![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Yikes, I'm only running +130 Core too... man, maybe it is a hardware issue then and I got lucky? What happens if you run stock on the FTW3? I think even stock my average clocks are higher than 1910Mhz... lol.
I'm on a FTW3 Ultra, and I do not have the issue. But I think the FTW3 is all the same hardware but different boost clocks in the vbios?
My 3090 XC3 with an undervolt/overclock of 1995@950mV (+135) maintains an average clock speed of ~1940 MHz in Time Spy (score was 20.9K I believe, close to 21K). And once again, that is while pulling an average of 360W, significantly less than the FTW3. I’ll need to double check as I have the result somewhere.
If your PCIe hits 78.9w your card will go into PerfCap PWR mode and heavily restrict the boost table. You can sorta manipulate this with very high clock offsets. If I offset with +360 I can see 2000 boost clocks with 84w PCIe draw until it crashes at a lower power scene. I've never seen average clocks in time spy over 1946mhz on either of my cards once they perfcap.
https://hardforum.com/threads/nvidia-rtx-3090-discussion.2001549/post-1044871281
There are people on the EVGA forum in the XOC BIOS thread that don’t have issues either. But there are still a LOT of FTW3s in that same thread that have the issue. I believe EVGA even said they recognized the load balancing issue and were looking into it. But that was back in October I think and now it’s January. They have been silent since.
China 2012 serials seem to have much better luck.
I believe Taiwan is 2014 and China is 2012. Mine is a 2012 China with Black Lips.
Corrected my post. Both of mine are Taiwan 2014s.
It’s basically a lottery at this point and my FTW3 lost...miserably.
Yeah, I bet it has to be a hardware defect as the vbios is the same. Does have me slightly worried though something may go wrong at some point, but the same was true with my 2080Ti when all those issues were happening and that card worked great for me over 2 years.
Who knows with this generation of EVGA cards. Quite a few FTW3s are suffering from load balancing issues and some are just randomly dying after working completely fine for days/weeks/months. Both at stock and overclocked. The cause is unknown to the public so we don’t know if only some FTW3s are affected by a defect or if all FTW3s are affected and are essentially ticking time bombs. For as good as EVGA has been with their customer service, they have been pretty quiet about the load balancing issues and random “red light of deaths”. If my XC3 dies I’m getting it RMA’d and selling the replacement to get a different brand this generation. But I hope it doesn’t come to that.
Ohhhh great....
Well crossing my fingers then for the both of us... lol. I guess worst comes to worst, I could RMA the 3090 FTW3... and I'm glad I kept my 2080Ti as a backup to use if that ever happens. I'd probably keep the RMA card though if I had too as hard as these cards are to get right now. The performance (for me anyway) is a must at 4K. My 2080Ti did pretty good, but running RT at the highest settings proved hard to maintain at 4K on many games, were i can fully max things out now with no problem.Sounds like for me keeping the XC3 totally stock would be the best choice. I just don't see the more expensive cards giving a meaningful gaming experience uplift. Shunt modding looks relatively easy on the EVGA cards, I would just do the two pcie connectors on XC3 shut resistors using conductive paint and some hot glue except I think that would be virtually a waste of time now. Going after meaningless 3dMark numbers from my view is pointless. Putting a water block or using an EVGA Hybrid kit would probably give the most performance uplift. Unless EVGA puts out a higher power limit, it stays stock for now.
Honestly, I’ve been the most satisfied with my 3090 XC3 by undervolting it. With Ampere you typically see more consistent high clock speeds and frame rates with undervolting.
How TF do you undervolt these things properly in PX1? Or do you all use something else? Every time I try, the damn curve only holds as long as I do not touch anything else. As long as it seems to take to set, I just give up and throw everything at full value and see how far it goes till it crashes...
I use the VF Curve in MSI Afterburner (accessed by pressing Ctrl + F). A Hardforum user created a nice guide to help people with undervolting these cards:How TF do you undervolt these things properly in PX1? Or do you all use something else? Every time I try, the damn curve only holds as long as I do not touch anything else. As long as it seems to take to set, I just give up and throw everything at full value and see how far it goes till it crashes...
Also, I don't think more power does much for these anyway beyond slightly higher clocks as you mentioned. The largest denominator is temperature. If you can keep it cool, the boosts stay higher.
Going from +500 to +1000Mhz on the memory netted me less than 0.5% increase in Timespy and almost nothing in actual game benchmarks in terms of FPS. As for Core Clocks, trying to push above 2000Mhz did not do a damn thing other than create more heat.
Man, thanks for pointing that out... same idea works for PX1 (just tried it on my "break"... lol). I had my clock pretty much LOCKED at 2040Mhz letting Valley just loop. Temps were slightly lower, which is probably why the boost held.
Its the way to go with Ampere. I prefer having more consistent sustained clocks than having the core clock jump all over the place so it can boost 50-100 Mhz higher for a split second.
(Did get a top 95% in the Timespy Upgrade benchmark test though so that's something at least)
Oddly, while the clocks were locked higher, by dipping less and raising less, it made 0 difference in benchmarks or game benchmarks... lol. So although its annoying to watch, nvidia boost must know what its doing?
The thing with benchmarks is that they just give you a result roughly based on the average performance. That’s why you won’t see much of a difference in the scores. If without the undervolt you jump between 1900-2000 with random spikes to 2100 but with the undervolt you’re locked at 1950 then the benchmark results will be nearly the same. But when playing some games I notice the increased frame rate consistency from locked clock speeds as well as slightly reduced thermals. I’ve noticed in some games Ampere has a habit of bouncing off the power limit (could be why it’s less effective on your FTW3 since it has a higher PL) so that’s where the inconsistency in clock speeds and frame rate can come from.Oddly, while the clocks were locked higher, by dipping less and raising less, it made 0 difference in benchmarks or game benchmarks... lol. So although its annoying to watch, nvidia boost must know what its doing?
Wish I could just tell it to run max voltage at a clock all the time in a game. Like, 70c is no where close to the limit, just let me sail man!
Yeah, I played some BFV tonight, but just used my standard "balls to the wall" OC for it... was not sure I had my "fixed" OC stable yet. Might have to try it tomorrow evening and see if it keeps a more consistent frame rate. I kept looking at GPUz though and my clock was pretty much 2025Mhz in BFV the entire time. Would sometimes bounce down to 1960Mhz or as high as 2040Mhz... but did not bounce nearly as bad as Timespy does. Hard to say if the effort will really be worth it for the fixed voltage thing... but I guess I will ensure stability first then try testing it in some games.
Are you on a 3090 FTW3 or FTW3U? If so, it does not void your warranty or anything critical. I had mentioned that it was one of the various bios(es?) that I had tried in my troubleshooting and overclocking attempts in my tickets to support. No pressure, but you can always flash it back since you have the dual bios switch. I can link the info on how to recover if you want to do more research.
I have a FTW3U... I saw the original evga thread from October for the 500W bios... is that the latest version of the bios or has there been any updates since?
Sadly it’s the only one they’ve released since then. There is apparently a new one coming “soon” according to the EVGA Germany forum but it may be updated just for resizable BAR support and no other improvements. We shall see.
It already does worlds better than my 2080Ti did!Well you lucked out to at least not be impacted by the load balancing issue. The biggest reason to get the FTW3 over the XC3 is for the significantly higher power limit. If you pay $200 more for a FTW3 that suffers from load balancing issues and performs about the same as an XC3 then you feel scammed.In the days of easy GPU buying, I'd say keep going back to the store till you get one you like! lol. Those days seem to be long over... I was just happy to snag a card period (even if it cost more). I'm not gonna squabble over 2~3% performance in the end (if that).
Yes. And technically evga does not have to replace the older cards unless you require an RMA from a dead card. evga simply needs to claim, does it hit the 1800Mhz guaranteed without crashing? If so, then you technically got what you paid for.
That's not entirely true. If your EVGA card is exposing other hardware to possible failure you have the right to get a product that works properly. Since this card advertises it's compatibility with PCIe Gen4 x16 in it's specs one can assume it follows the rules of the specification and thus purchase other hardware accordingly. It is very easy with the affected cards to generate a situation where the power draw over PCIe continuously exceeds 75w at stock clocks/power limits. I just cannot stress enough how bad this issue is. It's stressing the weakest part of the power delivery system in the card just as much as one of the three 8-pin power connectors. The 8-pins are rated for at least 150w; why stress the PCIe power so much? We're just lucky the default safety settings are about 80w so most mid to high end motherboards / PSUs should be ok at only 7% increase power draw. I really worry about people on the Kingpin bios with these cards ...
