Threadripper BUILDERS thread

[The Stilt]

Threadripper won't do too well in X265 at low resolutions (< 2160P) since there is no way you can saturate the cores. Generally at 1080P the scaling stops at 10 threads or so, due to the larger CTU size (64 pixel).

 

[Markfw]

I am backing down the OC until a better cooler is available. On the 4.1 ghz handbrake test, one core got to 97. sent down to 4.0 and it completed in 43 seconds (throttling ?) and 88c.
Going to try cinebench again

Well, I messed up, too much vcore.It was really over 1.4, I added the .2 offset. Now its at 4.0 and 1.3 vcore running 67-70c, and max of 77c

I am going to let it run at this today to see if its stable running seti and F@H, that 100%loaded CPU all cores and 2 GPU's 1080ti's), around 800 watts average from the wall

 

[Fir]

Download Realbench v 2.54 and run that.
For torture test select all ram and run for 8 hours.
Mine is stock and running now and CPU temp is around 58C.

This will also use your GPU so it will draw some serious watts if you have lots of GPU...

 

[ub4ty]

Had to design and 3D print a custom mounting bracket for TR, because there simply isn't any coolers (besides the Asetek ones) available.
Some of the cooler manufacturers I sent inquiries to hadn't even heard about AMD SP3 socket.

The bracket is for DeepCool Assassin II, which is probably the best air cooler for TR despite it wasn't even designed for it. Thanks to it's 60x45mm contact surface, it provides 100% coverage of the cores themselves and 78% coverage of the heatspreader.

Dat coverage !!!

Stilt... you madman !! Doing the proper engineering companies should have done !
Took a look at the cooler you reference.. Zero gaps on the heatpipes exiting off the contact plate :

So, the whole contact plate has transfer heat capability. This is very much unlike the Noctua that seem to have expanded the contact plate and done nothing to ensure the heatpipes contacted the expanded surface. The sheer number of pipes on that thing !!!

BTW, there seems to be a clearance issues w/ the Asus board and Noctua :

@4:51. Noctua's fin stack covers a whole PCIE lane. This issue is likely not present on :

I recall AMD giving these guys a nogo zone. I'm not sure how Asus managed to design a board with the 1st PCIE slot so ass close to the socket on an E-atx board where they have more room than everyone else. So yeah, Taichi looks more official by the day

 

[MarkPost]

Dat coverage !!!

Stilt... you madman !! Doing the proper engineering companies should have done !
Took a look at the cooler you reference.. Zero gaps on the heatpipes exiting off the contact plate :

So, the whole contact plate has transfer heat capability. This is very much unlike the Noctua that seem to have expanded the contact plate and done nothing to ensure the heatpipes contacted the expanded surface. The sheer number of pipes on that thing !!!

BTW, there seems to be a clearance issues w/ the Asus board and Noctua :

@4:51. Noctua's fin stack covers a whole PCIE lane. This issue is likely not present on :

I recall AMD giving these guys a nogo zone. I'm not sure how Asus managed to design a board with the 1st PCIE slot so ass close to the socket on an E-atx board where they have more room than everyone else. So yeah, Taichi looks more official by the day

good catch, that really sux
I've looked at Gigabyte and MSI boards and all of them have more clearance there (like the AsRock) than Asus Zenith. wtf asus?

 

[The Stilt]

I recall AMD giving these guys a nogo zone. I'm not sure how Asus managed to design a board with the 1st PCIE slot so ass close to the socket on an E-atx board where they have more room than everyone else. So yeah, Taichi looks more official by the day

Yes, the first PCI-E x16 slot is indeed a goner on Zenith Extreme with most of the larger tower coolers.
That was I price I was more than happy to pay, since there is another x16 slot and two x8 slots available.

 

[Campy]

Threadripper's Noctua air cooler made an appearance :

LOL, I love Steve but he really needs just a tad more polish on his video.. He just threw shit together on the board and didn't even give mention to the Noctua. It's not even listed on their site. I bumped into it by mistake. Steve ! Give the Noctua cooler a 5min video man... you're the only one w/ it. LMFAO

Haha, that build made me laugh Hope he calls it "The Medusa" or something... no redacted given ^^




No profanity allowed in the tech forums.


esquared
Anandtech Forum Director

 

[Fir]

Not sure about tower coolers with this platform. True, the plate is large enough to accommodate 8 heat pipes. But that heat has to go somewhere and somewhere is going to be a LARGE area of fins! A new breed of AIO will be in the works for these, I'm sure. And with a plate to raise it a few mm, those should work nicely with 2066. The 12 core and higher pieces will need that added TDP capacity especially when overclocked.

 

[Markfw]

OK, the latest is, that I don;t like the temps, even at 3900 and 1.3 vcore. I think its now stable, but 77c high ? and 75c most of the time ? I was in the upper 50's with the 3700 stock speed on all cores. I think I am going back to that. At least I know what this chip can do with better cooling. For 24/7 3900 is "ok" but not really worth it to me for the power and heat. 1.05 is the stock vcore. 1.3 goes up my like 50-100 watts. and like 20c delta in temps.

 

[formulav8]

TR has a +27* offset. Is your temp program taking that into the equation?

 

[Markfw]

TR has a +27* offset. Is your temp program taking that into the equation?

I think so. During the 4.1 testing I saw 97c AFTER taking away 27 and at 4.0, 89c, and the 3.9 was 77c. Stock 3700 is now at 57c and 1.05 average vcore (per cpuid) vs 1.3-1.41 vcore at the other frequencies. I have to say, I like the efficiency of the stock settings, and still pretty decent performance.

 

[MarkPost]

70-80ºc at full load is ok imo. I can live if its between 80-85º IF perfomance gained deserves it (normally it doesnt, because the gain is marginal), and if temps are beyond that, I definitely avoid them. You need to find the better balance between perf gained/temps/watts.

EDIT: What program do you use to monitor temps? I would recommend you HWINFO64 https://www.hwinfo.com/download.php You need to look at the tdie sensor (the +27ºc offset is subtracted already)

 

[Markfw]

70-80ºc at full load is ok imo. I can live if its between 80-85º IF perfomance gained deserves it (normally it doesnt, because the gain is marginal), and if temps are beyond that, I definitely avoid them. You need to find the better balance between perf gained/temps/watts.

EDIT: What program do you use to monitor temps? I would recommend you HWINFO64 https://www.hwinfo.com/download.php You need to look at the tdie sensor (the +27ºc offset is subtracted already)

HWMonitor and coretemp. Could also try the Ryzen utility.

3.9 was 77c but the power draw and the 20c diff in temps seemed too much. I will play later, kind of burned out after almost 2 days of testing.

 

[Reinvented]

I think so. During the 4.1 testing I saw 97c AFTER taking away 27 and at 4.0, 89c, and the 3.9 was 77c. Stock 3700 is now at 57c and 1.05 average vcore (per cpuid) vs 1.3-1.41 vcore at the other frequencies. I have to say, I like the efficiency of the stock settings, and still pretty decent performance.

What does core temp say? It should give you the correct temps with offset taken into account.

--Nevermind. Saw you say you used CT. Temps are really high, even for a HEDT system imo. But then again, you are stressing it pretty hard. Is 3.7 with or without XFR?

 

[Markfw]

Those are all from coretemp. I was wrong about hwmonitor. Getting dizzy with all this testing.

 

[MarkPost]

Dont know if actual coretemp and hwmonitor versions are TR ready. HWINFO64, is

 

[Fir]

Zenith has an AMOLED display on the back plate cover. In testing it's been around 58-60C. That's probably right, albeit average reading.
Funny thing is with latest BIOS it subtracted 27 off the real reading so idling it shows 0 to -3C core temp in BIOS. That brings back memories LOL.
A dice pot for this thing would be HUGE!

About half way into an eight hour RB torture test. All cores at 100% load with blender and handbrake running. It hasn't crashed but it's definitely warm. I need to find my cable so I can take some pix with the FLIR. This is 100% stock of course. Will not be doing this with any OC until a proper block is mounted. May have to pump some chilled glycol through it so full load is around 25C.

Also would be nice to have blocks for VRMs, those are getting warm too.

 

[IEC]

1950X processor arrived earlier today. This box makes my hands look tiny...

 

[Markfw]

1950X processor arrived earlier today. This box makes my hands look tiny...

OK, so how long until YOU do a few benchmarks ! I did my share I think, now I am crunching DC stuff. I love this sucker.

 

[urvile]

I got my case and cooler yesterday and the rest arrives today......

 

[FiLeZz]

My zenith arrives tomorrow. However I will still be held back waiting on the EK waterblock. I am pretty sure the aquacomputer 480 rad and 360 will keep the threadripper in check along with my dual 1080 ti's

http://www.frozencpu.com/products/i..._Fins_-_Single_Circuit_33051.html?id=JG8Jpbu6

 

[IEC]

OK, so how long until YOU do a few benchmarks ! I did my share I think, now I am crunching DC stuff. I love this sucker.

My AIO shipment got misdirected and is delayed.

At this point I either need to break down and 2 day an AIO from Amazon, or just wait for my EK water block to arrive so I can do some proper testing with a custom loop. Decisions, decisions...

 

[Markfw]

My zenith arrives tomorrow. However I will still be held back waiting on the EK waterblock. I am pretty sure the aquacomputer 480 rad and 360 will keep the threadripper in check along with my dual 1080 ti's

http://www.frozencpu.com/products/image/16823/ex-rad-424.jpg /ex-rad-424/Aquacomputer_Airplex_Modularity_System_480_Radiator_-_Copper_Fins_-_Single_Circuit_33051.html?id=JG8Jpbu6

@stock my 360 AIO in push/pull does just GREAT ! Its the overclocking that then makes it run a little warm (hot) Depending on how far you go.

It hit 97c@4.1, but it completed cinebench and set a record, beating a 7900X@6 ghz on LN2 !

 

[Reinvented]

My AIO shipment got misdirected and is delayed.

At this point I either need to break down and 2 day an AIO from Amazon, or just wait for my EK water block to arrive so I can do some proper testing with a custom loop. Decisions, decisions...

Wait for the EK. It'll look better...and less work!

 

[Fir]

Ok the base on the Thermaltake AIO is so convex I' thinking about pulling the H115i off the 6950x system and using that.
I plan on going the EK route. Performance PCs has them on their site, preorder of course. I guess if I got really desperate I could strap one of the several dozen blocks we have lying around in the shop here like Kyle did!