Was tweaking and looking at what could still be improved to raise my Time Spy score, and hit 21000 exactly. That was good for #10 for the Ryzen 5900x and 3080ti.
 
Pretty pleased with that. I'd only run a mild OC while the card was air, but since the Optimus block showed up I've got the freedom to push it much harder, and have been doing so for these benchmarks. I don't usually run it like this while gaming because the game I've been spending most of my time in lately doesn't benefit from it at all. When I get back to Star Citizen again I'll be curious how it runs, and whether this OC is perfectly stable in a game that demanding.

#1 with 21109.
 
Wowzers. That was done with Precision X1 set to +1675 VRAM and +210 GPU with power set to 113%.
 
My machine doesn't use any extraordinary cooling measures like dry ice, chillers, or anything more than that the CPU, GPU, and RAM are all conventionally water cooled and overclocked and I've got the rough equivalent of around three 360s' worth of radiator surface area for decently low water/air deltas. The 3080ti FTW3 is water cooled with the Optimus Absolute water block, the Ryzen 5900x with the Optimus Foundation block, and the RAM with an EK RAM block. The RAM is water cooled because I hand-tuned all the settings and run it at 3800/cas14 at 1.52V VDIMM with hand-tuned primaries, secondaries, and tertiaries. It's nearly as fast as 3800MT/s RAM can run, and since my 5900x can only do 1900MHz on the Infinity Fabric it's about as fast as RAM could possibly go with my particular 5900. If my 5900 could do 2000MHz on the Infinity Fabric the RAM would go that fast. I've run it up to 4400MT/s before, but desynched from the IF the latency penalty is killer, so 3800MT/s it is. My CPU uses an aggressive and nearly maximal set of PBO2 and Curve Optimizer settings. I say "nearly maximal" because my Curve Optimizer per-core settings are only tuned to a granularity of +/- 5. Between the CPU's own PBO2/CO push and the aggressively OCed RAM my CPU score is actually what pushed my score over the top. This can be seen in comparisons with other scores in the top 10.
 
Yesterday I got stuck in the 20600-20800ish range just trying to reproduce the 20943 I'd once gotten right after I installed the Optimus water block on the 3080ti and was playing around with the OC. I did some reading and manually shut down some system services and some other services that were running that I didn't need. Most of these, like the Microsoft Fax service, will stay off forever. Some of these weren't actively disabled, just stopped, and will restart when I reboot the machine. These included services related to software licensing, updating, and some other things from SolidWorks and Adobe. I don't know if stopping or disabling any of those services made much of a difference, though I did find myself climbing back up into the upper range of the scores I was getting more frequently.
 
The real kicker was figuring out that I could restart Steam in offline mode. I had 3DMark started through Steam, so I can't have Steam not running while benchmarking. Precision X1 was also started through Steam. I figured out I could load an OC profile, make any tweaks and apply them that I wanted to in Precision X1, and they'd stick even after I closed PX1 until I rebooted the machine. Once I restarted Steam in offline mode I shocked myself getting that 21000 that I OPed about in this thread. Apparently Steam would do some random network pings of various Steam services during the benchmark runs and it was just enough to cost me 200-300 points in Time Spy. Going offline erased that. I did a couple more tweaks of the VRAM speed and got low 21000ish score, then one more +25MHz VRAM tweak crossed 21109 points for #1 for machines using a Ryzen 9 5900x and a 3080ti from any company. I'm actually kind of shocked about this.

sethleigh
#1 with 21109.
 
Wowzers. That was done with Precision X1 set to +1675 VRAM and +210 GPU with power set to 113%.
 
My machine doesn't use any extraordinary cooling measures like dry ice, chillers, or anything more than that the CPU, GPU, and RAM are all conventionally water cooled and overclocked and I've got the rough equivalent of around three 360s' worth of radiator surface area for decently low water/air deltas. The 3080ti FTW3 is water cooled with the Optimus Absolute water block, the Ryzen 5900x with the Optimus Foundation block, and the RAM with an EK RAM block. The RAM is water cooled because I hand-tuned all the settings and run it at 3800/cas14 at 1.52V VDIMM with hand-tuned primaries, secondaries, and tertiaries. It's nearly as fast as 3800MT/s RAM can run, and since my 5900x can only do 1900MHz on the Infinity Fabric it's about as fast as RAM could possibly go with my particular 5900. If my 5900 could do 2000MHz on the Infinity Fabric the RAM would go that fast. I've run it up to 4400MT/s before, but desynched from the IF the latency penalty is killer, so 3800MT/s it is. My CPU uses an aggressive and nearly maximal set of PBO2 and Curve Optimizer settings. I say "nearly maximal" because my Curve Optimizer per-core settings are only tuned to a granularity of +/- 5. Between the CPU's own PBO2/CO push and the aggressively OCed RAM my CPU score is actually what pushed my score over the top. This can be seen in comparisons with other scores in the top 10.
 
Yesterday I got stuck in the 20600-20800ish range just trying to reproduce the 20943 I'd once gotten right after I installed the Optimus water block on the 3080ti and was playing around with the OC. I did some reading and manually shut down some system services and some other services that were running that I didn't need. Most of these, like the Microsoft Fax service, will stay off forever. Some of these weren't actively disabled, just stopped, and will restart when I reboot the machine. These included services related to software licensing, updating, and some other things from SolidWorks and Adobe. I don't know if stopping or disabling any of those services made much of a difference, though I did find myself climbing back up into the upper range of the scores I was getting more frequently.
 
The real kicker was figuring out that I could restart Steam in offline mode. I had 3DMark started through Steam, so I can't have Steam not running while benchmarking. Precision X1 was also started through Steam. I figured out I could load an OC profile, make any tweaks and apply them that I wanted to in Precision X1, and they'd stick even after I closed PX1 until I rebooted the machine. Once I restarted Steam in offline mode I shocked myself getting that 21000 that I OPed about in this thread. Apparently Steam would do some random network pings of various Steam services during the benchmark runs and it was just enough to cost me 200-300 points in Time Spy. Going offline erased that. I did a couple more tweaks of the VRAM speed and got low 21000ish score, then one more +25MHz VRAM tweak crossed 21109 points for #1 for machines using a Ryzen 9 5900x and a 3080ti from any company. I'm actually kind of shocked about this.

You can take your keys from Steam. Download 3dmark from 3dmark. Input your keys and not need to run steam ever.
 
Last night I hit 22668: https://www.3dmark.com/3dm/70303044
I feel my CPU should be pulling a better score o.O