RTX 3090 FE scores lower than reviews

[BlueDragon94]

Hi All,



So I recently finished my build and got some time yesterday to play around with my PC. I noticed right out of the box, my RTX 3090 FE was underperforming compared to the benchmarks/scores people have posted to forums. Here my PC specs:


CPU: Ryzen 5900X (Stock)

Cooler: NZXT X63

Ram: GSkill Tidentz NEO 3600MHZ

SSD: Samsung 1 TB EVO PLUS

GPU: RTX 3090 FE (Installed in PCIE first slot with Gen 4)

Motherboard: Asus Crosshair VIII Hero X570

PSU: Corsair RM1000X

Case: NZXT H710


I change my GPU's power management to: "Prefer high performance" and results improved a bit, but I am still not able to achieve 60 FPS 4k lock in games where I should.


Here is the timespy's best result (graphics only) with a very aggressive fan curve: 19,165

With out of the box setting: 19,020

Link: https://www.3dmark.com/spy/15559531

Ac:O

Setting: 4k, Ultra high

Issue: FPS drops to 54 during open world gameplay

Control:

Setting: 4k, DLSS: 1440p, RTX - High, Graphic- High

Issue: FPS drops to 50 during any action/new area


I am not sure what the issue is.... max temp for GPU is 70 with default fan curve... Is my card defective? Should I RMA?
During runs, the max power draw is 350 W whereas in reviews I see it's around 365 W and my PerfLimit reason is "POWER".


Thanks for your inputs!

 

[BlueDragon94]

350 watts is the default power limit for the 3090FE. You can increase it to 400w using MSI Afterburner.

Most people using benchmarks will be overclocked in timespy tests, few rarely run it at stock which is what you seem to be doing. I can see that your average clock speed is 1770 Mhz with peaks of 1830 Mhz which shows its performing above the clock speeds expected, therefore your card is working fine and boosting above the specs listed on the box. As mentioned, if you dial into the results people post on these forums, they usually tweak the cards using say MSI AB to increase power target / clock speeds etc so those numbers are not always representative of prformance out he box. Usually also do some other tips and tricks to eek out every bit in those benchmarks. If you check reviews from the normal places, timespy graphical performance is around 19500 - 20000 out the box with these on graphical score, abit with a overclocked CPU. so your down a couple of %, but margin of error + some of them using open bench's which will let card run slightly cooler = slightly higher boost clock speed.

You mention few games. Assassin's creed Origins on ultra high, not surprised with that performance and about right when you select ultra high. Some settings in those games just tank FPS when set to the highest settings. Check some guides. Even on newest AC game a 3090 will average about 60FPS at 4k.

Control, not played too much of it and my 3090FE is RMA's right now (HDMI 2.1 broken) so can't provide numbers, but may want to tweak DLSS to 1080, recall did not see much visual impact with that when RTX was on at 4k.

Thanks for the information!

I get that a lot of folks have overclocked GPUs when they post the timespy scores, hence I am not looking at them and just comparing against reviews which have stated that they run stock settings. My clock speeds are at lease 100 MHZ behind the review units, is this solely because of temps?
Also, the reason I mentioned that my power consumed is 350W only is because in these reviews I see the total power consumption crossing 360W when they say they are at stock. Here are few links:
https://www.guru3d.com/articles_pages/geforce_rtx_3090_founder_review,7.html
https://www.guru3d.com/articles_pages/geforce_rtx_3090_founder_review,27.html

Is there any setting that I am missing?

 

[BlueDragon94]

350 watts or round abouts is normal stock for FE does not mean you cannot go +/- by few %, so link you mentioned is showing a peak of 360 watts would be entirely normal and not worrying to see. Them hitting 360 watts for a second and yours sticking to 350 watts does not indicate your card is broken. Next your clock speeds as mentioned are normal from the timespy extract you provided. In Fact at 1771 average with heavy timespy and it well beyond the box specs so if you send the card back they would report there is nothing wrong with it.

Now as to why it may be behind other reviews, could be down to there benchmark running cooler which does affect boost clock speed, silicon lottery etc. It can vary. some reviewers see around 19500 score out the box so couple of % higher then yours. Some of that could be down to other tweaks in the system, etc.

You can see on the very benchmark you provided, you can compare results with similar systems is about where you would expect 545 systems scoring similar numbers to you. 15600-18199. Next block up where most are is 936 systems in the 18200 - 20799 block which no doubt is where most people have overclocked things.

So TLDR, your clock speed behaviour is above box speeds. your clock speed may not as high as some samples out the box some reviewers got, that's silicon lottery. Your 3D mark scores are within the norm for stock compared to others with similar systems. If you are worried, can send it back, but imagine they will report back there are no issues with it. Can try messing with power target / fan speed / offset if you want to extract more performance.

Ok that makes sense. I understand your point that the card is performing better than the advertised speeds so no point in sending it. It is just that I was really hoping to play AC:O in 4k Ultra at stock settings. I was able to get the rsults reviews get by raising the power limit to 114% but not sure how safe that is.
I have seen only one reviewer with benchmarks close to mine: https://www.igorslab.de/en/nvidia-g...-and-common-decadence-if-price-is-not-all/17/

 

[BlueDragon94]

Raising power limit is fine, these cards are pretty well limited in terms of getting them into dangerous territory with something like MSI Afterburner alone. Similarly the 3090 FE cooler is pretty solid, so temps kept in check usually.

Its unfortunate not seeing games maxed out getting a specific FPS, but unfortunately just a function of these sorts of games! Assassins creed since reboot is one of those and am well used to it pouring in 500 hours in Origins, Odyssey and Valhalla. But do note, in these games there are some settings you can tweak for little visual impact for a nice gain in FPS. Some settings just hurt FPS for no real improvement. https://steamcommunity.com/app/582160/discussions/0/2860219962082106581/ could try checking ingame settings in this guide and tweaking.

Thanks a lot for your inputs

 

[BlueDragon94]

You are also forgetting the reviews were done on the drivers that allowed higher power and clock speeds, but also caused crashing and NVIDIA fixed the drivers and lowered some of the performance to stop the huge spikes in power use and frequency.


New drivers you will be on and everything you have shown here looks about right. So nothing to worry about, some of the games you mention will not do 60fps on even a 3090 with 4k ultra settings, time to go SLI mate if the games you play support it.

Yup I did suspect drivers being an issue here so did roll back two versions back but nothing really changed so dropped the idea.
I think the scores look ok. One more factor is that I live in a country which is hotter than the temps found in US/CA where most of these reviews occur, so temps could be a factor here.
About SLI.... I think it is such a niche thing now, I don't ever see me going that route. I had always wanted to though

 

[BlueDragon94]

If it's a new system might also be worth looking at other parts like the CPU, how does that bench in cinebench? Not sure if needed on 5000 series but 3000 certainly needed amd chipset drivers for the correct power management settings etc. Also ram, make sure its got it's xmp profile set.
Might already be in place but you never know

Also, who did you have to bribe to actually get your hands on both a 5000 series cpu and 3000 gpu?!

Yup that was the first thing I checked, CPU is benching higher than reviews in Cinebench and RAM is on DOCP.

To get a 5000 series CPU, I had contacted a local vendor 1 month ago and asked him to let me know as soon as stock arrives. For RTX 3090, I was waiting for 3080 FE for a long time which didn't come in stock and only the 3090 FE was available from official NVIDIA vendor so got that instead. Didn't pay a single cent over MRP for both.

 

[BlueDragon94]

My 3090 Fe should land tomorrow so i can post some results to see how it stacks, I don't think you have anything to worry about as others have said.

That would be very helpful, thank you. I have one question, just out of curiosity not concern. Is it possible the difference is showing because all the reviewers have used a Intel core i9 clocked at 5Ghz while mine is max 4.7 Ghz. I am seeing only the graphics score and not the CPU score so I don't think it should matter much but still, I am not sure.

 

[BlueDragon94]

For sure, or memory type, speed, timings, bios on the MB, drivers on the system. Could be a million things to be honest, but you sound a little let down? I wouldn't be its an amazing card and performance even the numbers you have. I mean there are about 30 million Geeks all crying into the waifu tonight because they cant even get a card and prob wont this side of April 2021.

That makes sense! Yeah I am a little let down as I was not sure if my new card that I got after soo much of effort was ok or not. It looks ok thought so I am not so worried anymore now!

 

[BlueDragon94]

I had a look, and your boost speed seems a little low My average cpu speed was 1840 on my run, try upping your gpu core clock by 100 90 mhz, I take it you have the power limit at 100%?
Also have a look In Nvidia Control Panel > Manage 3D settings> Texture filtering - Quality and set that to High Performance
also while you are there set Power management mode to Prefer maximum performance
See if that makes a difference

Yeah boost speed was the main cause for the lower scores. My power limit is at 100%, I did try to increase it to 114% so see what can my card do and I was running 1870 Mhz stable avg clocks. I also offset the core clock by 90 Mhz and it was stable and I was able to get the scores shown by review units. When I tried an aggressive fan curve, I got this result: https://www.3dmark.com/spy/15396046. I live in a country with a higher ambient temp so even that could be a reason.
I tried both high perf and default settings, I am getting almost same numbers. One strange this, in Assasin's creed Odyssey, my average clock speeds are around 1940 Mhz.

 

[BlueDragon94]

Your clock speeds look solid bud.

Not surprising, some stress tests can place a different load on these GPUs which can change the clock speed. Not surprising to see in heavy GPU benchmarks for your clock speed to be lower then a game.

Thanks! This is one of the most helpful and friendly communities I interacted with for this query. Thanks everyone for your inputs, I will definitely stick around on this forum!