Prototype 4x Titan V

Venturi, THANK YOU for taking the time to post pictures and answer all of the questions.

Beautiful and impressive rig.
After this thread, I am good to go for another year of splurging on computer parts without feeling guilty.
Small update

Using 1TB of RAM and using Titan V 32GB CEO edition

How's the rig been running for the past year?

It has been running great. Its gotten a raid nvme upgrade, 1tb of ram, and titan v ceo enhancments.

Its also my daily rig for play, work, email, school, etc.

I bought 4 rtx titan cards, but on my rig the titan v was a better performer, so the trex is not in it anymore.. That's not to say the Trex isn't a good card, it is, I just get more out of the titan v.
Last edited:
Dude wahtever it is that you do you must be bringing in crazy amounts of excess cash. More power to you! I'd love to know what you are running that needs a TB of ram and benifits from 4 Titan V cards! Holy crap.
medical imaging nuc med, DBs, isotopes, etc ...and schoolwork, games, email, web.

This is my home rig, personal, not company owned, I just do work on it to get ahead.

I would love to see some raw output numbers... just to get an idea of how those Titan V chips stack up. I bet you are putting your work systems to shame. And to have the cash to invest... do you get a grant for doing this work to afford systems like this?
I would love to see some raw output numbers... just to get an idea of how those Titan V chips stack up. I bet you are putting your work systems to shame. And to have the cash to invest... do you get a grant for doing this work to afford systems like this?

Because really if I can get a grant to do imaging work like this on the side I would totally love to build a box like this and keep it current like what you are doing. Thats just amazing.
Your home box has more ram than the three nodes my company uses for its Kubernetes cluster...
this is from the adjacent thread, but I figured it may suffice for some gaming benchmark.

Before going any further let me say I don't overclock anymore, this rig works fine as is and overclocking on prior builds always had consequences. I'm into visual quality and stable. This box is rock solid. With that said, I do play games on it as well -and when I set up a game I max the visuals -except AA, for that TAA or SMAA seems to be all I need at 4k. The remaining visuals are maxed as well as post processing and the sort. I do turn off BS such as film grain, vignetting, DoF and Motion blur. All else maxed. I go further into the config files and eek out any remaining visuals I can.

Lastly, I don't steam, ea, origin, Ubi etc and no DRM. I actually BUY the games in their box whenever possible (usually collector editions) and then proceed to run drm free. Despise denuvo and other performance and privacy robbing third party apps. I like running right from the game executable in pure mode.

further, running base benchmarks in the canned settings is not very satisfying. Most benchmarks don't do mgpu or SLI (I run mgpu and auto diff sli). Most benchmarks don't thread enough to take advantage of 44 cores (88HT) or the resources of the box. So benchmark software is not very private, is not optimized, is for generic configs, and seldom makes use of real resources.

It was me that was first ran quad SLI on pascal, fully featured. There are plenty of forum posts I made with detailed how toos, and relative benchmarks. Without the micro stutter, hitching etc.

Lastly, benchmark results are unreliable. So many of the high scores will have artifacting, instability, etc. A pc should be able to run for months and months without refilling cryogen, not do ANY damage to the components, and be rock stable. Also, its very easy to make a video profile that redacs many features to get higher frame scores and fool benchmark engines. Forgive the rant. Most people don't understand.

with that said

I believe I will be disappointing some folks as the scores are not earth shattering. On my rig the individual cards faired as follows:

Time spy downloaded from techpowerup (I don't steam)

time spy - no changes just hit the "run" button:
(and no overclocking)

Titan V 12GB 12422

Titan V 32GB 13863

nothing spectacular, and somewhat of a let down, but also NO overclocking.

There are a few things I noticed:
3dmark didn't really make use of my processors, and they didn't turbo up much. (44cores/88threads 3.7ghz turbo doing very little)
I don't think it really made use the titan Vs because the temperature never got higher than 69c on the TV32gb and 71c on TV12gbs, never saw any thermal throttling,
room temp 72F

Also, the extra ram of the TV32GB edition never got used, I think the only advantage was the 4096 bandwidth, but the amount of ram did not pose any advantage on the benchmarks.

While I was installing benchmarks I also tried Superposition 4k optimized setting single gpu and no customizations

The score was 12790

I uninstalled the benchmark stuff and the little things they left behind.

I'm not very impressed with the scores, unless all those 14k, and 15k scores are some extreme overclocking (I don't overclock)

I just returned the config back to normal (I had to move cards around to benchmark) all 4 cards back in system in the correct order

On my rig 2 cards go to CPU 0 and 2 cards go to CPU1, with QPI, COD, above 4G decoding, SR-IOV Support, MMIOH Base, MMIO High Size, MMCFG Base, Maximum Payload, IOU non posted pre-fecth - all 4 cards work well. I have to do some end of year reports for work, some school work, and earn a living to pay for this rig. I need a less expensive hobby.
Last edited:
What windows did u use? did u check out how many cores were used when testing?

Windows server 2016 Data Center with optimizations (win 10 without telemetry, includes dx12, no forced updates, no specter mitigation, etc) running as a workstation

3 cores were used in testing but there were 85/HT more that sat quiet and bored (well, some activity showed on the other cores but all less than a few % and thats ripple of 1-2% percent usage is not unusual)
less than 1% on ram usage
less than 8Gb of vram usage