Threadripper BUILDERS thread

[Fir]

HEDT is essentially the bridge between gaming systems and workstations.
In the past I've used lots of gaming boards like the EVGA Classified and Asus ROG/Rampage boards to build formidable workstations that were very powerful and considerably cheaper than similarly configured Dell Precision workstations. The main thing they lacked is ECC RAM. Some software for CAM/CAD requires the use of a "blessed" system and drivers to be properly supported by the software vendor but in these cases the expense of a certified workstation is completely acceptable.

$1550 for a sixteen core CPU and board is quite laughably cheap when you think about it. But it wasn't exactly what I consider turnkey out of the box either. I've had mine since the first week of August and still it seems there are growing pains associated with this new platform. That is to be expected.

X99 wasn't exactly a saint either! Had lots of issues and blown processors/sockets. The more recent revision of X99 seemed to fix some of the cold boot issues and such but I'm glad I didn't buy a lot of them.

I do want to build a dual socket EPYC with dual 32 core chips though. 128 threads, the task manager display will look like a hard disk defrag screen with 128 logical processors!

 

[StefanR5R]

I can't user the Noctura air cooler for TR4, as its too big with my 2 video cards (that can't be moved)

Noctua's own mainboard compatibility list for NH-U14S TR4-SP3 mentions only the Asus Zenith as having the 1st PCIe slot too close. All other motherboards should allow you to populate the 1st slot. On the other hand, I am not up to date whether typical GPU back plates still fit.

 

[MarkPost]

OK< so I can't user the Noctura air cooler for TR4, as its too big with my 2 video cards (that can't be moved), so I got this:

https://www.newegg.com/Product/Product.aspx?Item=N82E16835214080

I heard its the only thing NOT custom water that beats the Noctura. I will let you know in a week or so....
It will be replacing a Arctic cooling AIO 360 rad with 6 fans...

That Enermax is a very good choice, I've read very good reports from users. They state about -10ºc in comparison with the not proper TR AIO.

Let us know if this turns true for you too!

 

[MarkPost]

Noctua's own mainboard compatibility list for NH-U14S TR4-SP3 mentions only the Asus Zenith as having the 1st PCIe slot too close. All other motherboards should allow you to populate the 1st slot. On the other hand, I am not up to date whether typical GPU back plates still fit.

yeah I think so too. Perhaps may be needed to applying the offset option, but it should be possible with any TR MB except ASUS ones.

 

[MarkPost]

I would be curious how it compares and if it would be worth getting over the H115i I'm using now. I know if I push 4GHz I hit throttling which is no good.

https://linustechtips.com/main/topi...ch-tr4-360-threadripper-full-coverage-cooler/

It seems there is a huge improvement

 

[Markfw]

OK, so I got the Enermax 360 for TR4. I was runnign 63 something with the old 360 rad and 6 fans in push/pull. The new one with only 3 fans is only running 1c cooler, 62 something, but I am also only at 1.25 vcore/3.8 ghz. When I get the time, it will be interesting to see what happens when I go to 4 ghz at (not sure) vcore.

 

[MarkPost]

for apples to apples comparison, you should test the same 6 fans installed in the Enermax and with similar rpms.

Did you spread the thermal paste across the whole cpu surface? it seems it gives the best result

 

[ddogg]

Noctua's own mainboard compatibility list for NH-U14S TR4-SP3 mentions only the Asus Zenith as having the 1st PCIe slot too close. All other motherboards should allow you to populate the 1st slot. On the other hand, I am not up to date whether typical GPU back plates still fit.

I have an ASrock Taichi and I am unable to fit my 1080Ti in the first slot due to the backplate. Wasn't a huge deal as I have only one card but had I gone with dual GPUs I would have had to remove the backplate.

 

[Markfw]

for apples to apples comparison, you should test the same 6 fans installed in the Enermax and with similar rpms.

Did you spread the thermal paste across the whole cpu surface? it seems it gives the best result

Yes, the WHOLE surface. The problem with the 6 fans is that I hate them sticking out above my case (pictures way up in this thread) it looks like crap.

 

[Markfw]

I have an ASrock Taichi and I am unable to fit my 1080Ti in the first slot due to the backplate. Wasn't a huge deal as I have only one card but had I gone with dual GPUs I would have had to remove the backplate.

?? I have the Taichi, and I have a 1080TI FTW3, with a backplate, and it works fine......

 

[ddogg]

?? I have the Taichi, and I have a 1080TI FTW3, with a backplate, and it works fine......

Is your backplate raised? I have a custom backplate that is slightly raised so that might be the issue. Either way didn't really bother me

 

[Markfw]

OK, 4 ghz testing. 1.375 vcore might be stable@77.5c (so far)

Edit: for this@100% load and 2 1080TI's@100%, now drawing 920 watts from the wall.

I may go back to 3,8... 130 watts more for 200 mhz ???

 

[Markfw]

Is your backplate raised? I have a custom backplate that is slightly raised so that might be the issue. Either way didn't really bother me

Well, here is a pic with the new Enermax 360 for TR4 currently 80c@ 4 ghz 1.375 vcore, and yes, it looks like the backplate is raised a hair.

 

[Markfw]

WOW, I just rebooted into stock, and with both 1080TI cards@100%, and the CPU @ 3.3 ghz (all cores) and vcore of 1.05 @ 100% load, it only draws 630 watts ! Probably 500 of that are the 2 1080's. So 3.3 to 4.0 takes 290 watts more ?

 

[ddogg]

OK, 4 ghz testing. 1.375 vcore might be stable@77.5c (so far)

Edit: for this@100% load and 2 1080TI's@100%, now drawing 920 watts from the wall.

I may go back to 3,8... 130 watts more for 200 mhz ???

You need 1.375V for 4? I thought you said you only required 1.2V in the other thread. I'm at 1.381V for 4.1.
Also, my backplate is raised a little more than yours.

 

[Markfw]

You need 1.375V for 4? I thought you said you only required 1.2V in the other thread. I'm at 1.381V for 4.1.
Also, my backplate is raised a little more than yours.

That was bios. I think I had llc on, and it might have really been higher.. Now using Ryzen master. And I have not done extensive testing, but I know it locked immediately @1.35

I paused the video cards. its was at 275 watts, just CPU@100% stock full load. So they take 350 watts basically.

 

[Charlie22911]

WOW, I just rebooted into stock, and with both 1080TI cards@100%, and the CPU @ 3.3 ghz (all cores) and vcore of 1.05 @ 100% load, it only draws 630 watts ! Probably 500 of that are the 2 1080's. So 3.3 to 4.0 takes 290 watts more ?

You experience shooting for 4GHz is strikingly similar to what I experienced trying for 4.3GHz on Broadwell-E, voltage and power draw run wild compared to even 100mhz lower.
Are your 1080tis SC2? If so HWinfo32 will show power in watts for those cards.

I haven’t really messed with my Threadripper build yet, if I did it would probably be RAM for infinity fabric performance. Need to research how Threadripper behaves with 8 sticks of ram...

 

[MarkPost]

Yes, the WHOLE surface. The problem with the 6 fans is that I hate them sticking out above my case (pictures way up in this thread) it looks like crap.

haha yeah I know, Its just to know perf dif between both AIO with the same fan config

 

[Markfw]

New idea, and experiment. I want the highest OC with the best balance of power. Right now@3.6@1.1 vcore. Will try to up the speed, I think this is a nice voltage.

And the draw with everything@100% is only 700 watts, vs 3.3 ghz stock taking 660. Remember that with the video cards@100%

 

[Markfw]

You experience shooting for 4GHz is strikingly similar to what I experienced trying for 4.3GHz on Broadwell-E, voltage and power draw run wild compared to even 100mhz lower.
Are your 1080tis SC2? If so HWinfo32 will show power in watts for those cards.

I haven’t really messed with my Threadripper build yet, if I did it would probably be RAM for infinity fabric performance. Need to research how Threadripper behaves with 8 sticks of ram...

I am using hwinfo64, and I can't find that power usage in the output.

 

[Charlie22911]

I am using hwinfo64, and I can't find that power usage in the output.

I'm not sure what the difference between the SC2 and FTW3 is but it shows for me...


EDIT:

Ran a single pass in Heaven all settings maxed at 2560x1440 both stock and overclocked, screen shots below if that helps you with finding card power draw. I think your estimate is pretty spot on though.

Overclock:

Stock:

 

[Markfw]

I'm not sure what the difference between the SC2 and FTW3 is but it shows for me...


EDIT:

Ran a single pass in Heaven all settings maxed at 2560x1440 both stock and overclocked, screen shots below if that helps you with finding card power draw. I think your estimate is pretty spot on though.

Overclock:

Stock:

I found it. You have to click on the sensors button (power usage in the sensors ??) But it says 200-211 watts each, and CPU at 217 watts currently, with wall draw at 705 watts. So about 85 watts used by the nvme, motherboard, ram, fans and AIO pump ?

Currently at 1.1 vcore and 3650 mhz

 

[ddogg]

Been playing Shadow of war and noticed that gaming mode vs creators mode makes a HUGE difference in FPS in this game. In creators, my average FPS was around 75 and it jumped to 105 when I switched to gaming. It's the first game I've noticed such a massive difference in FPS.
I generally keep it on creators as I have a bunch of VMs running but may have to switch when playing this game.

 

[IEC]

I found it. You have to click on the sensors button (power usage in the sensors ??) But it says 200-211 watts each, and CPU at 217 watts currently, with wall draw at 705 watts. So about 85 watts used by the nvme, motherboard, ram, fans and AIO pump ?

Currently at 1.1 vcore and 3650 mhz

Wall draw also includes power supply efficiency losses. With Gold/Platinum rated PSUs you are still losing between 8%-13% as heat depending on the load and efficiency at that load.

 

[urvile]

Been playing Shadow of war and noticed that gaming mode vs creators mode makes a HUGE difference in FPS in this game. In creators, my average FPS was around 75 and it jumped to 105 when I switched to gaming. It's the first game I've noticed such a massive difference in FPS.
I generally keep it on creators as I have a bunch of VMs running but may have to switch when playing this game.

I found I get a big frame difference in both Far Cry 3 and Deus Ex MD. I have also found that virtual box does not like 12 cores/24 threads.