Dual CPU system or Single 5995X CPU system?

Good afternoon everyone! I’m back on here again trying to figure out the best workstation parts I can currently muster with today’s available selection of hardware (3/14/2023). This build is going to be primarily used as a data-image analysis parallel processing system.

I’m looking for easy maintainability, reliability, and upgradability. Cost is definitely a factor as well, but I think I’d be willing to upgrade certain components every 2-3 years if the price is right.

Please let me know if there is better or newer hardware if you see something amiss!

For CPU: Deciding between a dual CPU system or a single CPU system.
-Either, Dual XEON E5-2699 v4 (22-Core/ea.) OR Single Threadripper PRO 5995WX (64-Core);

-For MOBO: Supermicro MBD-X10DAX for Dual Xeons OR ASUS WRX80E-SAGE SE for TR-5995WX;

-For RAM: 8 x 128GB LRDIMM (2400Mhz) for MBD-X10DAX MOBO OR 8 x 128GB LRDIMM (maybe 3200Mhz, depends what’s supported) for WRX80E-SAGE SE;

  • at least for the MBD-X10DAX MOBO, I can always slot in another 8 x 128GB versus, I’m not sure when 256GB (3200Mhz) DIMMs will be available in the 8 x 256GB (3200Mhz) config for the WRX80E-SAGE SE;

-For GPU: Either… RTX 6000 ADA 48GB VRAM OR Dual MSI Suprim Liquid X RTX 4090 24GB VRAM/ea. (thinnest 4090 w/ 240mm AIO attached);

  • I’m aware that NVLINK is no longer supported, so… I’m still trying to figure out if they can be parallelized for virtually combined VRAM-pools). I will say that RTTX 6000 ADA is extremely expensive and I fear the depreciation is exponential considering next year’s model will most likely out perform the 6000 ADA by a huge margin…

For PSU: With the single RTX 6000 ADA I think I can get away with a Silverstone 2050W PSU (Platnium-Rated) VS. With the Dual MSI RTX 4090 Liquid X, I’m pretty sure I’d have to run 2 x 1600W Titanium-Rated power supplies;

For Case: Any of these configurations should fit in the Phanteks Enthoo Pro 2 (non-TG);

That’s pretty much it! Thank you again for your feedback and any advice or clarification would be greatly appreciated!

Short answer … 2 x e5-2699v4 scores around 41000pts on CPUmark

A single 5975wx scores about 80000pts

Two architectures that are stratospheres apart

A single 5995wx is around 95000pts

Passmark’s aggregate scores are meaningless because most submissions are from systems running Windows. For example by running Linux the performance of my old dual 2697a increases some 20% over the mostly-Windows aggregate score. I’ve observed even larger discrepancies with my Skylake, Rome, and Milan systems. That being said, a 5995WX will easily be 2X faster and likely use less power to boot. So, when comparing processors this way it’s best to look for an individual result that best resembles the system you’re planning.

If price is no object it’s worth noting that you won’t be getting current technologies like PCIe 5 or DDR5 on either system (Broadwell or Threadripper Milan), which you may want down the road on a system with such high PCIe bandwidth requirements.

Especially for a system that you plan to use for 3+ years, and if you’re looking to upgrade to future GPUs that will support PCIe 5, you might look into a 32 core Genoa system.

2 Likes

I hear what you are saying and agree by and large however consider the following :
CPUmark “meaningless” scores on e5 CPUs are equally meaningless on threadripper 5000based.cpus … I guess what I’m getting at is its really just a tool/yardstick and it faithfully demonstrates there is simply no comparison between legacy xeon e5 and threadripper 5000 . They are not even in the same ballpark .

My 2c, the op needs to ask this question
Is he going to on save money and build this system out of cheaper used bits and pieces and will 1/3rd of the performance suffice ?

If budget isn’t a consideration then spend the money and go new and future proof somewhat …

It’s not just the CPU speed to consider but total architecture gains …

A previous lab system I strung together was a dual xeon e5-2680v4 with 256gb ddr4/2400 registered ram ,I ran my same Nvidia a4000 GPU on it , ran a 4 x nvme bifurcation setup and also a Intel PCIe 750 nvme …

I moved to a wrx80 threadripper 5975 w/ ddr4/3200 ram… Not only is the CPU more than double the prev config ,the architecture is so much faster … the PCIe 4.0 lanes etc …the comparison simply put the 2 machines are stratospheres apart … Unfortunately so is the cost

I agree the Threadripper system will be faster but that’s hyperbolic. My 64 core Epyc Milan system is exactly twice the speed of my Dual CPU 2697a v4 system in heavy SSE and AVX2 code - that is to say, scales proportionately to core count. Milan is faster clock for clock, more efficient, etc. but Broadwell is hardly the Stone Age :laughing:

If the system is mostly for GPU computing, which is what I gather from the original post, it may be worth saving the money to put towards another A6000 or 4090.

There’s also platform quality to consider; I’ve lost count of all the PCIe cards I’ve personally tried that are either unrecognized or non-functional in my AMD machine. This isn’t an uncommon issue.

The WRX80E board seems particularly problematic based on the number of issues I’ve seen on this forum:

And that’s only a tiny fraction of them :face_with_spiral_eyes:

It might be worth considering other motherboard options if you opt for the Threadripper system.

Hello there ! I’ve been using the WRX80E-Sage non-stop since I got it and fixed the issues I detailed. It’s absolutely solid. While this board was not my first choice, it’s definitely robust. Best proof you may want is that I’ve been so busy getting sh… done I haven’t posted on this forum in several months :smile:

Just thought I’d let you know.

2 Likes

This topic was automatically closed 273 days after the last reply. New replies are no longer allowed.