Threadripper BUILDERS thread

[ddogg]

For me, there are something like 2 games that I've tried so far that crash. The rest do great. Threadripper is actually a beast at gaming if you tweak everything.

Agreed. I game alot on it and even with a lot of background tasks running, FPS don't drop at all. When gaming I'm usually streaming using OBS, have 2 VMs running and Plex transcoding. Really love the computational power of this chip.

 

[urvile]

For me, there are something like 2 games that I've tried so far that crash. The rest do great. Threadripper is actually a beast at gaming if you tweak everything.

I find that the dunia engine (far cry 3/cry 4) performs worse in creator mode. I was doing some testing today. It doesnt seem to make any difference with wolfenstein the new order.

I also noticed that the gpu is utilised less in creator mode regardless of game.

 

[krumme]

Posting this here, so as not to anger coffeelake people. Someone was bragging about an 8600 (k?) 6 core using 200 watt and only 66c. Well I have 16 cores and 32 threads@200 watts (about) @57c, and about 4 times the computing power. I think that threadripper is a very efficient CPU !

The key is also this processor was delidded.

 

[Gator762]

It's been a while... I got everything put together. I got the Thermaltake View 71 case which I like, it's well thought out for watercooling.

The really good news what I wasn't expecting is that it's running the mem at 3200 well with no CPU overclocking. I had Prime 95 running for hours with no errors. From my earlier posts, G.Skill support told me they didn't have 64 GB of mem that would run at 3200. They would only support up to 32GB. It's G.Skill Trident z 64GB kit (4x16GB), CL14. I just loaded up the XMP profile in the BIOS.

 

[Markfw]

The key is also this processor was delidded.

But seriously, he can get 6 5ghz cores@200 watts, and I can get 16c/32t@200 watts ? I think I have him blown away in efficiency.

 

[StefanR5R]

To us who own high core count CPUs, and have a real use for them, and pay the electric bill that comes with it, those low core count/ high frequency CPUs and their associated power draw appear hilarious or almost obscene.

But let's not be offended; such CPUs are simply not what we need and want. And vice versa, high performance (let alone high efficiency) is not a thing for owners of the desktop Coffeelake type of CPUs... yet. Maybe some of them will discover the wonders of high performance computing some day and see the light.

 

[eek2121]

To us who own high core count CPUs, and have a real use for them, and pay the electric bill that comes with it, those low core count/ high frequency CPUs and their associated power draw appear hilarious or almost obscene.

But let's not be offended; such CPUs are simply not what we need and want. And vice versa, high performance (let alone high efficiency) is not a thing for owners of the desktop Coffeelake type of CPUs... yet. Maybe some of them will discover the wonders of high performance computing some day and see the light.

Maybe for you, not for me. I live in the US and didn't notice a change in my electric bill. Then again, my machine is only fully loaded when encoding videos, and due to the TIME difference in encoding these videos, I expected nothing less. Those running F@H, mining XMR, or other such schemes that load all 16 cores 24/7 will most definitely see a power bill increase due to 4X as many cores as a quad core.

 

[eek2121]

I finally pulled the trigger on an Enermax Liqtech TR4 360. It will be here this week. My current AIO was not designed for threadripper (Thermaltake Riing 240). With upgraded fans I could load it to 3.8, but temps still got up to 74C. To stay below 68C I had to back things off to 3.7GHz, in which point, maximum temps were only 62C...very annoying because bumping things up by even 25 mhz required a voltage bump that pushed the CPU up into the 70s.

When I get the new cooler later this week, I will post my voltages/speeds used in Ryzen Master as a reference for others to compare. I encourage others to do the same. Also, this MSI board SUCKS for overclocking.

 

[Markfw]

I finally pulled the trigger on an Enermax Liqtech TR4 360. It will be here this week. My current AIO was not designed for threadripper (Thermaltake Riing 240). With upgraded fans I could load it to 3.8, but temps still got up to 74C. To stay below 68C I had to back things off to 3.7GHz, in which point, maximum temps were only 62C...very annoying because bumping things up by even 25 mhz required a voltage bump that pushed the CPU up into the 70s.

When I get the new cooler later this week, I will post my voltages/speeds used in Ryzen Master as a reference for others to compare. I encourage others to do the same. Also, this MSI board SUCKS for overclocking.

with that cooler, I am at 1.1vcore, and 3650. and 57c. I can do more easily, but the watts go up quickly from here. Running 24/7/365 I worry about that.

 

[krumme]

with that cooler, I am at 1.1vcore, and 3650. and 57c. I can do more easily, but the watts go up quickly from here. Running 24/7/365 I worry about that.

Those tuned tr you are running there is just mighty impressive. I wouldnt think binning could do so much.
You will have to wait for 7nm to get something that can just remotely compete with that efficiency. Its 14lpp at its finest imo. Well done tuning on top of it !

 

[Paratus]

I finally pulled the trigger on an Enermax Liqtech TR4 360. It will be here this week. My current AIO was not designed for threadripper (Thermaltake Riing 240). With upgraded fans I could load it to 3.8, but temps still got up to 74C. To stay below 68C I had to back things off to 3.7GHz, in which point, maximum temps were only 62C...very annoying because bumping things up by even 25 mhz required a voltage bump that pushed the CPU up into the 70s.

When I get the new cooler later this week, I will post my voltages/speeds used in Ryzen Master as a reference for others to compare. I encourage others to do the same. Also, this MSI board SUCKS for overclocking.

Anything in particular about OCing the MSI board or is just the temps get out of control quick?

(I'm looking at getting the MSI board)

 

[Charlie22911]

I finally pulled the trigger on an Enermax Liqtech TR4 360. It will be here this week. My current AIO was not designed for threadripper (Thermaltake Riing 240). With upgraded fans I could load it to 3.8, but temps still got up to 74C. To stay below 68C I had to back things off to 3.7GHz, in which point, maximum temps were only 62C...very annoying because bumping things up by even 25 mhz required a voltage bump that pushed the CPU up into the 70s.

When I get the new cooler later this week, I will post my voltages/speeds used in Ryzen Master as a reference for others to compare. I encourage others to do the same. Also, this MSI board SUCKS for overclocking.

I don’t think you will see all that much of a change, I haven’t really pushed my 1950x that much due to its role... but when overclocking it thermals and power behave strikingly similar to my 6900k; they even clock similarly. Under custom water I still get 70-80c on the i7 depending on workload with P95 at the high end.
That is the cost of shrinking feature sizes, regardless of chip vendor.


This threadripper chip is very efficient at stock though, I’m really looking forward to the next iteration of Zen. If AMD separates the fabric and dram clock domains I’ll be jumping on it quick.

 

[eek2121]

Anything in particular about OCing the MSI board or is just the temps get out of control quick?

(I'm looking at getting the MSI board)

It only allows you to change the multiplier and core voltage as far as I can tell.

 

[eek2121]

Those tuned tr you are running there is just mighty impressive. I wouldnt think binning could do so much.
You will have to wait for 7nm to get something that can just remotely compete with that efficiency. Its 14lpp at its finest imo. Well done tuning on top of it !

It depends on the chip. My sweet spot is 3.7GHz @ 1.125. 3.8 GHz Requires me to pump the voltage to 1.1875 and it results in a temp of 76.8C under Prime95 max heat settings. This is a Thermaltake Riing 240 in push/pull configuration with upgraded (Cosair SP120) fans.

I don’t think you will see all that much of a change, I haven’t really pushed my 1950x that much due to its role... but when overclocking it thermals and power behave strikingly similar to my 6900k; they even clock similarly. Under custom water I still get 70-80c on the i7 depending on workload with P95 at the high end.
That is the cost of shrinking feature sizes, regardless of chip vendor.

This threadripper chip is very efficient at stock though, I’m really looking forward to the next iteration of Zen. If AMD separates the fabric and dram clock domains I’ll be jumping on it quick.

Oh, I'll see a difference for sure. First, the Enermax has a more powerful pump. Second, I'm going from a 240mm radiator to a 360mm radiator. Third, my current cooler doesn't cover the entire die, and it has a slightly concave shape. The Enermax is flat and covers the entire die.

Before upgrading to the SP120s, I could run prime95 with 3.8 GHz at all, and 3.7GHz was rather warm. Just by using the SP120s, I am able to use 3.8GHz for day to day usage, because nothing except Prime95 and other stress tests has managed to generate that kind of heat.

EDIT #2: Remove list for now. There were a couple typos and I'm re-testing anyways.

Note that tuning the Vcore is important, since a higher vcore raises temps even if the CPU is stepping down the Vcore. For example, when I first tested 3.6GHz, I started with a VCore of 1.175 and temps were 6C higher than the eventual 1.075V result.

 

[eek2121]

I found Threadripper's thermal throttling point. 86-87C. After that point, cores start throttling to 550 MHz. HWInfo64 did NOT see this. However, HWMonitor shows it.

 

[The Stilt]

86-87°C tDie (excluding the offset) or tCTL (including the offset)?
If the latter one, then it is VRM induced throttling and not the CPU.

 

[eek2121]

86-87°C tDie (excluding the offset) or tCTL (including the offset)?
If the latter one, then it is VRM induced throttling and the CPU.

The measurement was including the offset. It very well could have been VRM throttling, I didn't think to check the VRM temps. My cooler was clearly not able to handle the OC that caused it and I did not want to leave it at such a high temp for that long.

 

[eek2121]

The measurement was including the offset. It very well could have been VRM throttling, I didn't think to check the VRM temps. My cooler was clearly not able to handle the OC that caused it and I did not want to leave it at such a high temp for that long.

You know, I went back to look for replies to the thread, and I noticed you got Tctl (which excludes the offset) and Tdie (which includes the offset) mixed up. Are you okay? I know you know better. Or maybe I'm reading your post wrong...in which case, my fault.

 

[Charlie22911]

Steve did some testing on air versus liquid, included in the test were two CLCs with oversized and traditionally sized cold plates.

It is a worthwhile read for anyone curious. I didn’t expect the difference in cold plate size to be as large as it was.

https://www.gamersnexus.net/guides/3089-threadripper-cooler-comparison-full-coverage-liquid-vs-air

 

[Markfw]

Steve did some testing on air versus liquid, included in the test were two CLCs with oversized and traditionally sized cold plates.

It is a worthwhile read for anyone curious. I didn’t expect the difference in cold plate size to be as large as it was.

https://www.gamersnexus.net/guides/3089-threadripper-cooler-comparison-full-coverage-liquid-vs-air

They didn't use the Enermax 360, just the 240, didn't even include it ! The 360 would have won hands down. (I have one)

 

[eek2121]

My Enermax 360 has arrived. What a beast! I'm going to attempt to install this monster in push/pull config using 3 corsair SP 120s. It's going to be a bit of a challenge. We'll see how it all works out.

 

[MarkPost]

My Enermax 360 has arrived. What a beast! I'm going to attempt to install this monster in push/pull config using 3 corsair SP 120s. It's going to be a bit of a challenge. We'll see how it all works out.

some pictures once installed will be welcome too

 

[Markfw]

some pictures once installed will be welcome too

How about this one ?

 

[eek2121]

How about this one ?

Mark, we both have the exact same GPU and Cooler now, what the heck?

Running Prime95. 4 GHz, 1.375V. Temps are 70C and rising VERY slowly. Will see how it goes. Need to doublecheck the fan profiles. If I have issues I'm going to re-apply the thermal paste, I may have used a bit too much. It's hard to judge on such a large die.

EDIT: This might take a bit. Turns out that 1.375 is well above what is needed for 4 GHz. It appeared to top out at 73 C @ 1.375, so I'm trying 1.35 right now.

EDIT #2: 1.2875V appears to be the lowest I can go maintaining stability, of course I still need to stress test for quite a bit longer.

 

[eek2121]

So, at 4 GHz and 1.3V, things appear to be stable after 20 minutes of testing. Max temp reached was 68.75C. Very nice! The Enermax is a beast though, it's 15.5 inches long. The only place I could mount it is the top of my case.