Discussion in 'Frontpage news' started by Hilbert Hagedoorn, Aug 13, 2018.
Hilbert, 'thanks for a lengthy review on both - kudos '
And how many people here do so much rendering that they'd need a dedicated system just for it? The article from Puget Systems said:
My 1950X rig is a dedicated computing system - I've never run a single game on it, nor do I ever intend to (not even Steam is installed on it). People who buy such chips typically have multiple systems, and they often times use them for specific purposes.
Of course, I could game on it if I really wanted to (I always game with a GPU cap so it probably won't be an issue), but that's not the purpose for which I bought this chip - and if I were to get a 2990WX I would use it the same way. The only other tasks I use my TR system for is watching videos and browsing, usually while I'm gaming on my gaming system (easier than alt-tabbing).
I personally would agree with this, though I think the 12-core models also make for a decent choice. The 8-core models are in a weird situation. If you really want the PCIe lanes, Epyc is better. If you don't care about an abundance of PCIe lanes, the AM4 models are better.
Quite a lot, actually. If lengthy heavy computations is a regular thing for your workload, I would say it's actually more wise to have a separate dedicated system.
On this forum there are not many who use PC only for rendering, I do use my PC like for rendering and Folding@home but I do game when time allows me which is rare but I do play games
On some forums where I'm member there almost everyone is rendering or modelling on their PC, there are small group which using their PC for animation which I don't do
Regarding the Puget Systems and their review on 2990WX I agree with them on animation in C4D where this CPU wouldn't perform as good as other Intel CPU but when you do more rendering then this 2990WX or 2950X would do much better job
Its depends in most cases on workflow and what are you do
I just wish there is X399 board with 6-7 PCI_E slots
There are lots of people who use their PC for rendering, just not on this forum or OCN
Hope this helps
Shame AMD don't support dynamic overclocking.
The time-scale in the Vray chart is confusing or plotted incorrectly. The title states "minutes:seconds" but for example 1:55 minutes:seconds is obviously NOT the same as 1.55 minutes like it's plotted. 1:55 minutes is very near 2:02, not halfway 1:00 and 2:00.
Checking other benchs which I usually don't care about (like Corona, or Vegas PRO), it seems they're also badly plotted !! Any chart that has results in time is WRONG! You didn't convert minutes:seconds to minutes with decimals, it's blatantly obvious when the results are near X minutes with 50-something seconds!!!
I hope HH notices this...
Good catch. @Hilbert Hagedoorn : This causes visual grouping of results between x.00 and x.59 as there are no results between x.60 and x.99.
A Quick Look At The Windows Server vs. Linux Performance On The Threadripper 2990WX:
So now we can see that windows is the problem(Pro/Server), AMD must fix this issues with Microsoft so the HEDT platform can really show it's strength.
With all compilation related tests. One simply has to go to Windows Defender settings and exclude compiler and all project files/folders which are being used.
Negative effect Windows Defender has on compiling performance is crazy. And it has likely similar effect to some other workloads.
Threadripper 2990WX works better with Radeons
"Because like all other testers, we used a Geforce card, because models like the
GTX 1080 Ti simply count faster than all pixel accelerators from AMD and we want
to avoid a graphics limit as far as possible in a CPU test. However, a note from
OC_Burner in the 3DC forum has made us suspicious: According to him, the performance
of the Nvidia driver breaks down massively when CPUs with 64 threads under
Windows 10 are used.
So we grabbed a Radeon RX Vega 64 and measured it - with positive results."
So, nVidia is offloading something to CPU. Massively enough for it to spawn 64 threads if available?
And then their procedure for handling data of those threads has pretty negative scaling?
Do they still have software geometry discard of something?
Or is it actually usual suspect called "Windows 10"?
This is fantastic news for me. As I wanna to buy 2950x and I have a Vega 56
HH if you could do some additional tests with AMD card and Threadrippers it would be awesome
The 2950x is the best choice here imo , two 8 core chips with no restrictions like sharing the memory , otherwise the 2990wx would be the better choice . Well done AMD what a value and visionary approach again .
Btw I cannot help but noticing that loads of people don't understand that u cannot really judge performance based on tests that are run on badly written software that is way behind the hardware . Multicores are the future, the performance of these chips will only improve as the software will start catching up .
Your encoding/transcoding to x.264 test results would be much different if you would disable SMT (hyper threading) and set the threads number around 90 using x264 parameter (threads=90) option. Give a try and see the real result.
How many people use single thread for rendering purpose? Is Puget Systems high or what? Rendering requires massive parallel processing and as "a single socket processor" 2990WX is doing a fantastic job comparing to Intel's 8180 (10K USD list price). Single thread performance is not important unless you use 2990WX for gaming purpose which I believe it's a insult for this CPU.
thanks Hilbert very interesting and detailed review I especially appreciated two things :
1) the part about coverage of the smaller AIO watercooling blocks, since even today there are still not many specific TR4 solutions sold/reviewed and in my country almost and zero available (unlike the cpu,motherboard and ram)
2) the power consumption part where suddenly you realize that those "280w rated" giant air towers may not be enough
Really depends where are you located, but only AIO which does have full coverage cold plate is Liqtech from Enermax and Coolermaster I think too does have full coverage cold plate, but most of known manufacturers like Corsair or NZXT doesn't have full coverage cold plate on their AIO
Enermax Liqtech TR4 is not bad, just their reliability is not the best and failure rate of their AIO is pretty much high
Good air cooler can be better than AIO on TR4 just from reliability point of view and usually they're cheaper as well, I would recommend have look on this review of air coolers on ThreadRipper
But for best cooling you want put that CPU under water with good water block like is XSPC, Heatkiller or Bykski
Hope this helps
8 days remaining for TR 2920x.
I can't wait to see what it is capable of. Easy 9900k overall killer in its price range.