My impressions of the 2990WX after building and using it.

[Markfw]

OK< 4 dimms. Taichi will do 3600 on 3600 memory at 1.35, but not stable, 3466 is stable. MSI will not do 3466 to even boot, and you have to remove the video card, and then the battery, and reset everything to even try another speed. I hate MSI, I only got this for the 19 phase power. Eventually, I may get 3200 to work.

stay tuned

 

[mattiasnyc]

@mattiasnyc, thanks for the additional info.

Re: unrecoverable situations observed with Ryzen and Threadripper: It's certainly not the CPU that gets stuck; rather the software stack doesn't recover properly from the under-/overrun. As you indicated, there is a big matrix of parameters (audio device driver, buffer size : sample rate, channel count, plugin count), whose combinations just cannot be entirely tested by the developers of the software stack's x-run recovery routines.

However, this test has been done as I mentioned for over a decade and this is atypical behavior. The typical behavior of the software(s) is that they function "fine" and it is the audio stream that is interrupted. In other words processing is "fine" but late and the buffer empties and we get these glitches. That's the normal error. CPU usage is anywhere from 80-100% when this happens depending on the CPU in question.

So for the software to just hang on this CPU seems like an anomaly, and the parameter that has changed is the CPU (as far as I know).

I asked for buffer sizes earlier because it makes a difference if live manipulation of an audio stream is the goal (with humanly imperceptible delay between live recorded input and live played back output - which needs very small buffers, such that system latencies can become an issue, e.g. from badly behaving drivers, regardless of CPU power) --- or if merely manipulation of a prerecorded stream is the goal (such that moderately large audio buffers can be set).

Yes, I know. That is why the test is run at those varying buffer sizes.

All in all, I believe that this sort of testing requires a basic experience with the setup of an audio workstation, otherwise you are bound to get confusing results. At the very least, the rather large parameter space needs to be narrowed down. Edit: And the software better detects and reports x-runs itself, because having to rely on the user to perceive x-runs as clicks or pops is not really objective testing.

The clips/pops/crackling are pretty obvious though, in addition to which one interesting aspect is if this benchmark would actually hang at all on Mark's system. If it doesn't then it doesn't matter whether or not there are clicks because it is then established that the other test system may have had issues specific to the system itself, not the CPU.

Either way Mark can let me know if he want to give it a whirl or not. It may be more work than he would like to spend time on.

 

[dnavas]

OK< 4 dimms. Taichi will do 3600 on 3600 memory at 1.35, but not stable, 3466 is stable.

This is CL15? [I thought I saw a link to the CL15 memory, but your sig only speaks to CL16.]

 

[Markfw]

This is CL15? [I thought I saw a link to the CL15 memory, but your sig only speaks to CL16.]

They are cl15. Thats a typo, all the regular threadrippers are cl14.

At 3200, the memory is down to 67.8

from
8 x dimms@3066 at 69.9

Here is the pic:

 

[DrMrLordX]

OK< 4 dimms. Taichi will do 3600 on 3600 memory at 1.35, but not stable, 3466 is stable. MSI will not do 3466 to even boot, and you have to remove the video card, and then the battery, and reset everything to even try another speed. I hate MSI, I only got this for the 19 phase power. Eventually, I may get 3200 to work.

stay tuned

Damn, those results on the MSI board are troubling.

To get your RAM stable @ DDR4-3600, you may need to push 1.4v vDIMM and some extra vSoC. Unless you're blowing air directly onto your DIMMs, I would not go past 1.45v . That's my usual limit anyway.

Still, the latency results you're showing at just DDR4-3200 are pretty damn good.

 

[Markfw]

Damn, those results on the MSI board are troubling.

To get your RAM stable @ DDR4-3600, you may need to push 1.4v vDIMM and some extra vSoC. Unless you're blowing air directly onto your DIMMs, I would not go past 1.45v . That's my usual limit anyway.

Still, the latency results you're showing at just DDR4-3200 are pretty damn good.

Well, I am blowing air directly on the dimms, but on the Taichi, X399, it does 3466@1.35 v which its rated for..MSI just blows. I will wait for a bios update or live with it.

 

[Markfw]

OK, after today, I have to comment (in case some of you noticed some things)

First, this sucker will do 4 ghz@3200 memory all day every day at 100%load.

BUT ......

It runs HOT and sucks a lot of juice (like 450 watts just for the CPU) @ 4 ghz

SOOOOOO, I run it at 3.7 ghz @ 100% load all day every day doing DC cancer research, and its a lot cooler, and takes like 100 watts less. 1.3375 (or something like that) versus 1.15 vcore. and temps go from 83c down to 65c, and power usage from 450 to 350 or less.

I spent too much on this to kill it, and I need a real workhorse, not to just kill it, but it WILL do that it if I need it to.

Edit: One last comment. Since this is almost an EPYC, but only 4 memory channels, I can see why they are so good. They run 3 ghz on almost no vcore, and run VERY cool. THATS what is king in the data center. I was running 3550 at 1.04 vcore on air@65c and 250 watts, think what it could do at 3 ghz ? 0.8 ? (just a guess)

 

[DrMrLordX]

The input voltage on those chips is pretty low. I'm thinking you could probably handle stock speeds at around .9v? Just a guess.

Also if you are having problems with heat, you may need a bigger rad and/or higher coolant flow. The local WC experts can probably help you more than can I in that department.

 

[lightmanek]

As a new Threadripper 1920X owner I too was surprised how low it can run on vCore. Mine is stable at 1.15V at 3.9GHz and I tried XMR mining with 1V at 3.4GHz no problem. Haven't had time to find the true limits, but it rocks already compared to my 2700X!

Thanks Mark for all your work! I follow this thread very closely

 

[aigomorla]

I hate MSI, I only got this for the 19 phase power.

I told you MSI is garbage.
They used to be good, back when the Z97 and X79's were out.
After that each successive generation there boards and quality has turned into garbage.
They are more geared at bling then being solid stable.

Its not just motherboards either.
I would rank them just as bad as Zotac in Video cards as they dont know how to make a proper heat sink.

You were better off on your TaiChi even if it has lower vregs.
I bet you the quality of the VRM's on the Taichi are even better although fewer.

 

[Markfw]

I told you MSI is garbage.
They used to be good, back when the Z97 and X79's were out.
After that each successive generation there boards and quality has turned into garbage.
They are more geared at bling then being solid stable.

Its not just motherboards either.
I would rank them just as bad as Zotac in Video cards as they dont know how to make a proper heat sink.

You were better off on your TaiChi even if it has lower vregs.
I bet you the quality of the VRM's on the Taichi are even better although fewer.

Well, I have a brand new Taichi. I was not going to return it, and use it for a 2950x (when I pay off the $6000 2990WX system), but who knows, maybe I will use it for the 2990WX. I can only get 3200 out of the memory, instead of 3466 on Taichi (same memory, 3600 cl15)

And yes, I knew it was probably crap when I ordered it, but I thought I really needed the 19 phase power.

 

[Jerky_san]

Hey Markfw I am a long time lerker and read your whole thread on the 1950x build and all of this one so far as well.Signed up today to ask you if you've attempted undervolting the 2990wx on stock speeds to see how low you could go before instability hit? I haven't seen where you mention it so just asking. Thanks a lot for doing all your research and I hope you get cancer free soon. Never give up the fight!

 

[Markfw]

Hey Markfw I am a long time lerker and read your whole thread on the 1950x build and all of this one so far as well.Signed up today to ask you if you've attempted undervolting the 2990wx on stock speeds to see how low you could go before instability hit? I haven't seen where you mention it so just asking. Thanks a lot for doing all your research and I hope you get cancer free soon. Never give up the fight!

Since I want to run as fast as I can, as low vcore as I can (for a decent speed), the answer is no.. But I think 3550 on 1.04 vcore (stock) is pretty darn good ! So I took it up one small notch, and happy with this.

 

[Jerky_san]

Ok thanks for the reply.