W790 Memory Overclocking

I got my hands on the CPU. 3435X, 3465X, and 3475X are now available as boxed, but 3495X is tray-only.

Sadly, G.SKILL still doesn’t seem to have a 256 GB kit available for purchase, so I’ve got DDR5-4800 RDIMM 64 GB x 4 for now. Looking at Hardwareluxx.de’s benchmark, the Xeon seems to scale pretty well with memory speed, so I hope to get at least DDR5-5600…

I haven’t got this thing running yet, I ran out of coolant for my watercooling loop…

3 Likes

Nice.
I’m on the look out for a 3435X now, still don’t see any stores selling them in the states yet.

I’m going to try to hold out for the G.Skill 256GB DDR5-6400 kit, F5-6400R3239G32GE8-ZR5K, which isn’t even listed on their website yet… we’ll see how impatient I get

2 Likes

I’ve got the machine to boot & everything working. Turns out, Supermicro board does support memory overclock beyond 5600 MHz, but I can’t find anywhere to adjust memory voltage (or does voltage work differently for RDIMM? Given it has an on-DIMM voltage regulator?)

From what I gathered, is that the stock is extremely limited. Here in Japan, there are 3 retailers listing them for sale, and each of them seems to only have a single unit of each 3435X, 3465X, 3475X, and 3495X.

(p.s. look at all my cores!)

2 Likes

Done any benchmarks yet? I’m curious about cpu-z single core benchmark and I haven’t seen anyone post one online yet.

The Asrock W790 board has memory voltage settings in it’s BIOS:

alittle more complicated than how I remember DRAM voltages settings being.

I’ve done some quick benchmark runs. Everything running under the default PL1=350W, PL2=420W, and with Octa DDR5-4800.

Cinebench R23


(Yes, I haven’t activated Windows. I’m just using it to set up my loop hardware and a few benchmark runs, please don’t judge :smile:)

CPU-Z:

Stepping information:


(“E5/S3”? The sample Derbauer got on his video was “E5”, and ES that overclockers got their hands on was “E3”. Is this the case of v5_Final_Finalv2_REALLYFINAL?)

Some initial findings:

  • The w9-3495X is clocked so low that Windows’ default “Balance” power profile makes the entire system unresponsive. Every interaction took 2+ seconds, e.g., to open Windows Explorer. HWiNFO reported a very low number of 180 MHz for “Core Clocks” with Uncore VID of 0.690V.
    • Setting it to “High Performance” makes “Core Clocks” runs at 400-500 MHz range. The system is much more responsive at the expense of higher power consumption (80W vs. 150W). I think anything below w7-3465X may do much better in this regard.
    • CPU-Z and HWiNFO report quite a different number for clock speed at low power states. I’m not sure whom to trust here.
  • A simple bumping PL1/PL2 to 800W boost the multicore performance by about 10% (CB R23 score went up to 65000-66000 range), but I found that stock clock settings for w9-3495X never exceed 500W while running Cinebench R23. Some tuning is probably required here.
  • Since the die area is so huge, and I’m running a custom loop (360+480 radiators) my CPU never exceeds 50c even after 2 hours of y-cruncher run. Air cooling could have been a practical option.

Generally, I’m happy with the platform, but I think w9-3495X really requires overclocking. Intel seems to clock the CPU far too low to meet the power budget, and it would be interesting if Intel were to release Emerald Rapids-WS with DIVR. Please let me know if you want me to run any other benchmarks.

Ah, thanks for this. Memory overclocking on Supermicro is really a no-go, then. I can’t seem to find any way to adjust memory timings either. (There’s an option to play around with RAS, but not RAS timings).

3 Likes

hmmm… I would have expected the single core cpu-z score to be higher, maybe around ~760 points at max clocks. Did you happen to observe what clock speed the cpu was running at while the single core portion of the test was happening?
I’m suspicious that Windows is mishandling clock ramping and maybe even core scheduling on such a new architecture.

oh there is one benchmark I’d like to see if it isn’t too much trouble. It’s located in this thread:

Its a benchmark for a FEA problem I was solving that stresses memory particularly hard; It does require a comsol runtime installation (part of that big executable in the drive) though which is kind of a pain.

Looking to get one of these to upgrade from my Xeon 8280 machine, so thanks for the benchmarks!

if you dont mind, would you share some Build Pics always nice to see some pretty pics;)

Some quick overclocking; bumping All Cores Turbo Ratio to x36 (from x29) and setting PL1=500W, PL2=570W.

And 2 Cores Turbo Ratio at x50 (from x48):

Screenshot (45)
(Though, I don’t think it reaches max clock during this run)

The VRM on the Supermicro’s X13SWA-TF shoots to 100 °C and throttles immediately once it reaches 500W. But this is with very little airflow, adding some fans will probably help.

In that test, I believe it was running at around 4 GHz. I think the clock ramp-up behavior is very slow for some reason. In Cinebench I saw the score went up the longer I run in the multicore benchmark. Running CPU-Z benchmark in succession bumped the score up to around ~680, but it couldn’t go any higher due to the length of the test.

I think this may be a bug in Windows rather than a problem with SPR; on Linux the system stays very responsive in the lowest power state (consuming ~60W idle vs ~150W on Windows with “Performance” profile).

I couldn’t successfully run this on both Windows 11 and Linux; on Windows it crashes immediately after the splash screen (COMSOL installation was successful), and on Linux (both Ubuntu 23.04 and KDE Neon) it crashes after compute to about 40%. Is there anything I should try?

Sure :smile:

The build is still WIP, I’m still missing a GPU (or maybe GPUs) which is why there’s a tube looping from a Manifold distribution block into itself; otherwise it won’t work, 4 more RAMs, redesign the Manifold mount so there’s some more gap between the fan and the Manifold, etc., but I’m happy with how it turned out.

6 Likes

I agree

TBH I’ve never run into that before. Ignoring the Windows results, the only thing I can think of that would cause a crash is a system memory error which I would assume wouldn’t happen with ECC memory; did the IPMI log any memory errors around the time it was run?

Edit: Another thing that would be able to cause a crash is the system running out of memory during the run, it’ll silently crash if this happens.

​​​ ​ ​

​​​ ​ ​
Newegg now has the W-3400 SPR-WS cpus for sale, just unlisted. I should have mine in 2 days.

1 Like

I’ve heard from a Japanese reviewer that ASRock’s latest BIOS for W790 WS (3.04 released 2023/04/21) has fixed the responsiveness issue when running “Balanced” performance profile on Windows (although they were running 2495X). Still nothing from Supermicro or ASUS, however.

On the other hand, it looks like VRM on Supermicro board throttles at around 500W. I tried putting the fan on the VRM and gained 4,000 extra points from previous run (All Cores 3.4 GHz):

I think the Supermicro board may be more suitable for air cooling, since it looks to rely on some air going through the heatsink fins to cool the VRM (direct airflow is blocked by RAM).

Since I see it tries to create OpenGL context, maybe it has something to do with Intel ARC driver on both Linux and Windows. I didn’t see any memory-related errors on both Windows and Linux side (though I didn’t dig too far). On Linux running out of memory could may be the reason; I only have 256GB installed right now with no swap.

Congrats on getting the order through!

I’ve got the W790 SAGE w/ a 3475X and 128GB of G.Skill 6400 Zeta goodness. Works like a charm, immediately booted with XMP on, and haven’t had any faults yet. Once I can get a bit further along with some other parts, I can toss the other three 4090s in the rig w/ their respective waterblocks.

Currently using the EK Pro 4677 block and not a single lick of trouble from it. Using the third port to monitor liquid temps

3 Likes

I’ve got the same block and board, I’m even doing the same with the temp sensor.
Once you have final setup, you should post pictures. I’ll be posting pictures of mine tomorrow when I get my last pump bracket, it’ll be my first water cooled build.

Out of curiosity, what torque wrenches/screwdrivers to mount the blocks are you guys using?
I saw that EK made a screw driver for mounting the blocks but ~20USD seems too cheap to be a proper torque driver.

​​​ ​ ​

​​​ ​ ​

I watched the kitguru video on the w790 and they mentioned that the discrepancies in hwinfo reporting were due to Windows 11’s HVCI, Hypervisor-Enforced Code Integrity, being on by default, apparently turning it off will fix the hwinfo reporting issues and possibly improve performance.

If you’re interested in troubleshooting if the Intel ARC driver is the cause, you can force the application to use a software renderer instead of an opengl renderer by using the argument -3drend sw during launch.

Now that I think about it, running out of RAM is definitely the problem for the failure during run; RAM usage and core count usage are related and I originally designed the benchmark to use 200GB of RAM for a 32 core machine, going up to 56 cores likely pushed the RAM usage over 256GB. An -np 32 argument during application launch would force the use of only 32 cores and be leaner on memory.

​​​ ​ ​

​​​ ​ ​

​​​ ​ ​

​​​ ​ ​
The only component I don’t have now is RAM. I really wanted to wait for 32GB Hynix A-die memory to come out but I am getting impatient. Kingston has 32GB Hynix M-Die RDIMMs they sell that’ll run at 5600MHz.

2 Likes

I have the EK torque screwdriver. It worked fine for most EK stuff at 0.6Nm, although Intel seems to recommend 0.9Nm for LGA4189 (which I assume is the same for LGA4677). I’ve been looking to buy one of Wera 074700 (0.3Nm–1.2Nm) but it went out of stock a few weeks ago when I needed it… (perfect timing).

If you’re building a first custom loop, I recommend looking into Aquacomputer’s stuff and some external temperature sensors (e.g., an inline water temperature sensors). Aquacomputer Quadro (or Aquacomputer Aquaero) is an excellent PWM controller that can run autonomously based on an external temperature sensor. (The only downside is that they require a license to update Aquasuite software after 18 months or so, though you don’t really need to run the software).

EKWB’s recommended 0.6Nm seemed to be on the low end of all the recommendations I was seeing for the socket too. As far as I can tell all the sockets use nuts made out of the same PEEK material so it’s weird seeing such a wide variety of torque specs (Asrock recommends 0.7-1.35Nm for the socket, which is the widest range I saw).
I ended up getting a Tohnichi RTD120CN (0.2Nm–1.2Nm)

I got Aquacomputer’s dual pump top due to my paranoia of one of the pumps failing:

I might end up needing an external pwm controller because the fans I’m running for the radiator peak at ~15 watts each, SilverStone AP182s.

2 Likes

I’ll see if I can snag some photos this evening or weekend.

Currently the system is shoved into a CaseLabs S8S case and just the cpu is on water. I have a CaseLabs MH10 arriving next week that I might shove the system in for a bit until the parts for the TX10D become available.

I have most everything connected to an Aquaero right now, but not doing much with it atm besides fans and monitoring temperatures because the case is temporary.

The torque spec is interesting to me. The captured nuts on the 4677 pro block from EKWB seem to be plastic so I was very worried about over-tightening them and stripping the insides out.

Is caselabs back in business?

The nuts are Polyether ether ketone plastic which is basically the most expensive, performant plastic available… That being said now that I’ve properly torqued the nuts down, 0.6Nm feels like alot.

​​​ ​ ​

​​​ ​ ​
I haven’t been able to complete my loop yet due to running into some unforeseen issues with a reservoirless setup but I should have more fittings by Wednesday to remedy.

​​​ ​ ​

​​​ ​ ​

On the subject of memory overclocking,

I started looking at the Kingston FURY Renegade Pro DIMMs and I found out that even the 6000MHz SKUs are running the M88DR5RCD01 RCD which is first generation chip which means that it’s overclocking potential will be limited, the M88DR5RCD02 RCD is already out and is binned for 800MHz more nominal clock speed.
More reason to wait for the G.Skill memory to come out.

https://www.montage-tech.com/Memory_Interface/DDR5/M88DR5RCD01
https://www.montage-tech.com/Memory_Interface/DDR5/M88DR5RCD02

3495X has arrived. 256GB kit of 5600, Kingston Fury should arrive this evening. Still waiting on EK water block to ship, and until then I’ve the Noctua NH-U14S DX-4677.

@sirn looks like our processors are consecutive serial numbers.

6 Likes

Rofl that’s amazing!

I am having tons of fun with my falcon northwest build

Will you be doing an updated video? Interested what you’ve learned.

Got the epic deliveries this week. TRPro 5995WX arrived yesterday. Both machines are closely spec’d (RAM capacity + Storage + Video), for A/B comparison.