The Threadripper Report (or the lack of it)

[chew*]

As the title states.

I had a "report" written up as I refused to call it a review ( due to my inability or need to ever have 32 threads for my usage ) but I had to toss it.

The reason I had to toss it is because I was requested to by AMD. No need to get the pitchforks out yet till you hear the why.

The reason why is it appears a bunch of asshats violated embargo. No clue who or how and I really do not care.
The only thing I know is the domino effect reached as far as me which I should note is incredibly hard to do....so congrats on that. Someone really screwed the pooch.

Anway after i get a coffee in me i will offer what little info I am allowed to that may or may not have been covered in reviews.

[chew*]

Ok so as promised I had Built a Threadripper rig with 64gb ram in a case.
I ended up water cooling it with a swiftech h220x 2 with custom mounting and shoving it in a corsair case.

All newly purchased about $1k plus invested to do the report.

Horrible pic but at this point in time I just do not really care.


Anyway moving on to useful info.

Threading on socket is m3.5-6 for the DIY guys that want to do custom mounting and can be found at McMaster carr.

Alternatively if you end up in a situation with a chip and no "kit" a craftsman torque wrench will work @ 15inch pounds in a pinch to install chip.

I personally would avoid the AIOs unless you already own it as they quite frankly suck.

My system spent 90% of it time under prime 95 which btw works......read some info watched some vids......prime 95 29.1 works just fine.

The why I was under prime is to figure out the chip and throttle points.

Using just some common sense and a R7 chip scoring was 1700 3.8 - 1750 3.9 - 1800 4.0. extrapolate that to double the core count minus a small % scores should be about 3350 3.8 3450 3.9 3550 4.0 in cinebench.

Pretty much I can tell you the leaked scores around 4.0 suck and it was not because of ES it was because THROTTLE. Retails are even better.

Moving on from that I grabbed a pic off the net and modified it to what I believe are the active cores on TR....I might be wrong but my motivation is purely taming temps.

Take this pic with a grain of salt and keep in mind even if I did know I could not tell you anyway. I will just say from a connections point of view it makes sense.



The Board I am on is a Asrock x399 Gaming, it held up very well considering the amount of prime 95 I ran and the fact that it has no VRM sinks whatsoever ( ES sample for most part ). It never VRM throttled the entire time I used it but did run a tad warm under prime 95 worst case loads sub 90c temps were logged.

Anyway special thanks to ASrock AMD and swiftech ( for sorting out a problem quickly for me )

I will add more if I get permission and if I can think of anything else.

[KiSUAN]

I was actually waiting to see what you had to say

[chew*]

I had quite a bit but deleted it. At least 2 full hours of video and a hell of alot of results.

You can thank the embargo breakers.

AMD had no issue with me doing my report until that happened.

[StAndrew]

I don't understand. How can an embargo leak affect any review post NDA?

[chew*]

Quite simple. I was on ES parts. Embargo violators were on ES parts. AMD owns all ES and can at any point in time recall all chips and results. It is there right.

I had originally discussed not wanting to review TR with AMD. I felt resources were better allocated elsewhere.

That said and due to that decision I was not on Retail.

The only reason i am on TR at all is to help out Nick Shih and Asrock and I will continue to do so once on Retail parts.

I do have interest in the 1900x model btw just not the higher core stuff.

You can expect something from me on that and a comparison vs 1800x x370 platform vs 1900x x399. Pros/cons and which platform suits your needs best.

[StAndrew]

Quite simple. I was on ES parts. Embargo violators were on ES parts. AMD owns all ES and can at any point in time recall all chips and results. It is there right.

I had originally discussed not wanting to review TR with AMD. I felt resources were better allocated elsewhere.

That said and due to that decision I was not on Retail.

The only reason i am on TR at all is to help out Nick Shih and Asrock and I will continue to do so once on Retail parts.

I do have interest in the 1900x model btw just not the higher core stuff.

You can expect something from me on that and a comparison vs 1800x x370 platform vs 1900x x399. Pros/cons and which platform suits your needs best.

Oh, I see. Awesome, can't wait for your review.

I'm still looking for a review that includes, not just CPU bench, but total platform utilization, ie demonstrating the power of 60 PCI-E lanes.

[chew*]

Here is a quick overview on Asrocks Fatality x399 gaming board.

I will do far more in depth report when it has a cpu installed.

https://youtu.be/NaZJzPtWCQE

[BlindShot]

I'm really looking forward to any info you can provide Chew. I'm holding out for the 1900x and asrock mb if everything looks good.

[chew*]

Ok got permission for this one.

Chopped means don't ask. If I told you I would have to kill you.

Crossed out means irrelevant or incorrect.

Circled means correct and or relevant.

Cooled by a swiftech h220 x2 and to be quite frank I broke the rule #1 of water cooling.

If you feel heat coming out of radiator you need more radiator. I need more radiator........

Custom mount and the apogee cold plates only half cover die.

Anyway after seeing some results floating around I would say this particular chip was less than stellar so that is a good sign.

3.8 @ 1.2v without throttling is nothing to complain about. Prime 95 29.1 avx 512 version....

[chew*]

For those that may be curious or interested.

The only difference I can find on the x399 Fatality vs the Taichi is the sound and the 10gbs lan and a fairly significant price difference.

I just checked all the current offering and prices so for a benching/fun board.....looks like the taichi is the only one in my budget.





The Asus seems like a really nice board but I can not justify $500 for benching at this point in time, maybe for a daily user but not a disposable bench board.

Really miss RC Bluetooth/Rog connect...would like to see that brought back. I like monitoring my machine without any software installed on the machine.

[BlindShot]

The taichi looks like a good choice for me. Was it difficult to hit those memory speeds on that Asrock? I'm currently selecting all my parts for a new build and I'm not sure about the ram? I'm just looking for a stable rig for doing everyday stuff.

[chew*]

It was pretty much porting my timings from taichi x370 and it just worked save for the 2T command rate.

1T boots.

Dropped a thread in prime 95 896K iteration 1 hour in..... etc memory fault.

Set 2T ran for a day fine.

With 2 sticks only 1T works on my x370...4 sticks requires 2T.

I had hopes that 1T would work in quad channel with DR dimms and not all slots populated but whatever it is not a deal breaker.

[drmrlordx]

Good to hear that you managed 3.8 at such a low voltage using a block that didn't even cover everything correctly. We've had a few trolls on other boards saying that poor coverage of the IHS will lead to major temp problems.

I'm interested in seeing what throttling is happening in some of the other TR reviews out there.

[chew*]

The apogee almost has full IHS coverage and covers die completely but the coldplate pin coverage is about half die like many aio.

The drive x2 has a better coldplate (coverage over die) for TR but still not optimal coverage over die imo.

I have no delusions however.

My first issue was not enough radiator.

[StAndrew]

Does the intercore infinity fabric sync with memory speed like the inter CCX fabric?

I'm curious to see scaling on different memory speeds.

[Charles Wirth]

Thread selection test:

If you can please run 3D Mark Vantage and run just the CPU test with HT on and off.

Why? to see if the selection of the HT is being favored over the real cores.

You can also use Time Spy physx as it is limited to 20 threads.

I expect the cpu scores to decrease with HT on, not that we can fix core scheduling but we can disable HT.

I know it is not HT, just using as example.

[chew*]

Already confirmed it just can not post results with the above chip.

Both time spy and vantage cpu score went up on the 16c...with smt off.

I can not say the actual scores...but off the top of my head smt off went up by 2k in timespy cpu test.

Vantage cpu went up to even on the latest version which iirc is 12 core or thread limited unlike the older version.

Overall score in timespy iirc did not change.

Iirc when i was pushing for the top timespy score on bot with fury x around 9k ish cpu score was enough to turn it into a VGA bottlenecked score with the 1800x R7 cpu @ 4250mhz.

[Charles Wirth]

Seems like a Microsoft issue that needs to be resolved with both AMD and Intel.

I was wondering if the TR had addressed this issue with improved scheduling.

[chew*]

Unless it was due to my parts and somehow retails schedule better ( unlikely as afaik it is just ipc increase and better clocks ) then yah probably needs an update through ms or maybe chipset drivers?

[FlanK3r]

Threadripper on the way at my side also guys Im looking forward as small kid at a toy

[chew*]

Micocenter east coast NJ,NY,MA have various TR boards in stock now.

Asus zenith, gigabyte k7, msi pro carbon.

Iirc longisland has 5 asus zeniths in stock.

Cheers.

[chew*]

Current stock @ microcenter boston.

[chew*]

Newegg has stock of taichi finally. Ordered mine. I will order my swiftech mounts over the weekend. I need a replacement fan anyway..

[Buckeye]

Have a 1950x TR here now for a few days, MSI Gaming Pro Carbon for my first board, more incoming.
My cooler is a Thermaltake Water 3 Ultimate, handles 4GHz OC just fine, no hot air coming out the back... yet.