The Threadripper Report (or the lack of it)

[FlanK3r]

Ouch, I guess everything is subjective then

It pisses me off that AMD essentially forces TR users to purchase a POS Asetek based AIO solution, which is simply whiskey tango'ed to be compatible with the platform.
Pretty weak if that happened on a mainstream platform, but on a halo platform? Totally unacceptable and absurd.

The TDG specs for this platform were completed exactly a year ago, yet none of the 3rd party cooler manufacturers had their products available at launch.
I wonder if all of the cooler manufacturers decided not to produce compatible coolers for the launch, or if the people who are (supposed to be) communicating with the manufacturers simply didn't do their jobs.

Since the Asetek TR4 mounting brackets are in short supply, I had to give mine away. These brackets cannot be purchased separately and none of the Asetek based AIOs ship with one.
At the moment my only option would be purchasing a Noctua heatsink (which is scarcely available and insufficient) or go with a custom loop (several blocks are available). A custom loop is not an option, as the system needs to be portable and easy to throw aside (I have too much crap laying around).
Three weeks after the launch and I cannot even boot the system.

Also while trying to make a custom mounting bracket for a aircooler (3D printing) I ran into another issue.
The moron who designed the socket decided to use non-standard PEM fixtures for the cooler mounting. The cooler mounting on SP3 socket uses M3.5 PEM nuts. M3.5 fixtures are pretty much as common as dodo birds nowdays.
Based on the SP3 MDG the M3.5 fixtures are also totally uncalled for in terms of strength. Even a M2 (let alone M3) screws would outlast the spot welded M3.5 PEM nuts. Finding any M3.5 screws was a total PITA, I had to order ones intended as the head bolts of a RC car engine (HPI) from Taiwan.
Despite all the effort I spent into the custom bracket, it ended up failing too (my fault).

From AMD's perspective TR was exactly the right thing to do. In the current situation juicing out all they can from a single design (Zeppelin) was essentially the only option.
However in my opinion a design like this (MCM) simply doesn't belong anywhere near the desktop or a consumer product in general.

If I'd find a person (crazy and / or stupid enough) who would be willing to swap the 1950X to a 7900X (or even 7820X) I wouldn't have to think it for a second.
Mind you that I currently use a 18C/36T Haswell-EP in my workstation.

Getting this pissed over a platform that I didn't pay anything for. If I had paid the retail price for the 1950X and a high-end motherboard I'd be *ucking furious.

Each on their own of course.

Try to ask Enermax, your visibility and popularity is great, I think, they could you help with Enermax Liqtech TR4 cooler.

[pumero]

The TDG specs for this platform were completed exactly a year ago, yet none of the 3rd party cooler manufacturers had their products available at launch.
I wonder if all of the cooler manufacturers decided not to produce compatible coolers for the launch, or if the people who are (supposed to be) communicating with the manufacturers simply didn't do their jobs.

It's true that the TDG for SP3 was more or less (some details missing) available from AMD about a year ago, however, there was no word about a HEDT platform back then.
Since AMD was very tight on any performance hints, nobody knew what to expect, especially from a possible Opteron-successor, so this wasn't a high-priority project, especially as AM4 was sh*tty enough with late changes to details, confusing communications and a lack of samples to get a feeling if this is going to be a good product or not -> that's one of the reasons why AM4 coolers were in short supply at the beginning.
The information about an SP3 based HEDT platform came only very late (spring), which was barely enough time to get something existing adjusted to SP3/TR4 and ready for mass production in time (more or less).

It seems AMD's communication with 3rd party vendors is really hampered by their internal structure, which negatively impacted availability of AM4 compatible solutions, even though they were much easier to do, and now TR4 again.

[chew*]

Not much to say that is not already said in the screenshots.

h220 x2 + apogee drive x2 with TR mounts artic mx2 because its $26 for 65 grams....small 80mm fan on top vrm sink.

Geardown enable because disabled and 1T CR is not stable and 2T is slower. Enabled is equivalent to roughly 1.5T CR

[chew*]

Bench Stable

[Buckeye]

So how are you liking TR Brian ?

[JPQY]

And is it Chess stable
Would like to see a few benches..

JP.

[chew*]

Nope not stable. Only mission critical server stable but not chess stable

The system took 1 10 min break from prime and hci to run 1 bench. Its back on prime but @ 3333 on ram and @ 22 hours in I am not stopping it for a benchmark....

Enduser support > random benchmarks.

[chew*]

So how are you liking TR Brian ?

Has some quirks with 2 fury x vgas yet to pin it down. The issue is annoying more so that the problem is not when your using them but when your not using them.

Pull a card and this system just works no matter what i toss at it providing A memory is tuned right. B i keep volts/clocks tuned to a lvl that i can maintain proper temps.

Overkill for my use. I had to employ Aaron to actually make use of system. Somehow he managed to exceed 64gig memory use lol. Needs 128gb for him.

Unlike in the past I have enough time to tune/tweak then walk away and run tests as I am a very busy person.

Thats why you commonly see well over 24 hour stability. Because i was busy with life for 31 hours before i could get back to the machine.

[chew*]

Same cpu clocks 1.38v to memory. 1.025 soc set in bios.

Now its running HCI so tied up for another day or so.........

[chew*]

Memtest, 4x16gb DR 1.36 vdimm 1.050 SOC .88 vddp

Vcore jacked to produce heat that Prime has no problem producing. AMD needs heat to trigger IMC instability IME.

[Buckeye]

When you get time... Run AIDA64 Stability Test.
IMO it is a better stress test program as it stress more circuits in the CPU instead of a single repeating loop of Prime.

Reason I now use AIDA64 Stability Test is back in the day when I was building HFT systems I used to run Prime95, get everything nice and dialed in and stable. Off goes the machine and guess what happens ??? yup, they start running the apps they use and BAM blue screen. AIDA64 seemed to work better for this.

[chew*]

Ive run it already. Prime and hci seem to pick up errors better. Also prime tends to peg chip harder based on temps.

Asrock told me to run hci due to my dual vga issue. I told them I was stable its not my settings...now i showed them.

Running prime same settings as above now.

Ill toss aida at it for kicks after. I do not mind running them. This rig is going to Aaron tuned by me so it needs to be bullet proof.

Last thing to do is add a 1x 120mm radiator to the 2x 220 and swap the block out see if it buys any headroom temps/clocks wise.

No clue which one Aaron picked. I sent him a few options said your the one with the phd in thermal dyanamics you pick it.

Of course he said you know this is going to annoy me and i will have to design one now lol.

[Buckeye]

Lucky Aaron !!

[chew*]

Sort of I guess.

I do not know how to utilyze a rig of this magnitude and i have 2 boards 2 cpus plenty of ram.

Double the tits on a bull factor.

Might as well make 1 usefull while the other is used to help endusers.

[bluestang]

I could utilize it by running WCG on it 24/7. Oh, and of course video encoding and photo editing when needed lol.

[chew*]

Prime done Aida running now as suggested by Buckeye.

Once it passes that I will give 3333 another shot but not to optimistic.

it failed hci earlier @3333 and I have voltage limits I refuse to exceed for this build.

[bluestang]

Hey chew, you think that MB (or others) is capable of running a 1950X with all cores at full load 24/7/365? Of courser with proper CPU cooling and airflow in case. Are the VRMs and their heatsinks up to it? Or do they need active cooling pointed directly at them.

[chew*]

I ended up strapping a small fan to top vrm sink. I have a peak of 60c vrm temps now.

Just a little cpu fan from one of those old square amd coolers will suffice.

60c is barely breaking a sweat for vrm temps but...this system is meant to be used for the long haul.

Lower temps longer life imo.

Not necessary but does not hurt.

I see no reason why it can not handle it @ 3.8.

Beyond that until I experiment with cooling more I honestly can't say.

Ive seen. 69c on an enermax aio @ 1.35v @ 3950 from a user who did a hour prime run for me. Fans were flat out full 2300 rpm so keep that in mind.

Users who claimed 4.1 @ 1.45v on it(enermax aio) with 68c full load by that above result i am referring to...are full of . or not really full load for hours. Either way...misleading = full of

[bluestang]

Cool. Thanks for all the info here and the hard work you do!

[Opteron96]

Last thing to do is add a 1x 120mm radiator to the 2x 220 and swap the block out see if it buys any headroom temps/clocks wise.

No clue which one Aaron picked. I sent him a few options said your the one with the phd in thermal dyanamics you pick it.

Of course he said you know this is going to annoy me and i will have to design one now lol.

LOL, nice one

[Buckeye]

Ive run it already. Prime and hci seem to pick up errors better. Also prime tends to peg chip harder based on temps.

Asrock told me to run hci due to my dual vga issue. I told them I was stable its not my settings...now i showed them.

Running prime same settings as above now.

Ill toss aida at it for kicks after. I do not mind running them. This rig is going to Aaron tuned by me so it needs to be bullet proof.

Last thing to do is add a 1x 120mm radiator to the 2x 220 and swap the block out see if it buys any headroom temps/clocks wise.

No clue which one Aaron picked. I sent him a few options said your the one with the phd in thermal dyanamics you pick it.

Of course he said you know this is going to annoy me and i will have to design one now lol.

He needs to use one of those Phantom Evaps with a water block head on it

[chew*]

Ne needs to use one of those Phantom Evaps with a water block head on it

yah we discussed that its a lot of weight on the socket and the mixed metals alum top is sort of a problem or was last time.

took me forever to dealuminize the evap last time.

[chew*]

Here you go Bill.

[Buckeye]

Good deal man

[chew*]

Ok think we have the multi card issue nailed down, bug in chipset drivers for onchip USB........unplugged usb mouse/keyboard and plugged into another usb so far so good.

Threw this together to show what kind of real world power this system has now @ the proven stable speeds.

Most likely not the best per app but considering its going to be multitasking real world better to have all benchmarks open to get a better real world indication.

Unified memory access mode as that appear to offer better performance the way this machine is being used.



Now to wait on water blocks. Aaron ordered one and then I said what the heck I will order the mystery brand more so because I am a cheap bastard.