Author Topic: xeon v4 cpus  (Read 49397 times)

2016-07-19, 14:44:35
Reply #75

Juraj

  • Moderator
  • Active Users
  • ***
  • Posts: 4446
    • View Profile
    • studio website
You know it's funny, because in actual rendering performance, it's fast like it didn't even underclock :- ).

So I wonder how much faster it would be otherwise... (maybe not at all, maybe the AVX does speed-up something, although all my friends working with it assured me it's minimal )
talcikdemovicova.com  Website and blog
be.net/jurajtalcik   Our studio Behance portfolio
Instagram   Our studio Instagram, managed by Veronika

2016-07-19, 14:50:33
Reply #76

fobus

  • Active Users
  • **
  • Posts: 388
    • View Profile
So I wonder how much faster it would be otherwise... (maybe not at all, maybe the AVX does speed-up something, although all my friends working with it assured me it's minimal )

May be. But I would like to seу maximum clock speed as it can be without AVX.

2016-07-19, 14:58:14
Reply #77

Juraj

  • Moderator
  • Active Users
  • ***
  • Posts: 4446
    • View Profile
    • studio website
Yeah, that's the biggest crap "feature" ever. Thermal design my ass..
talcikdemovicova.com  Website and blog
be.net/jurajtalcik   Our studio Behance portfolio
Instagram   Our studio Instagram, managed by Veronika

2016-08-05, 15:59:41
Reply #78

peterguthrie

  • Active Users
  • **
  • Posts: 250
    • View Profile
    • Peter Guthrie Visualisation
In case anyone was wondering, the recent Windows 10 anniversary update thing FIXES the start menu and the runtime broker issues we were having! woohoo!

(still cant set affinity though)

so microsoft were obviously aware of the issues, they just didnt make people aware of them

2017-09-28, 19:43:19
Reply #79

Fluss

  • Active Users
  • **
  • Posts: 553
    • View Profile
Just received my dual 2696 v4 OEM. Max all cores turbo @2.8GHz.

Run @2.8GHz all cores in absolutely any app (Vray, cinebench, mining etc etc...), except in corona -> all cores @2.6GHz. So are you really sure this is not a corona related issue?

2017-09-29, 11:01:30
Reply #80

FrostKiwi

  • Active Users
  • **
  • Posts: 686
    • View Profile
So are you really sure this is not a corona related issue?
No commercial renderer before was based off of Embree from the ground up. This is the first time Instruction sets have been used this extensively and optimized, by the very people who build the processors themselves. Even Vray does not fully integrate it.
This collides with this stupid TDP limit "feature" however, as we near a proper 100% full utilization.
It's actually quite goddamn impressive what modern Instruction sets can do.
So it really is not corona's fault. (Maybe for being this fast...)

2017-09-29, 11:42:47
Reply #81

Fluss

  • Active Users
  • **
  • Posts: 553
    • View Profile
OK I understand better. I've just noticed Arnold is running @lowered turbo too.

2017-12-20, 19:36:45
Reply #82

Vuk

  • Active Users
  • **
  • Posts: 109
    • View Profile
Just tried Corona benchmark with 2698v4 es Xeon cpu. I'am guessing this is AVX issues since in Cinebench all cores are running on 2.3 ghz while in Corona benchmark they run on 1.9 ghz? Also tried in distribute rendering in a standard scene not Corona benchmark same problem. Is this issue fixable? Motherboard is Asus z10pe-d8 ws. Haven't tried to play with bios.

2017-12-20, 19:49:50
Reply #83

Juraj

  • Moderator
  • Active Users
  • ***
  • Posts: 4446
    • View Profile
    • studio website
According to recent posts from devs, not really since Intel injects AVX instruction automatically apparently.

Nonetheless if you do some scaling comparison you'll see it's not actual (negative) performance hit. So it's just Intel's decision to keep low turbos for their server cpus (non-WS xeons) to avoid thermal issues.
So it's unclear if ignoring AVX would actually bring faster rendering in Corona. Would be good to test it somehow but not possible it seems.
talcikdemovicova.com  Website and blog
be.net/jurajtalcik   Our studio Behance portfolio
Instagram   Our studio Instagram, managed by Veronika

2017-12-20, 19:58:30
Reply #84

Vuk

  • Active Users
  • **
  • Posts: 109
    • View Profile
Just tested 2 other Xeons that are oem and qs. Oem works exactly at same frequency in both CB and Corona benchmark. QS is overclocked about 100 mhz and only shows in CB while in Corona is 100mhz less. Apparently the biggest hit is taken by ES cpu's. Don't know how its not a negative thing since it is 400mhz less per core on 20 cores so its also slower rendering time. Thnx for the answer Juraj but I dindn't really understand if this can be bypassed in bios or not?

2017-12-20, 20:04:18
Reply #85

Juraj

  • Moderator
  • Active Users
  • ***
  • Posts: 4446
    • View Profile
    • studio website
Were the QS/OEM also v4 generation ?

The one thing that ES has trouble is that it doesn't have the adaptive AVX turbo that QS/OEM/Retail has, which is under-clocking only the cores that contain AVX instruction. ES will always underclock the most.

It can't be adjusted in bios, the only bypass is by Corona to restrict the AVX. But Ondra said recently he doesn't know how.
talcikdemovicova.com  Website and blog
be.net/jurajtalcik   Our studio Behance portfolio
Instagram   Our studio Instagram, managed by Veronika

2017-12-20, 20:45:13
Reply #86

Vuk

  • Active Users
  • **
  • Posts: 109
    • View Profile
Thnx for the quick answer Juraj. No qs is v3 while oem is v2. Seems that vray ignores AVX instrucions and works at same clock speeds as CB.

2017-12-20, 21:58:20
Reply #87

Juraj

  • Moderator
  • Active Users
  • ***
  • Posts: 4446
    • View Profile
    • studio website
Vray also isn't based fully on embree, although I am not sure if the Embree support isn't on by default there too.

It's only V3/V4 that have the AVX-Turbo concept and mainly V4 where it's most prominent.

By perfomance comparison I meant that I compared the ratios of my v2 and v4 xeons in Cinebench and CoronaBench. The fact that my 2698v4 ES underclocked didn't skew this ratio negatively, which means that the AVX instruction did potentially offset performance in positive way.
This is actually the argument that Intel keeps making. That despite lower clock while using AVX instruction, the actual net performance is higher.

This doesn't take into consideration that ES have worse turbos in general so the net performance is probably worse, but not as worse as pure frequency difference would hint at. The end takeaway is that frequency simply isn't everything when it comes to all the modern instruction sets.
talcikdemovicova.com  Website and blog
be.net/jurajtalcik   Our studio Behance portfolio
Instagram   Our studio Instagram, managed by Veronika

2018-01-12, 04:10:47
Reply #88

Zray

  • Active Users
  • **
  • Posts: 43
    • View Profile
    • 3darcspace studio
I have 2 computers: dual 2683v3 and Giga X99 + 2696v3. The dual xeon computer seems that it doesn't have any problem with turbo boost speed (2.5 when rendering - full load). However, the 2696v3 turbo boost is just 2.6 (actually, it should be 2.8 - same issue with some other people in this topic). That's really annoying and I suppose the rendering is a little bit slower.

2018-01-15, 17:36:37
Reply #89

Rimas

  • Active Users
  • **
  • Posts: 156
  • Please refer all complaints to a brick wall nearby
    • View Profile
I know this is a wee bit off-topic, but I'd much rather prefer to work with top-end i7 chips than Xeons... Had a 20-core Xeon machine - it was slow in single-threaded tasks (say....MAX viewport!) at 2.6GHz and got superseded by a heavily overclocked 8-core i7-5960X (from 3.2GHz all-core to 4.5GHz all-core, on water). cheaper, much more responsive and a pleasure to extract more power from by overclocking. You do lose out on being able to have hundreds of gigs of RAM and having more than one socket, but for a workstation I wouldn't ever buy a Xeon anymore...at least not with those pathetic clocks...
A morning of awkwardness is far better than a night of loneliness...