RTX 4090 Port Royal Scores - Post your Runs!

III_Slyflyer_III

[H]ard|Gawd
Joined
Sep 17, 2019
Messages
1,240
As the title says, I am curious what everyone is pushing on the Port Royal benchmark with these 4090's. I noticed a jump of about 1000 points on this test simply by overclocking the memory from 21GHz to 23GHz... core made very minimal difference for me. Seems to land where I should be, but would be fun to compare, and compete?

Score = 27109 | 125.51 fps

http://www.3dmark.com/pr/2134184

With OC and Re-Bar Forced on...

Graphics score = 28028
Graphics test = 129.76 FPS

http://www.3dmark.com/pr/2142490
 
Last edited:
https://www.3dmark.com/pr/1902482

I seriously don't get how people are getting well past 28000 in this. My average clock frequency is pushing right up against 3Ghz and I'm still in the low 27000s. Unless memory is what really matters for this specific benchmark.
 
https://www.3dmark.com/pr/1902482

I seriously don't get how people are getting well past 28000 in this. My average clock frequency is pushing right up against 3Ghz and I'm still in the low 27000s. Unless memory is what really matters for this specific benchmark.
Possibly? I have noticed a jump in a lot of game tests too by upping the memory speed, not so much with the core. Some people out there are getting +2000Mhz offsets (which would be 25Ghz) easy. TBH, I have not tried to push past +1000Mhz yet, maybe I should? lol. It feels so unrequired for games though, but the thought of leaving any performance on the table just bugs me deep down.
 
Possibly? I have noticed a jump in a lot of game tests too by upping the memory speed, not so much with the core. Some people out there are getting +2000Mhz offsets (which would be 25Ghz) easy. TBH, I have not tried to push past +1000Mhz yet, maybe I should? lol. It feels so unrequired for games though, but the thought of leaving any performance on the table just bugs me deep down.

Wow 2000Mhz? That's pretty insane. I forgot how much I increased mine by but it was nowhere near 2000, I think around 1200 or maybe 1400 and going higher than that value for me resulted in lower scores so I had to dial it back. Otherwise though I am super happy with this FE model. It's by far the best reference design Nvidia has made, runs cool and quiet with minimal coil whine.
 
Wow 2000Mhz? That's pretty insane. I forgot how much I increased mine by but it was nowhere near 2000, I think around 1200 or maybe 1400 and going higher than that value for me resulted in lower scores so I had to dial it back. Otherwise though I am super happy with this FE model. It's by far the best reference design Nvidia has made, runs cool and quiet with minimal coil whine.
Yeah, I might try to push more just to find the memory limit. It seems this generation has a bit of headroom on that front with some halfway decent gains to be had from it as well. Last generation OC'ing did not do a whole hell of a lot on either the core or memory front. It seems this GPU benefits from the increase in memory speed much more than recent generations.
 
There are a couple of tweaks that make big gains. One is the same as 30 series, which is to force reBar with Nvidia Profile Inspector. The other is to flash the Galax HOF bios. That bios holds significantly higher effective clocks than any of the other stock bios. Both of these tweaks are worth 400 pts or so. Some saw even more with the Galax bios. So that's almost 1000 pts right there.

https://www.techpowerup.com/vgabios/252087/252087

Then there are some golden bin cards floating around that can complete PR with 3100+ core clock.

In my case with a TUF, it's not golden bin by any means. I'm just using other means to get there. Watercooled with a chiller and my latest run I used Elmor Labs EVC2 to bump the core voltage to around 1.15v. This run was on the Asus 1kw bios though which is on par with the Galax bios, but higher power limit.

https://www.3dmark.com/pr/2122080

But there are a few cards that can score slightly higher than that without watercooling or volt mods. Problem is that gains are relatively small with core clock on the 4090 because it's much more memory bandwidth starved than 30 series.
 
Last edited:
I did not even think for forcing ReBar... interesting. So far looking at these scores, they seem to go up in tandem with memory overclocking... which is what I was observing in my own tests. Might have to see how high I can push the memory. I also read these memory modules prefer to run warm as well.
 
I did not even think for forcing ReBar... interesting. So far looking at these scores, they seem to go up in tandem with memory overclocking... which is what I was observing in my own tests. Might have to see how high I can push the memory. I also read these memory modules prefer to run warm as well.

Memory and core both contribute in PR. Every bin (15mhz) is around 50-75pts. Memory is maybe 50pts for every +100, I can't remember the exact numbers when I tested. Something like that. Speedway OTOH is very heavily memory bottlenecked. Every bin on the core is only like 20pts.

Yes the VRAM chips need to be 50C or a little more to maximize the OC. I've had to fight that while using a chiller. I literally have a heater on the backplate to heat the mem through the PCB. Even with that I lose 50-100 on the mem with water temps below 10C. I really need to remount with crappy pads.
 
I assume like previous generations I can max out the voltage slider and be fine in afterburner? Default is 1.05V, I noticed maxing out the slider gives me 1.1V and higher boosts. I assume these cards are rated for at least that 24/7; of course, as long as heat is managed and all (but I'd guess the vbios handles throttling anyway if that was not the case).
 
I assume like previous generations I can max out the voltage slider and be fine in afterburner? Default is 1.05V, I noticed maxing out the slider gives me 1.1V and higher boosts. I assume these cards are rated for at least that 24/7; of course, as long as heat is managed and all (but I'd guess the vbios handles throttling anyway if that was not the case).
Yea that won't hurt the card at all.
 
Memory and core both contribute in PR. Every bin (15mhz) is around 50-75pts. Memory is maybe 50pts for every +100, I can't remember the exact numbers when I tested. Something like that. Speedway OTOH is very heavily memory bottlenecked. Every bin on the core is only like 20pts.

Yes the VRAM chips need to be 50C or a little more to maximize the OC. I've had to fight that while using a chiller. I literally have a heater on the backplate to heat the mem through the PCB. Even with that I lose 50-100 on the mem with water temps below 10C. I really need to remount with crappy pads.

I don't think it's that simple. Looking at the difference in core and mem clock frequencies (and even temperature as well) between my score and LittleBuddy's, there is a difference of 41Mhz in avg core and 62Mhz in avg memory. By your calculations every 15Mhz on the core should be +75 points and every +100Mhz on the memory is +50 points so he should be getting +225 more points for the core and we'll just say +100 more points for the memory. So his score should be no more than 400 points higher than mine and yet its higher by 1300. Even if we account for ReBar adding in another 400-500 I'm still well below the extra 1300 points. I don't think a bios flash is it either since our avg clocks are barely that far apart. Pretty crazy how less than a 100Mhz difference in core and memory frequencies result in such a big difference.


1676058948746.png
1676058609807.png
1676058958902.png
1676058645374.png
 
I don't think it's that simple. Looking at the difference in core and mem clock frequencies (and even temperature as well) between my score and LittleBuddy's, there is a difference of 41Mhz in avg core and 62Mhz in avg memory. By your calculations every 15Mhz on the core should be +75 points and every +100Mhz on the memory is +50 points so he should be getting +225 more points for the core and we'll just say +100 more points for the memory. So his score should be no more than 400 points higher than mine and yet its higher by 1300. Even if we account for ReBar adding in another 400-500 I'm still well below the extra 1300 points. I don't think a bios flash is it either since our avg clocks are barely that far apart. Pretty crazy how less than a 100Mhz difference in core and memory frequencies result in such a big difference.
It's likely because I have a 5800X3D. My score beats out everyone on the 5800X Leaderboard but is 57th on the 5800X3D Leaderboard. You can check on the results by system page. https://www.3dmark.com/search

When I ran the test my CPU was also on eco mode, so likely I can increase my a little score further.
 
It's likely because I have a 5800X3D. My score beats out everyone on the 5800X Leaderboard but is 57th on the 5800X3D Leaderboard. You can check on the results by system page. https://www.3dmark.com/search

When I ran the test my CPU was also on eco mode, so likely I can increase my a little score further.

KickAssCop also has a 5800X3D and his clock speeds are very close to mine (difference of 49Mhz on the core and 24MHz on the mem) and we both basically have identical scores so I don't think its a difference of 5800X vs 5800X3D.
 
KickAssCop also has a 5800X3D and his clock speeds are very close to mine (difference of 49Mhz on the core and 24MHz on the mem) and we both basically have identical scores so I don't think its a difference of 5800X vs 5800X3D.
Well then it could be down to system optimization, I play a lot of competitive games so I heavily tweak my OS to reduce latency, that could also play a role. Also what PowerLimit the card is running at, I ran my test at 133%.

Let me know your GPU settings and I'll match it and rerun the test.
 
Last edited:
Well then it could be down to system optimization, I play a lot of competitive games so I heavily tweak my OS to reduce latency, that could also play a role. Also what PowerLimit the card is running at, I ran my test at 133%.

Let me know your GPU settings and I'll match it and rerun the test.

It's likely related to this. I have 2 Win11 installs. One is my bloated up daily driver, the other is a bare bones install just for benching. The bench OS scores quite a bit higher (like maybe 500pts).

It all adds up. The two big gains I already mentioned, but there are quite a few other things that can effect the score. CPU and memory latency has some effect, setting the desktop to 1080p, etc...
 
I think my next goal is to see how high this memory can go before artifacts! Anyone know the max safe temp of GDDR6X off hand for these modules? I'm sure shit throttles anyway, but I'd also hate to damage a small fortune of a card... lol.
 
Just picked up my MSI Gaming X today and started messing with it. 27346 with +175 core and +1300 memory. https://www.3dmark.com/pr/2137680
Interestingly I can go +1400 memory, but only if the core is lower. I think maybe I am running into a power wall since this card has only +6% max power limit? Card has some coil whine, nothing bad, but worst I have had on any card recently.
 
Last edited:
I think my next goal is to see how high this memory can go before artifacts! Anyone know the max safe temp of GDDR6X off hand for these modules? I'm sure shit throttles anyway, but I'd also hate to damage a small fortune of a card... lol.
GDDR6X starts throttling somewhere around 110-120c, if I recall, I doubt it will get that hot on a 40 series card.
 
Oh FFS!!!!! I just discovered when playing Forza 5 and watching Gpuz that my fans have been locked at 30% the entire freaking time cause MSI Center description is f*cking backwards!!! Now I am back to software control of fans through afterburner and I'm holding 3015Mhz Core and +1250Mhz memory and GPU is not evening getting above 50C... lol. Going to have to rerun ALL of my benchmarks... good lord........
 
Oh FFS!!!!! I just discovered when playing Forza 5 and watching Gpuz that my fans have been locked at 30% the entire freaking time cause MSI Center description is f*cking backwards!!! Now I am back to software control of fans through afterburner and I'm holding 3015Mhz Core and +1250Mhz memory and GPU is not evening getting above 50C... lol. Going to have to rerun ALL of my benchmarks... good lord........
:LOL: :LOL: Oops! :)
 
Temps have a huge effect on memory OC for sure... tried 1400+ while it was idle at like 35C and I got space invaders and an instant lockup on windows desktop... guess I did not win the memory silicon lottery. It worked after it it 52C tho, but obviously not entirely stable. :( It seems for me, gaming stable is around 1200+ at idle temp. Oh well, better than nothing i guess!
 
Best I got while tinkering is 28403

https://www.3dmark.com/3dm/87489901?

Wouldn't call it game stable, but I got the score with 250+ core and 1750+ memory.

Lowered the oc to 210+ core 1700+ memory and have run it for hours playing TW warhammer no issue.

$1800 is way too much for a consumer GPU, but can't deny the 4090 is an absolute monster
That 13900K really helps it seems.
I thought this was only a GPU bench.
 
29171

https://www.3dmark.com/pr/1947805

Inno3D X3 OC with Galax 666W and under water, ReBar enabled. Sustained clock was 3090 MHz and memory is +1250. Daily settings are 3060/+1100, sometimes even +1000, because this memory really doesn't like cold water, score basically right at 29000.

With the right BIOS, cooling and ReBar enabled pretty much any average sample can break 29K, to go above that you need +1500 or higher memory bin and good core at the same time, so it is rather unlikely.
 
29171

https://www.3dmark.com/pr/1947805

Inno3D X3 OC with Galax 666W and under water, ReBar enabled. Sustained clock was 3090 MHz and memory is +1250. Daily settings are 3060/+1100, sometimes even +1000, because this memory really doesn't like cold water, score basically right at 29000.

With the right BIOS, cooling and ReBar enabled pretty much any average sample can break 29K, to go above that you need +1500 or higher memory bin and good core at the same time, so it is rather unlikely.
I have a 4090 inbound as my first team green GPU since the GTX 970. With my 6900XT, SAM/ReBar is automatically active so long as its on in the BIOS. Is this not the case with Nvidia?
 
I have a 4090 inbound as my first team green GPU since the GTX 970. With my 6900XT, SAM/ReBar is automatically active so long as its on in the BIOS. Is this not the case with Nvidia?
It isn't universally beneficial on Nvidia, so it's on a whitelist that can be edited.
 
I have a 4090 inbound as my first team green GPU since the GTX 970. With my 6900XT, SAM/ReBar is automatically active so long as its on in the BIOS. Is this not the case with Nvidia?
It is not, it is disabled by default and only enabled by NVIDIA in drivers for games that they tested and whitelisted. You can enable it in NVIDIA Inspector though, either globally or per game. It is like a more advanced control panel that gives you access to various settings that are not exposed in NVIDIA Control Panel. You will need it anyway for other things, like SGSSAA.
 
Back
Top