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
Anarchist Mae
Nov 5, 2009

by Reene


Lipstick Apathy

A SWEATY FATBEARD posted:

Hmmm there seems to be a shortage of fast RAM over here in Croatia; the supply is spotty at best and you have to preorder your sticks if you want anything faster than 2400mhz. I bet this has something to do with Ryzens selling like hotcakes.

e: in theory you could get 16 gigs of 3600mhz RAM for around $300. The sticks are listed as "in stock" on the website but when I went to the physical store they told me that these sticks are sold out - ie they list them as available just to lure people into their brick and mortar store. Well I ain't buying a R7 system just to put 2400mhz RAM into it, even though my aging FX system already has a serious hernia thing.

Hmm, is amazon.de an option?

Adbot
ADBOT LOVES YOU

A SWEATY FATBEARD
Oct 6, 2012

Entering of the person whom the nasty smell does is refused.


Measly Twerp posted:

Hmm, is amazon.de an option?

Importing hardware on your own is always an option, and I'll probably do just that. Buying a R7 system in the late fall or so, by then the mainboard/RAM compatibility is going to improve a great deal and bugs in BIOSes will be ironed out.

MaxxBot
Oct 6, 2003

you could have clapped

you should have clapped!!


People were saying that the 1920X didn't have the best pricing, looks like there's gonna be a plain 1920 part that should be more appealing.

http://www.pcgamer.com/amds-partner...n-support-docs/

quote:

AMD has been slow playing the launch of its Threadripper lineup by first announcing the 1950X ($999) and 1920X ($799), then later unveiling the 1900X ($549). Now it appears there is a fourth SKU, one that will not support AMD's extended frequency range (XFR) technology.

The folks at Tweakers noticed that several motherboard makers are referencing an announced Threadripper 1920 CPU in their support documents. So far there are reference from ASRock, Asus, and Gigabyte.

Palladium
May 8, 2012


MaxxBot posted:

Now it appears there is a fourth SKU, one that will not support AMD's extended frequency range (XFR) technology

Which means its cheaper and better right?

EmpyreanFlux
Mar 1, 2013

The AUDACITY! The IMPUDENCE! The unabated NERVE!


Palladium posted:

Which means its cheaper and better right?

Depends actually. There is a rumor going around that the XFR enabled SKUs can hit 4.2Ghz, it's possible there might be legitimate binning differences between X and non-X SKUs for TR4.

MaxxBot
Oct 6, 2003

you could have clapped

you should have clapped!!


The 1920 also has a lower TDP but I'm not sure if it's binned for lower TDP like the 1700 or just has a lower listed TDP due to the lack of XFR.

eames
May 9, 2009



e: wrong thread

eames fucked around with this message at 09:29 on Aug 5, 2017

repiv
Aug 13, 2009



eames posted:

I suspect Threadripper (B1 "stepping") will suffer from the same problem and Epyc will have the fix (B2 stepping)

Someone on Reddit claims his Epyc server is segfaulting on the Phoronix stress test

ufarn
May 30, 2009


God dammit, why must the available CPU options be like this.

Pablo Bluth
Sep 7, 2007

I've made a huge mistake.


Mr.Radar posted:

Phoronix has released a test suite that can reliably reproduce the Ryzen GCC crash bug within minutes both on stock configurations and with both SMT turned off and the memory down-clocked to 2133 MHz. Unfortunately AMD isn't giving them Threadripper or Epyc samples so there's no word yet if they also suffer from this bug.
I've been quite close to hitting buy on the bits for Ryzen upgrade a few times as my current (AMD) windows rig is painful in Lightroom, but I'd stopped myself each time given how poorly Lightroom is written when it comes to making use of cores. This issue has kind of made up my mind to hold off.

I fear this might prove to be rather costly issue for AMD.

Pryor on Fire
May 14, 2013


AMD CPU discussion: gently caress, maybe version 2 will be better

eames
May 9, 2009



AMD CPU discussion: 64C/128T EPY Threadripper gaming CPUs available soon

wargames
Mar 16, 2008

official yospos cat censor


where is the fault, amd or people not updating their software in the past 8 years because amd hasn't had a good cpu since then?

ufarn
May 30, 2009


AMD CPU discussion: Trying to B1 step ahead of Intel.

Ihmemies
Oct 6, 2012



Will be interesting to see how well the new 6core intels work. That gamernexus 1700x vs 7700k review was quite bleak, ryzen still leaves a ton of gpu performance unused.

PerrineClostermann
Dec 15, 2012

by FactsAreUseless


Ihmemies posted:

Will be interesting to see how well the new 6core intels work. That gamernexus 1700x vs 7700k review was quite bleak, ryzen still leaves a ton of gpu performance unused.

Weird, I got the opposite impression

wargames
Mar 16, 2008

official yospos cat censor


Ihmemies posted:

Will be interesting to see how well the new 6core intels work. That gamernexus 1700x vs 7700k review was quite bleak, ryzen still leaves a ton of gpu performance unused.

non-x, and how much ram OC did he do?

repiv
Aug 13, 2009



Phoronix posted part 2 of their segfault investigation: http://www.phoronix.com/scan.php?pa...continues&num=1

Phoronix posted:

So that's where I am at now. Yesterday's testing showed the problem can be reproduced even if SMT is disabled and even if forcing the memory to DDR4-2133. This latest testing shows the problem isn't isolated to GCC that it can also happen with LLVM Clang (in fact, worse) as well as when using the latest GCC 8 development code. Disabling ASLR did not help the situation and running the other stressing concurrent workloads for server/multimedia/scientific didn't yield any segmentation faults in an hour unlike the compilation workloads yielding 50+ per hour. Clearing the BIOS and leaving the defaults also hasn't yielded any change. Some have also said they have been able to reproduce this problem on Epyc CPUs, but unfortunately it doesn't look like we will be receiving any Threadripper / Epyc review samples.

Sinestro
Oct 31, 2010

The perfect day needs the perfect set of wheels.


It's stunningly specific. They should really take the microcode that checks if it's running a C compiler and creates random errors out, I don't think that's going to be their silver bullet versus Intel.

Wait, gently caress. The conspiracy theorist in me almost imagines some sort of insane "Reflections on Trusting Trust" scheme that's not quite working right.

Malloc Voidstar
May 7, 2007

Fuck the cowboys. Unf. Fuck em hard.

From the linustechtips video going live in a few hours:





HalloKitty
Sep 30, 2005

Adjust the bass and let the Alpine blast


Shame the Ryzen system is hobbled with slow RAM... AMD really needs to tweak the architecture in this regard.

Combat Pretzel
Jun 23, 2004

No, seriously... what kurds?!

Would have been nice to see how the TR system would have acted with DDR4-3200 memory. But I guess they compared the system completely stock to their test bench.

SourKraut
Nov 20, 2005

POST QUALITY UNDER CONSTRUCTION




HalloKitty posted:

Shame the Ryzen system is hobbled with slow RAM... AMD really needs to tweak the architecture in this regard.

There's no reason it can't use DDR4-3200, beyond what Combat Pretzel mentioned about using a stock system for the test bench. I'm actually impressed that it did so well when not using DDR-3200.

Anarchist Mae
Nov 5, 2009

by Reene


Lipstick Apathy

Sinestro posted:

It's stunningly specific. They should really take the microcode that checks if it's running a C compiler and creates random errors out, I don't think that's going to be their silver bullet versus Intel.

Wait, gently caress. The conspiracy theorist in me almost imagines some sort of insane "Reflections on Trusting Trust" scheme that's not quite working right.

It seems that what Phoronix is experiencing is in fact normal behaviour while compiling PHP:

quote:

conftest segfaults might be normal, part of autoconf trying ( on purpose ) to find broken configuration options so it knows what to avoid. This would align with my conftest segfaults appearing in my dmesg when test-suite starts a new php run a segault in bash or gcc would be a different story...

However, he was able to cause Clang to crash, but it seems that the kill_ryzen.sh script is still the best method for testing. After 1.5 hours I've not been able to reproduce it yet.

Combat Pretzel
Jun 23, 2004

No, seriously... what kurds?!

^^^ So premature ejaculation about nothing then? Come to think of it, autoconf does generate those configure.sh scripts, which invoke gcc tons of times compiling and running test files? Kind of make sense some of that output breaks.

SourKraut posted:

There's no reason it can't use DDR4-3200, beyond what Combat Pretzel mentioned about using a stock system for the test bench. I'm actually impressed that it did so well when not using DDR-3200.
Yeh, going from 2666 to 3200 would have boosted IF bandwidth by 20% and reduced IF latency a bit. Given how everything's hung up on it, there'd probably a non-negligible effect.

Combat Pretzel fucked around with this message at 18:16 on Aug 5, 2017

TheJeffers
Jan 31, 2007



Malloc Voidstar posted:

From the linustechtips video going live in a few hours:


So they're going to run Rise of the Tomb Raider at 4K on a GTX 1080 Ti and call that a test of CPU gaming performance? :iamafag:

repiv
Aug 13, 2009



Combat Pretzel posted:

^^^ So premature ejaculation about nothing then? Come to think of it, autoconf does generate those configure.sh scripts, which invoke gcc tons of times compiling and running test files? Kind of make sense some of that output breaks.

Confirmed, I'm running the stress test on my Xeon E5-2650L VPS and conftest is throwing segfaults just like the Ryzen systems in the article.

welp i expected better from phoronix

repiv fucked around with this message at 18:43 on Aug 5, 2017

Malloc Voidstar
May 7, 2007

Fuck the cowboys. Unf. Fuck em hard.

TheJeffers posted:

So they're going to run Rise of the Tomb Raider at 4K on a GTX 1080 Ti and call that a test of CPU gaming performance? :iamafag:
It at least shows that Threadripper isn't garbage at games

Anarchist Mae
Nov 5, 2009

by Reene


Lipstick Apathy

Malloc Voidstar posted:

It at least shows that Threadripper isn't garbage at games

Tadaaaaa
https://www.youtube.com/watch?v=decz1N9YpOw

SourKraut
Nov 20, 2005

POST QUALITY UNDER CONSTRUCTION





"Threadripper is pretty good but holy poo poo the Area 51 sucks."

Cygni
Nov 12, 2005

raring to post



TheJeffers posted:

So they're going to run Rise of the Tomb Raider at 4K on a GTX 1080 Ti and call that a test of CPU gaming performance? :iamafag:

Realistically, i think the vast vast majority of people who buy these things for gaming will be using 4k. It's the real world test I guess,with 1080p essentially being a non-real world theoretical test.

TheJeffers
Jan 31, 2007



Cygni posted:

Realistically, i think the vast vast majority of people who buy these things for gaming will be using 4k. It's the real world test I guess,with 1080p essentially being a non-real world theoretical test.

If you're going to take a game that's not generally CPU-bound to begin with and play it at a resolution where it will assuredly be GPU-bound, you're essentially telling your audience that it's basically pointless to buy a $500+ CPU for high-resolution gaming, which is true—the GTX 1080 Ti is the far more important determinant there. It's also the best case for AMD because it basically tells you nothing about how the unique packaging decisions of Threadripper affect gaming performance.

That same graphics card can still bind up certain games on a single thread/core even at 2560x1440 for high-refresh-rate gaming, and I'd bet there's an equally large or larger contingent of loaded gamers with high-refresh 2560x1440 displays who are curious how the chip performs under those circumstances. In that case, you are at least doing something that's both representative and CPU-bound.

All I'm getting at is that if you're going to present a GPU-bound test as a CPU benchmark, you should at least be clear about that with your audience. It tells you basically nothing about CPU performance, and that seems kind of silly (at best) or disingenuous (at worst) for something that's ostensibly a CPU review.

Anime Schoolgirl
Nov 28, 2002






https://www.phoronix.com/forums/for...7437#post967437

quote:

I am running the test in a Core i5 laptop and the conftest segfault appear here too. Michael, did you get only conftest segfaults?

ufarn
May 30, 2009


Never trust someone whose forum URL reads https://www.phoronix.com/forums/forum/phoronix.

Pryor on Fire
May 14, 2013


Sudden plot twist, the empire strikes back. Err wait I don't think that metaphor works. gently caress star wars.

repiv
Aug 13, 2009




Yeah, the Phoronix article was junk and they've retracted it for now.

Phoronix posted:

As a result of feedback, currently working on some updated results. As some have pointed out, the conftest segmentation faults aren't specific to Ryzen, so updating the tests to avoid confusion. Though one area being explored now as well is the Clang segmentation faults shown in the original article, not originating from conftest as well as Clang being able to yield the system hanging hard where the system is unresponsive and SSH is not working. Plus also incorporating more Ryzen-Kill tests as outlined in the aforelinked article. As many readers have pointed out, BSD developers have also discovered a Ryzen bug. More details soon.

shrike82
Jun 11, 2005

Get the fuck in your cage and shut the fuck up, you stupid fucking Mexicans. Your voice doesn't fucking matter unless you support OUR rapist.

OBAMA DID NOTHING WRONG


Bizarre that he's never across this issue given he's spent a lot of time benchmarking Intel CPUs in the past.

Combat Pretzel
Jun 23, 2004

No, seriously... what kurds?!

Too bad, since if that issue would have been real, it'd have kept me financially responsible.

GRINDCORE MEGGIDO
Feb 28, 1985


Combat Pretzel posted:

Too bad, since if that issue would have been real, it'd have kept me financially responsible.

Getting THREADRIPPER?

Adbot
ADBOT LOVES YOU

Combat Pretzel
Jun 23, 2004

No, seriously... what kurds?!

So I'm reading some blurb that the Alienware box(es) that went around for testing might have had a preproduction CPU, and the retail CPUs have slightly better performance. Hothardware ran a review with a 2900 Cinebench score, and was then contacted by AMD, who noted that it might be a preproduction CPU and sent them a new one, which resulted in a score of 3000 and some pocket change. I've noted Linus' Cinebench score also is roughly 2900, so it might be another preproduction CPU, if that poo poo is real. So the current numbers still aren't final.

https://hothardware.com/news/ryzen-...a-51-transplant

I mean, Cinebench, yea. But it's indicative that other metrics might have changed, too.

GRINDCORE MEGGIDO posted:

Getting THREADRIPPER?
WALLETRIPPER!

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