A Look At The Windows 10 vs. Linux Performance On AMD Threadripper 2990WX

[wolfyrion]

Just chipping in to point out that the 1950X temperatures are fine if properly cooled. I idle at 30-40°C with the fans at ~500 rpm, and max out at 60-70° under full load (mprime), with the fans at ~1500rpm. While that may sound like an obvious statement, it at least means that the heat spreader design in the CPU does not appear to be a bottleneck (unlike e.g. skylake CPUs). The 1950X just generates a lot of heat, and that heat needs to be dealt with adequately. I use a water cooled setup with a 360mm radiator to do the job.

What cooling system are you using?

[MadCatX]

Haha all these peeps having a blast because a day 1 windows driver is not 100% perfect for once and doesn't win ALL the tests on these cherry picked, mostly Linux oriented, mostly synthetic benchmarks is absolutely hilarious.

ImageMagick, Blender and video processing are very real workloads and the synthetic test displayed the same trends so I wouldn't call these tests crafted nor handpicked.

Suddenly Linux desktop doesn't suck ass anymore, right? I mean, whenever you can actually build that awesome stuff for these superior OSes (to this day, you have "build issues" with this established framework and mature OS, it's so impressive).

The Linux desktop/workstation/whatever you want to call it has not sucked for a long time. It all comes down to what your reference for suckiness is.

They'll release a new driver (or a new scheduler, as they did with Ryzen), who the hell cares. That *will* happen.

*If* and when that happens we can revisit the issue. At this points the 2990WX performance under Win10 is truly subpar the OS is to blame.

Alright, let's forget about premiere pro, let's use x264, which might be possibly the only util here with an ACTUAL use case. Oh wait, that ran best on Windows. Too bad again.

You mean apart from Blender for instance? You may also want to read the results again because Win10 actually lost in x264 too even though that used to the one benchmark where its performance was adequate.

[Tomin]

While true, the WIndows performance numbers were in the same ballpark, and the LInux numbers were more than double the performance. Anandtech measured a slight improvement but unless you are suggesting that Linux performance was neutered while Windows performance remained stagnant, i think the point is clear enough even with the version being different.

These is also the difference of memory speed. Michael tends to use 3200 MT/s DDR4 but Anandtech runs theirs at JEDEC speeds by default.

[MadCatX]

These is also the difference of memory speed. Michael tends to use 3200 MT/s DDR4 but Anandtech runs theirs at JEDEC speeds by default.

Which might have been relevant only if Michael's tests did not include Windows results. Since they do it's pretty safe to say that different memory bandwidth didn't influence the results very much.

[Particle]

I have a 1950x and my main problem are the temperatures , I cant find a proper TR4 fan to reduce the temperatures , current idle temperatures are between 70-80 degrees Celsius with a thermaltake water cooler.
I wonder if its worth to upgrade to 2990WX and if the temperature issues are fixed on this cpu. I cant find anywhere on this article what cooling unit is used and what were the temperatures on these benchmarks.

Something is definitely wrong here, and it should be a solvable problem. I suppose the first question to ask is if you are running with stock settings or if you have done something like overclock, manually set voltages, disabled dynamic clocking for performance, etc. My 1950X experience has been like haasn's. I idle around 35C. I'm using a Noctua NH-U14S which isn't particularly exotic.

Also worth noting is that all-in-one water coolers can have issues with losing coolant over time which can lead to high temps. I had to discard one earlier this year on another system for that very reason. I could get the system to crash within 10 seconds of running Prime95. I replaced it with an air cooler and all was good again like the water had been three years earlier when new.

[wolfyrion]

Something is definitely wrong here, and it should be a solvable problem. I suppose the first question to ask is if you are running with stock settings or if you have done something like overclock, manually set voltages, disabled dynamic clocking for performance, etc. My 1950X experience has been like haasn's. I idle around 35C. I'm using a Noctua NH-U14S which isn't particularly exotic.

Also worth noting is that all-in-one water coolers can have issues with losing coolant over time which can lead to high temps. I had to discard one earlier this year on another system for that very reason. I could get the system to crash within 10 seconds of running Prime95. I replaced it with an air cooler and all was good again like the water had been three years earlier when new.

cough * stupid me , I was monitoring the TCL Temperature instead of the TDIE ... (I had put wrong ALIAS)

Well TDIE temperature is 44 and TCL 71 , so I guess I am ok

[chithanh]

Unfortunately, Windows is the 800lb. gorilla that everyone is familiar with. So if new hardware is not working well on that platform, then it is always the fault of the hardware, not Windows.

Ok, but we are not talking about consumer systems here. We are talking about HEDT and Workstation market, and Linux is very much known there.
Also, Anandtech, ComputerBase, etc. have tested on Linux in the past. ComputerBase even had launch-day Linux tests for the AM4 Ryzens. Just for Threadripper it appears that all the other sites' Linux folks were on vacation or whatever.

I mean, imagine that you are a reviewer, and you see such numbers which obviously indicate a problem somewhere. Would it be too much to ask to repeat those tests on Linux (where applicable)? For me, this would be the obvious way of telling my readers:

• This is a problem in the application software (handbrake etc.)
• This is a problem in the Windows operating system (7-zip on 2990WX slower than on 2950X)
• This is a problem of the memory controller configuration in the 2990WX (certainly there are some instances of this)

[Tringi]

Would you mind trying some tests again with High Performance power profile activated?

The default Balanced power profile is horrible for all AMD CPUs I've had over the past decade, whether on Windows 7 or 10. The numbers in the test are similar to what I'm used to see, and I'm pretty confident they would be on par with Linux if run with High Performance power profile active. For some reason it takes Windows' scheduler on Balanced profile super long (100's of ms) to detect high CPU usage, raise the frequency and unpark necessary cores. And then it's too eager to park them again. It's not that bad on 1800X, as on older CPUs (have tested 5350, FX-8150, dual 6282SE).

[Weasel]

Unfortunately, Windows is the 800lb. gorilla that everyone is familiar with. So if new hardware is not working well on that platform, then it is always the fault of the hardware, not Windows.

That's why Windows is so insidious; it is so widespread that it can make or break a product. Many (most?) people will look at the benchmarks and see that this new CPU performs poorly in Windows, and even though it's the fault of Windows and not the hardware/AMD, they still won't buy it. The insidious part is that it's Windows that's broken, but the hardware that pays the price. At least for users that aren't willing to look past Windows.

Replace "Windows" with "most modern software" and your post is way more accurate.

(note that it doesn't include the Linux kernel, just FYI, that's a case of good software)

Windows is a shitpile of legacy cruft whose only saving grace is the software library.

And they're extremely correlated, but not like someone like you would know that.

Most software is garbage though but that applies to both Windows or Linux... it's just that, a % of a way larger amount of total software means a lot more software is bad on Windows. Windows 10 itself is pretty bad itself tho.

[Weasel]

Suddenly Linux desktop doesn't suck ass anymore, right?

This is workstation benchmarks, desktop is in your head. So try again.