to you by the 'save our dead gay forums' foundation.


Risky Bisquick posted:

Dell/HP/SuperMicro confirmed OEMs so far, no cloud customers yet. If they announce even one

Azure on stage right now. Any vendor actually taking orders or are these not showing up until December?

eames
May 9, 2009



Can't help but be impressed to see this announced by a company that was on the brink of bankruptcy about a year ago






source:
https://twitter.com/TheRegister/sta...265375439749120

eames fucked around with this message at 20:54 on Jun 20, 2017

bobfather
Sep 20, 2001

I will analyze your nervous system for beer money

Just being slightly cheaper won't win them the server space, I don't think. The TDPs on most of the lower-end Epyc SKUs are insanely high compared to the Intel comparisons. Server farms care pretty dearly about power efficiency.

Subjunctive
Sep 12, 2006

careful now



drat.

eames
May 9, 2009



well tom's hardware wrote this but I couldn't find a slide to confirm it so

tomshardware posted:

AMD provided some basic benchmarks, seen in the slides above, that compare its processors to the nearest Intel comparables. The price and performance breakdown chart is perhaps the most interesting, as it indicates much higher performance (as measured by SPECint_rate_base2006), at every price point. It bears mentioning that Intel publicly posts its SPEC benchmark data, and AMD's endnotes indicates that it reduced the scores used for these calculations by 46%. AMD justified this adjustment because they feel the Intel C++ compiler provides an unfair advantage in the benchmark. There is a notable advantage to the compiler, but most predict it is in the 20% range, so AMD's adjustments appear aggressive. We should take these price and performance comparisons with the necessary skepticism and instead rely upon third-party data as it emerges.

Risky Bisquick
Jan 18, 2008

PLEASE LET ME WRITE YOUR VICTIM IMPACT STATEMENT SO I CAN FURTHER DEMONSTRATE THE CALAMITY THAT IS OUR JUSTICE SYSTEM.



Buglord

quote:

Cloud Datacenter and Enterprise Customers
Datacenter and cloud service providers also welcomed EPYC to the market today. Members of the “Super 7” datacenter services providers, including Baidu and Microsoft Azure, as well as 1&1, Bloomberg, Dropbox and LexisNexis, all voiced their support at launch.

“As the world’s largest Chinese language search engine and leading AI-Tech company, Baidu prides itself on simplifying a complex world through technology,” said By Dr. Zhang Ya Qin, president of Baidu. “The AMD EPYC processor powered one-socket server can significantly increase our datacenter computing efficiency, reduce TCO and lower energy consumption. We will start deploying with the launch of AMD EPYC and I look forward to our cooperation leading to scaled EPYC adoption this year, and ongoing innovations.”

“We’ve worked to make Microsoft Azure a powerful enterprise grade cloud platform, that helps guide the success of our customers, no matter their size or geography,” said Girish Bablani, corporate vice president, Azure Compute, Microsoft Corp. “To power Azure, we require the most cutting-edge infrastructure and the latest advances in silicon which is why we intend to be the first global cloud provider to deliver AMD EPYC, and its combination of high performance and value, to customers."

to the moon

MaxxBot
Oct 6, 2003

you could have clapped

you should have clapped!!


drat $600 for a 16-core server CPU, this is a really good time for people wanting to build a workstation.

Anarchist Mae
Nov 5, 2009

by Reene


Lipstick Apathy

bobfather posted:

Just being slightly cheaper won't win them the server space, I don't think. The TDPs on most of the lower-end Epyc SKUs are insanely high compared to the Intel comparisons. Server farms care pretty dearly about power efficiency.

Worst case the Epic at 32 cores comes in at 5.6w/core, and the Xeon at 22 cores comes in at 6.5w/core.

Anime Schoolgirl
Nov 28, 2002






the lower 16 core SKUs look like discount runt bins for cheap workstations.

eames
May 9, 2009



Anime Schoolgirl posted:

the lower 16 core SKUs look like discount runt bins for cheap workstations.

so that's where the R3 dies went...

GRINDCORE MEGGIDO
Feb 28, 1985


R7 1700+ working with DDR4 3000 out of the box, not mine, but a friends. Vengeance LPX.

The Spire cooler it came with is insufficient with the high ambient (for the UK, anyway, of 32degC) for any clocking prolly, it's hitting 75deg at 3.5Ghz all cores.

mmkay
Oct 21, 2010


eames posted:

well tom's hardware wrote this but I couldn't find a slide to confirm it so

The slides are on techpowerup

https://www.techpowerup.com/reviews...itecture/7.html (second to last)

Anime Schoolgirl
Nov 28, 2002






GRINDCORE MEGGIDO posted:

R7 1700+ working with DDR4 3000 out of the box, not mine, but a friends. Vengeance LPX.

The Spire cooler it came with is insufficient with the high ambient (for the UK, anyway, of 32degC) for any clocking prolly, it's hitting 75deg at 3.5Ghz all cores.
you still have to put in requests for am4 mounting kits on virtually every aftermarket cooler, literally nobody was prepared for ryzen

GRINDCORE MEGGIDO
Feb 28, 1985


It deserves a noctua for sure. It's quiet though idle, very quiet but ramps up.

It's good and stable though, pleased the ram hit it's spec. Although it only would in the two non recommended ram slots (it's dual channel though, 43GB/s read). Asus prime b350.

E- AM4 mounting system is great.

GRINDCORE MEGGIDO fucked around with this message at 21:59 on Jun 20, 2017

SourKraut
Nov 20, 2005

POST QUALITY UNDER CONSTRUCTION




mmkay posted:

The slides are on techpowerup

https://www.techpowerup.com/reviews...itecture/7.html (second to last)



Hmm..

Kazinsal
Dec 13, 2011






They ran the Intel systems with the RAM downclocked to 2133 MHz too? WTF?

Cygni
Nov 12, 2005

raring to post



Anime Schoolgirl posted:

the lower 16 core SKUs look like discount runt bins for cheap workstations.

The Epyc boards are gonna be really really expensive and complicated. A threadripper + X399 board will be cheaper, unless you need the 128 PCI-Ex lanes (like for deep learning or trying to become rich in cosbycoins) or the 2TB of memory.

wargames
Mar 16, 2008

official yospos cat censor


Kazinsal posted:

They ran the Intel systems with the RAM downclocked to 2133 MHz too? WTF?

Slower rams effect intel way less then amd.

VealCutlet
Dec 20, 2015

I am a marketing god, shave that shit


wargames posted:

Slower rams effect intel way less then amd.

Yeah I've noticed a lot of the x299 reviews comparing with Ryzen stuff running at woeful memory speeds... guess they are trying to keep intel happy.

I know just the jump from 2933 to 3200mhz on my 1600 in cinebench made a noticeable difference. Going from this 2133 to 3200 would be rather drastic I feel.

Paul MaudDib
May 2, 2006

"Tell me of your home world, Usul"


Subtracting performance from Intel's benchmark results is bullshit which completely invalidates all of those results. And just to be clear, they are cutting the performance of the Intel chips almost in half. That's an immense level of bullshit right there.

Yeah, ICC is good at extracting performance from Intel chips. But that's real performance that you would have in real-world usage. It's AMD's responsibility to push patches into LLVM or GCC or write their own compiler if they think they're leaving performance on the table.

Next up: AMD doctors the Vega benchmarks because "NVIDIA has better drivers "

Unfortunately this is par for the course with AMD. Like benching Vega against Titans with crippled CAD drivers, while losing to a card that's almost half the price using the correct drivers. Or the launch benchmarks of Fury X showing it winning against the 980 Ti at 4K with settings nobody would ever use in the real world, while in reality it undershoots by 10%+. AMD will do whatever it takes to put their best foot forward even if it's intellectually dishonest, and the AMD fanbase just eats it up.

Paul MaudDib fucked around with this message at 02:49 on Jun 21, 2017

VostokProgram
Feb 20, 2014



Paul MaudDib posted:

Subtracting performance from Intel's benchmark results is bullshit which completely invalidates all of those results. And just to be clear, they are cutting the performance of the Intel chips almost in half. That's an immense level of bullshit right there.

Yeah, ICC is good at extracting performance from Intel chips. But that's real performance that you would have in real-world usage. It's AMD's responsibility to push patches into LLVM or GCC or write their own compiler if they think they're leaving performance on the table.

Next up: AMD doctors the Vega benchmarks because NVCC emits better PTX than OpenCL

And if they really want an apples-to-apples comparison they could run all the benchmarks with gcc and use those numbers. Arbitrarily cutting numbers is so hilariously bad that I can't believe even the worst marketing idiot would expect it to fly.

Subjunctive
Sep 12, 2006

careful now


Paul MaudDib posted:

Yeah, ICC is good at extracting performance from Intel chips. But that's real performance that you would have in real-world usage. It's AMD's responsibility to push patches into LLVM or GCC or write their own compiler if they think they're leaving performance on the table.

Didn't AMD release a compiler that sucked?

Kazinsal
Dec 13, 2011






Subjunctive posted:

Didn't AMD release a compiler that sucked?

It was basically just LLVM with some patches.

Paul MaudDib
May 2, 2006

"Tell me of your home world, Usul"


Subjunctive posted:

Didn't AMD release a compiler that sucked?

Now does that sound like a thing AMD would do!?

yes

quote:

AMD advertises AOCC 1.0 as having improved vectorization, better optimizers, and better code generation for AMD Zen/17h processors. AOCC also includes an optimized DragonEgg plug-in for those making use of Fortran code-bases on Ryzen and still compiling them with GCC 4.8. There's also an AOCC-optimized Gold linker available too.
...
While we were very excited when first hearing of AMD's AOCC compiler for potentially better performance on Ryzen, with this 1.0 release we found very little change in performance. In most workloads of the dozens of tests ran, the AOCC 1.0 performance ended up being right in line with upstream Clang 4.0. In a few cases AOCC was faster, but it was generally about three percent or less. In some of the cases where AOCC was faster, it was only to then lose out to the GCC compiler being faster yet. In some benchmarks, the Clang 5.0 development code was also faster than AOCC.
http://www.phoronix.com/scan.php?pa...yzen-aocc&num=1

Risky Bisquick
Jan 18, 2008

PLEASE LET ME WRITE YOUR VICTIM IMPACT STATEMENT SO I CAN FURTHER DEMONSTRATE THE CALAMITY THAT IS OUR JUSTICE SYSTEM.



Buglord

Regardless of that, the margins on Epyc are going to be insane for AMD. If they can ramp up some sales volume this is nuts. If they do roughly the same thing to Navi

Twerk from Home
Jan 17, 2009

This avatar brought to you by the 'save our dead gay forums' foundation.


Risky Bisquick posted:

Regardless of that, the margins on Epyc are going to be insane for AMD. If they can ramp up some sales volume this is nuts. If they do roughly the same thing to Navi

It really looks like AMD's platform is focused heavily on i/o. 128 PCI-e lanes on a 1P system are insane. If datancenter nVME drives actually are a high volume product, AMD has a spot.

Malloc Voidstar
May 7, 2007

Fuck the cowboys. Unf. Fuck em hard.

eames posted:

well tom's hardware wrote this but I couldn't find a slide to confirm it so
https://www.spec.org/cpu2006/result...0503-46962.html
ICC is a Sufficiently Smart Compiler for the specific libquantum benchmark. In my experience trying ICC this is way beyond what it normally gives you.

Paul MaudDib
May 2, 2006

"Tell me of your home world, Usul"


Malloc Voidstar posted:

https://www.spec.org/cpu2006/result...0503-46962.html
ICC is a Sufficiently Smart Compiler for the specific libquantum benchmark. In my experience trying ICC this is way beyond what it normally gives you.

What is the significance of libquantum being 10x over in particular? How does that translate to AMD's particular 0.575 multiplier?

So AMD is basically cherrypicking one benchmark with huge gains that are abnormal even for this compiler?

What is the typical speedup in your experience? I've never used ICC. Can you describe the nature of the code (highly parallel/not, level of cross-thread interaction, math-intense/memory-intense, etc)?

Paul MaudDib fucked around with this message at 05:01 on Jun 21, 2017

3peat
May 6, 2010



http://www.legitreviews.com/one-amd...andwidth_195653

quote:

One of the storage demos that they were showing off was of on an AMD EPYC 7601-based 1-socket system with 24 3.2TB Samsung PM1725a 2.5-inch PCIe NVMe SSDs that were over-provisioned to 1.6TB. The end result of this 24 drive setup that costs upwards of $168,000 (24 x $6,999.99 per drive) was pretty impressive.

The live live demo was run using Flexible IO benchmark (FIO) v2.16 and AMD was able to get ~57 GB/s of bandwidth (128KB Random workload) and 435,355 IOPS.

The test was run in an HPE Cloudline CL3150 server running Ubuntu 17.04 4.10 kernel with 256GB of memory (8x32GB PC4-2666). This is actually higher than the speeds they noted in the accompanying presentation slide shown below.


When they switched to a 4K 100% Random test they got an insane 9,178,000 IOPS on the Read and 7,111,000 IOPS on Write. Amazing numbers and that is on a single-socket system! The AMD EPYC 7601 processor was at 75% load during this test, so it takes a tremendous amount of processing power to get 9.1 million IOPS.

Anime Schoolgirl
Nov 28, 2002






and this here is why cloud providers are excited about epyc

Adbot
ADBOT LOVES YOU

3peat
May 6, 2010



Power consumption is looking good

https://www.servethehome.com/amd-ep...n-observations/

quote:

With the level of power/ performance of the new systems, you can essentially replace four Intel Xeon E5-2600 (V1) servers with a single dual socket EPYC node and get more performance (in most cases) in a single node that uses half the power. That is absolutely stellar. The AMD EPYC platform is still seeing major updates to BIOS for power and performance which is why we are calling these preliminary results. At the same time, we are already seeing some impressive figures.

  • 1
  • 2
  • 3
  • 4
  • 5
  • Post
  • Reply