Threadripper BUILDERS thread

[dlerious]

You should have put this in the 2990wx builders thread, but there are only 2 of us that I know, and I run linux

It's a video from Wendell. He mentions something about Linux.

 

[Markfw]

It's a video from Wendell. He mentions something about Linux.

I watched it. Yes, the bottom line ? Linux has no problems with it, Windows DOES

 

[dlerious]

I watched it. Yes, the bottom line ? Linux has no problems with it, Windows DOES

Didn't mean to imply there was. I'm looking at running Linux on my 1950X, just haven't really had time to spend doing much with my system.

 

[dnavas]

I hope that was helpful! My 2950X arrives Tuesday.

Or not, given that Amazon has lost it. Much like they also lost four case fans and a wifi range extender just last week. :smh:
The Taichi board looks awesome, though!

 

[dnavas]

My 2950X is limping. I had to force Win10 install from UEFI for reasons I don't really understand, and then I had to repartition my drive because MBR vs GPT something or other. I'll have to try reinstalling linux at some point.

Good news: I ran prime, all cores ran above 4Ghz w/o me changing a thing. 85C ... 86C ... 88C -- I thought TR throttles at 85C? Hmm, well, with a good cooler, this is going to be amazing, because it was using less power to run 16 cores than my 1800X uses at 3.9G! Of course, I don't have all my hardware installed just yet, because....

Bad news: the taichi HATES my RAM. @Markfw I got those same 3600CL15s you pointed at, and when I boot into linux with them, 4 times out of 5 I lose my mouse and keyboard post login. I pulled the 3200CL16s from my 1800X (I only have a pair :sigh: ), and those work fine, unless I try to run them at 3200, and then all holy hell breaks loose. Reset button doesn't appear to clear the XMP settings, so I had to stick the 3600s in there, and reset the RAM that way. Yeesh -- that seems like a brick waiting to happen? Anyone have any timings I can try for the 3600s? I'm a little concerned that they didn't even work at 2133, never mind 3600 or 34xx.

 

[Markfw]

My 2950X is limping. I had to force Win10 install from UEFI for reasons I don't really understand, and then I had to repartition my drive because MBR vs GPT something or other. I'll have to try reinstalling linux at some point.

Good news: I ran prime, all cores ran above 4Ghz w/o me changing a thing. 85C ... 86C ... 88C -- I thought TR throttles at 85C? Hmm, well, with a good cooler, this is going to be amazing, because it was using less power to run 16 cores than my 1800X uses at 3.9G! Of course, I don't have all my hardware installed just yet, because....

Bad news: the taichi HATES my RAM. @Markfw I got those same 3600CL15s you pointed at, and when I boot into linux with them, 4 times out of 5 I lose my mouse and keyboard post login. I pulled the 3200CL16s from my 1800X (I only have a pair :sigh: ), and those work fine, unless I try to run them at 3200, and then all holy hell breaks loose. Reset button doesn't appear to clear the XMP settings, so I had to stick the 3600s in there, and reset the RAM that way. Yeesh -- that seems like a brick waiting to happen? Anyone have any timings I can try for the 3600s? I'm a little concerned that they didn't even work at 2133, never mind 3600 or 34xx.

Run the 3600cl15 at 3466 or 3200. I can't get a full 3600 from them. Make sure the bios is set correctly for 1.35 dram voltage.

Edit: I just saw your later stuff, so not sure, sounds like a dram voltage problem though,

 

[dnavas]

Run the 3600cl15 at 3466 or 3200. I can't get a full 3600 from them. Make sure the bios is set correctly for 1.35 dram voltage.

Edit: I just saw your later stuff, so not sure, sounds like a dram voltage problem though,

Voltage was set in XMP (maybe it didn't like running at stock voltage and speeds?!), but it's stable through AIDA @3200, so it looks like I'm just going to have to mess around a little. Thanks!

edit: and I just put in the second pair, and 3200 is a no go, but ... 3466 works again? What? :sigh: I'm going to have to get that linux install working again. Maybe it was just linux causing trouble.

edit2: max auto-boost all core is clearly tied to memory speed and count. Using Prime95:
2 sticks at 3200: 4.1G (I was thinking "wow, golden!")
4 sticks at 3466: 3.7G
4 sticks at 2933: 3.9G
4 sticks at 2400: 4.0G

...which is interesting, and not something I'd read before. I've read that clocking memory higher can boost apps, but not that downclocking memory can raise boost. Actually, it's probably a power draw limit -- I'm seeing a fair amount more power drawn with four sticks at 2400 than 2 sticks at 3200.

Something to keep in mind if you're running apps that aren't memory sensitive, anyway.

 

[Markfw]

Voltage was set in XMP (maybe it didn't like running at stock voltage and speeds?!), but it's stable through AIDA @3200, so it looks like I'm just going to have to mess around a little. Thanks!

edit: and I just put in the second pair, and 3200 is a no go, but ... 3466 works again? What? :sigh: I'm going to have to get that linux install working again. Maybe it was just linux causing trouble.

Well, I have 4 threadrippers, and 2 or 3 or them have that 3600 cl15 memory and all 4 are Taichi motherboard. (all 4 are also running dual-boot with linux mint), and all run at a minimum of 3200.

 

[dnavas]

Well, I have 4 threadrippers, and 2 or 3 or them have that 3600 cl15 memory and all 4 are Taichi motherboard. (all 4 are also running dual-boot with linux mint), and all run at a minimum of 3200.

Yeah, I'm using ubuntu. It was definitely not happy. Issue was addressed be editing /etc/gdm3/custom.conf and uncommenting out the WaylandEnable=false line.

System idle in Windows is nice -- slightly less than my 1800X system, which is better than I had expected. Idle on linux is a lot hotter (97W in linux vs 69W in Windows).

The BIOS fan PWM on the Taichi is difficult to use. "Monitor CPU" ... doesn't. I have no idea what that is monitoring, but it isn't CPU temp. Would love to be able to monitor VRM temp :|

Any kind of load IMMEDIATELY causes the CPU temp to sky-rocket - I see occasional 10C second-to-second jumps in CPU temps -- much faster than fan spinup can compensate for. I'm concerned I may not have a great mate on my Noctua cooler. Or maybe it's just insufficient. :|

Enabled PBO, and if I could keep the CPU from throttling, I can get 4G from all cores w/3466G RAM. System pulls nearly 400W under CPU-only load, which is 150W higher than without PBO. Throttling pulls that down to 350W, which is still far higher than the 250W the system peaks at w/o PBO.

If I can get a decent cooler, this little guy is going to rock. And cheap, too. Remember those $3k IBM PCs back in the 80s? That's $6k in today's money. This stuff is great.

 

[DrMrLordX]

@dnavas if you can get your RAM stable @ 3600 using 1.42v - 1.45v vDIMM, would you be willing to run some intercore latency tests via SiSoft Sandra for me? Oh yeah, don't forget, you may need to push 1.1v SoC voltage or more too. Not sure what the UEFI looks like though.

Also you will probably need a beefier cooler, that 2950X is going to (overall) chew up a lot more power than your 1800x, and it will (probably) have more hotspot issues. I would think watercooling would be a minimum. It would be nice if I were wrong though. Regardless you may just want to run the fans balls-out until you find a better solution.

 

[Atari2600]

Looks like the Enermax has bitten me too. Have a 280mm unit.

One of the fans is rattling, but still turning.

But the CPU clocks have been dropping (1900X) with temperatures pretty constant - but I never really paid heed.

Just put in a new 2950X, and its running all cores ~ 3 GHz ~ 68 degC.


Correspondence started with reseller. Lets see what they come up with.

Has anyone got the new Liqtech II model? Does it resolve the issues?

 

[Markfw]

Looks like the Enermax has bitten me too. Have a 280mm unit.

One of the fans is rattling, but still turning.

But the CPU clocks have been dropping (1900X) with temperatures pretty constant - but I never really paid heed.

Just put in a new 2950X, and its running all cores ~ 3 GHz ~ 68 degC.


Correspondence started with reseller. Lets see what they come up with.

Has anyone got the new Liqtech II model? Does it resolve the issues?

I have my 3rd going bad. One RMA'ed with the old one shipped, the second is on process, but a friend just got his RMA, and it was the new one.

 

[Atari2600]

I have my 3rd going bad. One RMA'ed with the old one shipped, the second is on process, but a friend just got his RMA, and it was the new one.

I've specifically mentioned the new one when I've made my enquiry, so hopefully that leads them to that being a satisfactory conclusion!

 

[Paratus]

I have my 3rd going bad. One RMA'ed with the old one shipped, the second is on process, but a friend just got his RMA, and it was the new one.

I've specifically mentioned the new one when I've made my enquiry, so hopefully that leads them to that being a satisfactory conclusion!

How do I know if I have a new one. (My RMA showed up a couple of weeks ago and I haven’t had time to install)

 

[Markfw]

How do I know if I have a new one. (My RMA showed up a couple of weeks ago and I haven’t had time to install)

They are Liquitech II

 

[Paratus]

They are Liquitech II

Crap. Liqtech not 2. Ah well better start using it to failure so I can get another RMA.

 

[Markfw]

Crap. Liqtech not 2. Ah well better start using it to failure so I can get another RMA.

The box should look like this : https://www.newegg.com/Product/Product.aspx?Item=N82E16835214089
(in the last picture)

 

[Paratus]

The box should look like this : https://www.newegg.com/Product/Product.aspx?Item=N82E16835214089
(in the last picture)

Thanks, but the new one they sent me is definitely the original liqtech.

Sigh.

 

[dnavas]

@dnavas if you can get your RAM stable @ 3600 using 1.42v - 1.45v vDIMM, would you be willing to run some intercore latency tests via SiSoft Sandra for me? Oh yeah, don't forget, you may need to push 1.1v SoC voltage or more too. Not sure what the UEFI looks like though.

Also you will probably need a beefier cooler, that 2950X is going to (overall) chew up a lot more power than your 1800x, and it will (probably) have more hotspot issues. I would think watercooling would be a minimum. It would be nice if I were wrong though. Regardless you may just want to run the fans balls-out until you find a better solution.

I'm going to back off from PBO until I get water-cooling. Planning to buy a liqtech ii 280 and stuff it in the front. I see a bunch of people unhappy with rev1, and I'm really hoping they've fixed things with rev2. I also want to measure and make sure things are going to fit!

Point me at a download page for what you want run and I'll give it a whirl. Might be awhile, though -- I need to prioritize getting the machine back up so the wife isn't left without working software. Ping me in a week if I haven't responded. Good point on the SoC voltage -- I'll try that if things get squirrely.

 

[jmmec]

Regarding the Enermax Liqtech TR4 cooler:

Although there are some reports of problems online (even in this forum), I don't overclock my 1950X, so although the cooler wasn't showing any signs of failures I swapped out the original Enermax Liqtech 240 TR4 (purchased Nov 2017) with their new Liqtech 360 TR4 v2 since that was all that NewEgg had in stock.

FWIW, I don't care about doing a lot of stress testing, so last weekend I ran a 20 minute Prime95 with small ffts with the original Enermax, and then with the new v2 version, with an ambient room temperature of around 17C:

• Original 240 Enermax: peaked at 55C (Tdie) at about 10 minutes into the run, and stayed at 55C throughout the rest of the 20 minute run @ 3.45 to 3.50GHz on all cores.
  
• New 360 Enermax v2: peaked at 50C (Tdie) around 10 minutes into the run, and stayed around 50C throughout the rest of the 20 minute run @ 3.40 to 3.5GHz on all cores.
  
• At idle the Tdie averaged < 23C for both coolers.
• Thermal Grizzly thermal paste was used for both coolers.

It would be nice if Enermax would explain what they know about the failed units and how large of a problem it is / was. I'll probably open up the original cooler in the next month or so to see what it looks like inside.

 

[Markfw]

Regarding the Enermax Liqtech TR4 cooler:

Although there are some reports of problems online (even in this forum), I don't overclock my 1950X, so although the cooler wasn't showing any signs of failures I swapped out the original Enermax Liqtech 240 TR4 (purchased Nov 2017) with their new Liqtech 360 TR4 v2 since that was all that NewEgg had in stock.

FWIW, I don't care about doing a lot of stress testing, so last weekend I ran a 20 minute Prime95 with small ffts with the original Enermax, and then with the new v2 version, with an ambient room temperature of around 17C:

• Original 240 Enermax: peaked at 55C (Tdie) at about 10 minutes into the run, and stayed at 55C throughout the rest of the 20 minute run @ 3.45 to 3.50GHz on all cores.
  
• New 360 Enermax v2: peaked at 50C (Tdie) around 10 minutes into the run, and stayed around 50C throughout the rest of the 20 minute run @ 3.40 to 3.5GHz on all cores.
  
• At idle the Tdie averaged < 23C for both coolers.
• Thermal Grizzly thermal paste was used for both coolers.

It would be nice if Enermax would explain what they know about the failed units and how large of a problem it is / was. I'll probably open up the original cooler in the next month or so to see what it looks like inside.

I have 4 of them. 3 have already failed within a year. I took apart the first one to see that is "gunked up" and hence the faile. I suspeact the rest have the same problem due to the metal reaction/

 

[dnavas]

Point me at a download page for what you want run and I'll give it a whirl. Might be awhile, though -- I need to prioritize getting the machine back up so the wife isn't left without working software. Ping me in a week if I haven't responded. Good point on the SoC voltage -- I'll try that if things get squirrely.

FWIW, I can boot with 3600, and although it dumped out on my first AIDA run, I was able to get a couple of clean runs after that:
3600 reportedly using 16-15-15-35 timings has a memory read of 85285MB/s and a latency of 80.3ns.
3466 (same timings) has a memory read of 81893 and a latency of 83ns.

I know that isn't core-core latencies, but you can see you're getting some benefit, though honestly not a whole lot. I should probably run in NUMA just for kicks, but ... later....

 

[eek2121]

Regarding the Enermax Liqtech TR4 cooler:

Although there are some reports of problems online (even in this forum), I don't overclock my 1950X, so although the cooler wasn't showing any signs of failures I swapped out the original Enermax Liqtech 240 TR4 (purchased Nov 2017) with their new Liqtech 360 TR4 v2 since that was all that NewEgg had in stock.

FWIW, I don't care about doing a lot of stress testing, so last weekend I ran a 20 minute Prime95 with small ffts with the original Enermax, and then with the new v2 version, with an ambient room temperature of around 17C:

• Original 240 Enermax: peaked at 55C (Tdie) at about 10 minutes into the run, and stayed at 55C throughout the rest of the 20 minute run @ 3.45 to 3.50GHz on all cores.
  
• New 360 Enermax v2: peaked at 50C (Tdie) around 10 minutes into the run, and stayed around 50C throughout the rest of the 20 minute run @ 3.40 to 3.5GHz on all cores.
  
• At idle the Tdie averaged < 23C for both coolers.
• Thermal Grizzly thermal paste was used for both coolers.

It would be nice if Enermax would explain what they know about the failed units and how large of a problem it is / was. I'll probably open up the original cooler in the next month or so to see what it looks like inside.

Since they didn't ask me to send mine back, I took the copper plate off. It was full of goop. I'm currently rocking the version 2 (360) and my machine is silent at stock, even under load. The headphones they sent me are great. They also sent an apology letter. IMO if this unit holds up, then they went above and beyond and they will keep me as a customer in the future.

 

[Atari2600]

edit: Making this general for anyone with Liqtech cooler:

Just looking to confirm your failure symptoms - you observed the CPU downclocks rather than exceeds the temperature settings within Ryzen Master*? (Which is 68degC)


*is this even adjustable? In work and can't remember.

 

[DrMrLordX]

Point me at a download page for what you want run and I'll give it a whirl. Might be awhile, though -- I need to prioritize getting the machine back up so the wife isn't left without working software. Ping me in a week if I haven't responded. Good point on the SoC voltage -- I'll try that if things get squirrely.

https://www.sisoftware.co.uk/download-lite/

Just get the lite version and run the processor multi-core efficiency benchmark. Run it in both best-case and worst-case mode.

FWIW, I can boot with 3600, and although it dumped out on my first AIDA run, I was able to get a couple of clean runs after that:
3600 reportedly using 16-15-15-35 timings has a memory read of 85285MB/s and a latency of 80.3ns.
3466 (same timings) has a memory read of 81893 and a latency of 83ns.

I know that isn't core-core latencies, but you can see you're getting some benefit, though honestly not a whole lot. I should probably run in NUMA just for kicks, but ... later....

I'm surprised your latency was that high.