RTX 3090 [H] Owner's Official 3DMark Time Spy Leaderboard

mnewxcv

Supreme [H]ardness
Joined
Mar 4, 2007
Messages
8,005
Your graphics score is around what I was seeing with my 3090 FE running at stock clocks. At stock on all settings, my graphics score is 19,720 and an average temp of 56c. With just the power limit increased to 114% I can get a graphics score of 20,336 and an average temp of 59c.

Pull back your memory clocks some or entirely and see if your score stays about the same or if it improves. Because of power limitations, I find that average core clocks are lower with higher memory clocks since they both share from the same power limit. For me, I noticed core clocks start to dip by 13-17MHz when adding +250 to +500. But the +500 memory did help increase my score by a whopping .5% despite the core clock dip, the +250 made little to no difference. Also, sometimes going too high on the memory clocks can have an adverse effect on performance but the memory ECC keeps it from crashing. So, instead, your performance drops. Those memory chips create a lot of heat when cranked up and could be having an adverse effect as well.

Also, take a look at the undervolting thread on here and look at applying a curve to your overclock instead of a full on offset. It might net you more performance and less heat. I am currently running a curve offset of +150 MHz. With that along with a +500 MHz on the memory and the fans set to 100%, I can get a graphics score of 21,169.

Is 104% the highest power limit your card will allow? I am kind of surprised by that considering it is water cooled. Even the FE allows a power limit of 114%. Of course, what would be a more accurate measurement is peak wattage. I hit around 410w or so at peak when my power limit is set to 114%.

I am also a little surprised by your temps considering you are on a water cooling hybrid and have the fans at 100%. I would have expected it to be lower. What's your ambient room temperature? Are the radiator fans pulling in cool air from outside the case or pulling air from within the case?

With fans at 100% I can keep the 3090 at around 50c when running this benchmark and around 56c-62c if fans are left at default settings. 69-72c is about the highest I see my GPU temps get when fans are at default when playing something like RDR2 for extended periods. My idle temps at night are around 24-26c and 28-29c during the day when in use at default fans.

Your card has more potential. You just need to get your settings dialed in.
yeah, his avg gpu clock is about 150mhz less than my peak run. CPU does affect graphics score too, so he likely won't be able to match higher core CPUs without some magic, but still should be able to get that score up.

I'm starting to think, which I believe was mentioned earlier, that we should probably be comparing graphics scores instead of overall scores to get a better idea as to what each of us are able to squeeze out of the graphics card. I am seeing some people's scores get really skewed because of the CPU they are using.

Also, and this maybe too much of a request, can the OP attach the URL next to the score in the link for easier reference. As this thread gets bigger, it becomes more and more difficult to chase down where the link to the score is so that we can do side by side comparisons.
The cpu affects graphics score as well. The CPU score is strictly CPU, but the graphics score is based on the FPS of the 2 graphics tests, which utilize the CPU. That being said, I do see from looking at results that graphics score does seem to scale better across CPUs. I will update the leader board at some point today, and sure I will throw in links. I wonder if it will disrupt the order of the leaderboard, will be interesting to see.
 
Last edited:

VirtualMirage

Limp Gawd
Joined
Nov 29, 2011
Messages
421
The cpu affects graphics score as well. The CPU score is strictly CPU, but the graphics score is based on the FPS of the 2 graphics tests, which utilize the CPU. That being said, I do see from looking at results that graphics score does seem to scale better across CPUs. I will update the leader board at some point today, and sure I will throw in links. I wonder if it will disrupt the order of the leaderboard, will be interesting to see.
While I am sure the CPU still factors in some for the graphics, I am not sure by how much. I don't know if it is effected more by pure clock speed or if it even utilized large core counts. Looking at UL's technical guide for the benchmarks, this is what they have to say about Graphics Tests 1 and 2:

Graphics tests are designed to stress the GPU while minimizing the CPU workload to ensure that CPU performance is not a limiting factor.

The person with the watercooled GPU, they are running a 6 core 5600X, which I have twice as many cores with my 3900XT, but they are claiming a clock speed of nearly 250 MHz higher than me. So from a IPC per core and clock speed, they have an advantage there if all cores are at 4.7GHz. But when it comes to multi-threading, I still have the advantage.
 

pclausen

Gawd
Joined
Jan 30, 2008
Messages
560
Appreciate the feedback. Yeah, I haven't really had a change to tune my EVGA Hybrid yet. Just did a couple of quick runs. Should have more time tonight. I have about 5 rigs in my office, so it is pretty hot, probably around 78F or so.

This is the setup:

IMG_1307.JPG


The XC3 Hybrid is just a stop gab until I can get my hands on a FTW3 Copper to incorporate into my existing loop.
 
Last edited:

mnewxcv

Supreme [H]ardness
Joined
Mar 4, 2007
Messages
8,005
While I am sure the CPU still factors in some for the graphics, I am not sure by how much. I don't know if it is effected more by pure clock speed or if it even utilized large core counts. Looking at UL's technical guide for the benchmarks, this is what they have to say about Graphics Tests 1 and 2:

Graphics tests are designed to stress the GPU while minimizing the CPU workload to ensure that CPU performance is not a limiting factor.

The person with the watercooled GPU, they are running a 6 core 5600X, which I have twice as many cores with my 3900XT, but they are claiming a clock speed of nearly 250 MHz higher than me. So from a IPC per core and clock speed, they have an advantage there if all cores are at 4.7GHz. But when it comes to multi-threading, I still have the advantage.
OP updated. As of now, all future scores will be focused on graphics score alone, and the leaderboard now reflects that instead of overall score. :)
 

sk3tch

[H]ard|Gawd
Joined
Sep 5, 2008
Messages
1,609
OP updated. As of now, all future scores will be focused on graphics score alone, and the leaderboard now reflects that instead of overall score. :)
Thanks for all of the hard work updating the list and such!! Much appreciated.
 

mnewxcv

Supreme [H]ardness
Joined
Mar 4, 2007
Messages
8,005
Thanks for all of the hard work updating the list and such!! Much appreciated.
You moved up to 4, you have a chance to place on the podium :D idk what kind of sorcery arestavo is up to though.
 

pclausen

Gawd
Joined
Jan 30, 2008
Messages
560
Best I can do with my EVGA XC3 Hybrid is 19,935.

https://www.3dmark.com/spy/16171374

Driver 457.51
Power Limit 104% (that's the limit in AB)
Core +125 Memory +250
Fan Speed 100%

Time Spy crash with core above +125, even with Memory at 0. And with core at +125, Time Spy crash with memory at 250.

Since I have a 1500i, I enabled logging to capture the power usage before and during the above run:

power_before_during.PNG


So the card consumed about 384 watts at the peak (5 second logging interval) minus whatever the CPU was pulling.
 
Last edited:

mnewxcv

Supreme [H]ardness
Joined
Mar 4, 2007
Messages
8,005
Best I can do with my EVGA XC3 Hybrid is 19,935.

https://www.3dmark.com/spy/16171374

Driver 457.51
Power Limit 104% (that's the limit in AB)
Core +125 Memory +250
Fan Speed 100%

Time Spy crash with core above +125, even with Memory at 0. And with core at +125, Time Spy crash with memory at 250.

Since I have a 1500i, I enabled logging to capture the power usage before and during the above run:

View attachment 307806

So the card consumed about 384 watts at the peak (5 second logging interval) minus whatever the CPU was pulling.
We will teach you, grasshopper. There's more in it. Wax on, wax off.
 

VirtualMirage

Limp Gawd
Joined
Nov 29, 2011
Messages
421
Best I can do with my EVGA XC3 Hybrid is 19,935.

https://www.3dmark.com/spy/16171374

Driver 457.51
Power Limit 104% (that's the limit in AB)
Core +125 Memory +250
Fan Speed 100%

Time Spy crash with core above +125, even with Memory at 0. And with core at +125, Time Spy crash with memory at 250.

Since I have a 1500i, I enabled logging to capture the power usage before and during the above run:

View attachment 307806

So the card consumed about 384 watts at the peak (5 second logging interval) minus whatever the CPU was pulling.
Probably a dumb question:

Are you using two separate PCIE cables for your card?

You can get the wattage capture of the card using Afterburner instead of estimating it from your AX1500i.
 

pclausen

Gawd
Joined
Jan 30, 2008
Messages
560
I checked out your custom curve thread, and was able to do +135 at each point of the curve and be stable. My score was slightly less @ 19,893 vs the 19,935 run though.

Yes, I'm using 2 separate PCIE cables. I'll see about capturing wattage using Afterburner. Chilly evening and my idle temp is currently 23C.
 

mnewxcv

Supreme [H]ardness
Joined
Mar 4, 2007
Messages
8,005
I checked out your custom curve thread, and was able to do +135 at each point of the curve and be stable. My score was slightly less @ 19,893 vs the 19,935 run though.

Yes, I'm using 2 separate PCIE cables. I'll see about capturing wattage using Afterburner. Chilly evening and my idle temp is currently 23C.
with the +135 curve up to the stock max boost, set the next incremental point another 15mhz higher. If it doesn't crash, then 30, etc. It can be a bit tedious but that is what tuning is.
 

VirtualMirage

Limp Gawd
Joined
Nov 29, 2011
Messages
421
I'm inching it a little further along. Latest graphics score of 21,309:
https://www.3dmark.com/spy/16230285

I bumped up my curve just a little more, to +165MHz. Otherwise, the rest of my settings are the same as my previous attempt (+500 Memory, 114% Power Limit, Fans at 100%).

I am monitoring my GPU voltage now to see where I sit at. When at 100% Power Limit, I see my voltage go between 800-875mV with the occasional spike to 900mV. When at 114% Power Limit, I see my voltage go between 900-950mV with the occasional spike to 975mV. When reviewing my core clock curve, my peak clock speed tops out at 975mV.

Last night I was able to get the card to run at this clock speed but at 0 memory overclock and default fan speed for over 30 minutes without issue on the Bright Memory benchmark. And repeated runs in Time Spy were no issue too.

This morning I just tried to increase the core clock curve to +180MHz with a 0 memory overclock but Time Spy would just hang at the loading screen. So, I guess I am starting to hit the ceiling of what my card can do. I wonder if this means I cannot make any more adjustments to the curve at all or if I can maybe tweak just the top end and leave the bottom end alone?

Update:

I went back down to +165MHz core clock curve adjustment and everything seems to run just fine. I then tried to up the memory to +700MHz but my score was actually worse than at +500 by .1% to .6% even though my average core clock was about 10MHz faster with +700MHz memory. So I think I am hitting the ceiling there as well.
 
Last edited:

mnewxcv

Supreme [H]ardness
Joined
Mar 4, 2007
Messages
8,005
I'm inching it a little further along. Latest graphics score of 21,309:
https://www.3dmark.com/spy/16230285

I bumped up my curve just a little more, to +165MHz. Otherwise, the rest of my settings are the same as my previous attempt (+500 Memory, 114% Power Limit, Fans at 100%).

I am monitoring my GPU voltage now to see where I sit at. When at 100% Power Limit, I see my voltage go between 800-875mV with the occasional spike to 900mV. When at 114% Power Limit, I see my voltage go between 900-950mV with the occasional spike to 975mV. When reviewing my core clock curve, my peak clock speed tops out at 975mV.

Last night I was able to get the card to run at this clock speed but at 0 memory overclock and default fan speed for over 30 minutes without issue on the Bright Memory benchmark. And repeated runs in Time Spy were no issue too.

This morning I just tried to increase the core clock curve to +180MHz with a 0 memory overclock but Time Spy would just hang at the loading screen. So, I guess I am starting to hit the ceiling of what my card can do. I wonder if this means I cannot make any more adjustments to the curve at all or if I can maybe tweak just the top end and leave the bottom end alone?
Instead of the UV curve, set afterburner to +165 in the main window. Then go into the curve editor, and pull down anything over 2100 to 2100. If it crashes, do 2085, 2070, etc, then once you pass (and likely break previous high score), go back on the curve and start raising the points by 15mhz to the right of the leftmost point on the straight line. As in, if 950mv is set to 2070mhz and everything to the right is as well, leave that alone but set 975mv to 2085, if it crashes set it back and set 1000mv to 2085, etc, until you don't crash. Don't forget to thoroughly cool the card before making changes in AB.
 

blade52x

2[H]4U
Joined
Sep 9, 2006
Messages
3,065
About ~20500 is what I am reaching on my watercooled card. I am power limited (107% on stock TUF bios) so this is about my limit until I flash to a higher one. My undervolted boost clocks fluctuate from 1920-1980mhz during the benchmark. No overclocks on the memory. No overclocks on 5800X either, but I did crank the IF to 1900mhz and lower my DRAM timings to CL14 (3800mhz).

https://www.3dmark.com/3dm/54802430 (20,544 score)
 

talon95

Weaksauce
Joined
Aug 13, 2006
Messages
119
This is my new best. Zotac 3090 w/ Gigabyte bios (390w). Max fans with the side panel off, but no other extreme cooling. 21035 graphics, 19220 total.

https://www.3dmark.com/spy/16556849

Keeping the card quite a lot cooler let me go up +30 more on the core clock.
 

German Muscle

Supreme [H]ardness
Joined
Aug 2, 2005
Messages
6,473
Nice memory OC! My FTW3U had gains up to +1548 but became unstable in anything but Doom Eternal. +1448 seemed nice for every day stuff.
thanks,
Thats as far as i could go with afterburner. The slider wont go any further. PX1 seems to have a couple hundred point loss just from using it but it will allow me to set the offset higher then +1500. I might try and start from ground zero using PX1 and see if itll go any higher If it wont then im good with that as ill just use afterburner instead.
 

III_Slyflyer_III

Limp Gawd
Joined
Sep 17, 2019
Messages
254
Barely touched this thing yet. Evga 3090 FTW3 Ultra with just 107% Power Slider and +130gpu +500mem. Air Cooling.

Still have a ton of pushing to do; but this is a good start for now... :)

21113 Graphics Score. Using the "Hot Fix" nvidia drivers BTW because of HDR.

https://www.3dmark.com/3dm/56068017?
 
Last edited:

noko

Supreme [H]ardness
Joined
Apr 14, 2010
Messages
5,902
EVGA 3090 XC3 Ultra, air cooled, power limit 104, Core +100, Mem +309, custom fan curve
System: TR 3960x PB only, 3733 16,16,16,16,36 32gb 4 sticks, SMT off (makes huge difference in TimeSpy scores)

Graphics score 20675

http://www.3dmark.com/spy/17330621
 

mnewxcv

Supreme [H]ardness
Joined
Mar 4, 2007
Messages
8,005
Barely touched this thing yet. Evga 3090 FTW3 Ultra with just 107% Power Slider and +130gpu +500mem. Air Cooling.

Still have a ton of pushing to do; but this is a good start for now... :)

21113 Graphics Score. Using the "Hot Fix" nvidia drivers BTW because of HDR.

https://www.3dmark.com/3dm/56068017?
I'd rather not mix unapproved driver results into the main air cooled category. Did the january 7th driver fix your hdr issue? That driver should be approved.
 

III_Slyflyer_III

Limp Gawd
Joined
Sep 17, 2019
Messages
254
I'd rather not mix unapproved driver results into the main air cooled category. Did the january 7th driver fix your hdr issue? That driver should be approved.
Yeah it did, I can run another run if you would like... but I figured everyone knew what the hotfix driver was at the time, so I had used it for the tests.

UPDATED with "Offical" drivers now...

Evga 3090 FTW3 Ultra with 107% Power Slider and +130gpu +500mem. Air Cooling.

21077 Graphics Score


New Run with Higher Overclock:

Evga 3090 FTW3 Ultra with 107% Power Slider and +150gpu +1000mem. Air Cooling.

21321 Graphics Score

https://www.3dmark.com/3dm/56702530?

Considering my average clock was 2000Mhz (2100Mhz boosted) and Vram was at +1000Mhz, not much higher than my "average" overclock previously. Certainly diminished return as you go higher.
 
Last edited:

Dayone

n00b
Joined
Jan 13, 2021
Messages
6
Hello,

New forum user and 3090 Aorus Master owner.

https://www.3dmark.com/spy/15991765
Graphics Score: 21632

CPU: 3960x: PBO +200
GPU: +225, +1200

On Air.

Just replaced the 3960x with a 3970x and some better B-die ram. Expecting worse results (3970x is slower for gaming) but going to try and beat that 3960x.
Side note: ended up getting the top score for 3960x and a single 3090 on 3dmark (it's not really a benchmark winning processor though lol).


EDIT: Just realized we care only about Graphics score.

Then my best is 21742 as seen here: https://www.3dmark.com/spy/15976541

Most likely same settings.
 
Last edited:

mnewxcv

Supreme [H]ardness
Joined
Mar 4, 2007
Messages
8,005
Hello,

New forum user and 3090 Aorus Master owner.

https://www.3dmark.com/spy/15991765
Graphics Score: 21632

CPU: 3960x: PBO +200
GPU: +225, +1200

On Air.

Just replaced the 3960x with a 3970x and some better B-die ram. Expecting worse results (3970x is slower for gaming) but going to try and beat that 3960x.
Side note: ended up getting the top score for 3960x and a single 3090 on 3dmark (it's not really a benchmark winning processor though lol).


EDIT: Just realized we care only about Graphics score.

Then my best is 21742 as seen here: https://www.3dmark.com/spy/15976541

Most likely same settings.
you're not supposed to beat me. :D I'll add it to the leaderboard in a little bit!
 

Dayone

n00b
Joined
Jan 13, 2021
Messages
6
you're not supposed to beat me. :D I'll add it to the leaderboard in a little bit!

Ha! Well I'm sure you'll tweak and do better.

Interestingly, the newest NVIDIA drivers seems to have messed with stability at higher clocks >2100 so doubt I'll be going higher anytime soon.
 

Lumpus11

n00b
Joined
Jan 15, 2021
Messages
1
here we can have some friendly competition for achieving the highest Time Spy GPU score. Overclock all you want. I chose Time Spy and not Port Royale because Time Spy is available for free here. Please make sure you are using the latest version.


We are focused on graphics score, not overall! This allows anyone with a 3090 to be competitive since CPU does not play a major role in the GPU score.

The rules are:
1. In your post, please include the score and a link to your validated result, and tell us if it is air cooling, water cooling, or extreme (AC, chiller, dry ice, LN2, hard mods, shunt mods, etc).
2. No double posting. If you would like to post a higher score than your old one, either edit your post or wait for someone else to post theirs.
3. Leaderboard will consist of top 10 scores per cooling type only.



Have fun!

Extreme:
1. 22247 (Falkentyne, shunt mod, unapproved driver) https://www.3dmark.com/spy/15397572

2. 21462 (NukeDukem, chilled air) https://www.3dmark.com/3dm/53349030
3.

Water Cooling:
1. 21425 (German Muscle) https://www.3dmark.com/3dm/55355577?
2. 20544 (blade52x) https://www.3dmark.com/3dm/54802430
3. 20251 (pclausen) https://www.3dmark.com/spy/16175040
4. 19328 (AP2) https://www.3dmark.com/spy/16167167

Air Cooling:
1. 22484 (arestavo) https://www.3dmark.com/spy/15580398
2. 21888 (Kenpomasta) https://www.3dmark.com/3dm/53448394
3. 21724 (Dayone) https://www.3dmark.com/spy/15976541
4. 21702 (mnewxcv) https://www.3dmark.com/3dm/54245955

5. 21530 (sk3tch) https://www.3dmark.com/spy/15120662
6. 21430 (lobstar) https://www.3dmark.com/spy/15623809
7. 21321 (III_Slyflyer_III) https://www.3dmark.com/3dm/56702530?
8. 21309 (VirtualMirage) https://www.3dmark.com/spy/16230285

9. 21035 (Talon95) https://www.3dmark.com/spy/16556849
10. 20892 (Monstieur) https://www.3dmark.com/spy/15111551
https://www.3dmark.com/spy/17048372
Its air cooled.
Score is 21.332 :)
 
Top