basically everyone, everywhere claims that the 4090s memory OCs like crazy.

[GerKNG]

reviews, videos, forums... on average i see between +1200 and +1800 on the memory. some people need the extended memory clock support in gpu tweak up to run +2200 fully 24/7 stable. (with evidence). I have three 4090s and these are my results. (and it does not matter which software or system i use)

Gainward Phantom GS from December: +450 stable, +500 flickering textures and broken geometry, +650 artifacting imediately after applying across the whole desktop, +700 = imediately locking up.

Zotac AMP Extreme Airo from December: +500 stable, +600 flickering textures, +650 desktop full with black squares and full system lockup.

MSI Gaming X Trio from August: +550 barely stable, +650 flickering textures, +700 broken geometry and artifacting on desktop, +800 full system lockup when applying.

how is every review and so many people are often far beyond +1200? three cards in a row perform basically the same is no longer coincidence.

for example the TPU reviews... the highest i can achieve on all cards is~1370MHz + 1390MHz which is lightyears behind everything else.

 

[lsevald]

Are these MSI Afterburner numbers (8xGPU-Z numbers)? My Gainward Phantom 4090 can run +1500, but beyond +1300-1400 benchmarks seems to go a little down or stay the same, and thinking it might be some error correction involved I stick with +1300 for max performance or +1000 in my undervolt profile for less demanding games.

 

[Steevo]

Sometimes reviewers get the ripest cherries.....

 

[GerKNG]

Are these MSI Afterburner numbers (8xGPU-Z numbers)? My Gainward Phantom 4090 can run +1500, but beyond +1300-1400 benchmarks seems to go a little down or stay the same, and thinking it might be some error correction involved I stick with +1300 for max performance or +1000 in my undervolt profile for less demanding games.

should be 8X (as always in every software except of GPU tweak where it's double: 500 in AB is 1000 in GPT)

 

[ir_cow]

You can go really high if you turn off ECC in the control panel.

I should see what the PNY 4090 I have can do. Haven't tried to OC it all yet.

 

[GerKNG]

You can go really high if you turn off ECC in the control panel.

I should see what the PNY 4090 I have can do. Haven't tried to OC it all yet.

ECC is always disabled by default.
500-600 is the absolute limit on all 3 cards.

 

[ir_cow]

ECC is always disabled by default.
500-600 is the absolute limit on all 3 cards.

I thought it was the other way around. Which is how all those 3dmark scores were broken last year. People had to intentionally disabled it.

 

[Space Lynx]

important to note here is the graph showing avg fps, where increased speeds show basically no gains in-game fps... i'd say leave it at stock and don't worry about it

 

[ir_cow]

important to note here is the graph showing avg fps, where increased speeds show basically no gains in-game fps... i'd say leave it at stock and don't worry about it

Correct. GDDR6X introduced with 30 series made memory OC pointless (unless LN2 OC). In fact if you went too high it would cause those errors and actually lower performance.

GDDR6 on the other hand does benefit from memory overclocking.

 

[GerKNG]

well it's heaven benchmark... memory OC does more than core in modern games.

 

[oxrufiioxo]

Mine does 1500 before perfomance regression. It also can do about 2950 ish on the core clock. I think if I where to test outside of a case I could get a bit higher too lazy for that lol....

CP 2077 rt overdrive stable but consumes nearly 600w doing it.

Really though it only gives me like 5-8% more performance vs stock so really not worth it especially when capping the card around 350w gives me nearly stock performance.

 

[freeagent]

No 4090 here but can do +1500 on mems, and core can run at ~3030-3060

 

[Space Lynx]

i ran lots of tests recently on my 7900 xt, and for my card anyway core is still king from what i can gather.

 

[chrcoluk]

As always OC is never assured, if every card could do these clocks, they would have shipped with them.

Take reviewer's clocks with a pinch of salt, a manufacturer will just be too tempted to send them golden samples. They also are time constrained on embargo etc. so dont know how much stability testing has been done.

People on reddit etc. you dont know if they been returning poor samples to retailer effectively doing their own binning process or fibbing, or running unstable, I have a friend who does this every time he upgrades, he buys 5-6 parts, tests them all and only keeps the best one.

Also is there any benefit to overclocking memory on a 4090?

 

[oxrufiioxo]

People on reddit etc. you dont know if they been returning poor samples to retailer effectively doing their own binning process or fibbing, or running unstable, I have a friend who does this every time he upgrades, he buys 5-6 parts, tests them all and only keeps the best one.

I only do this for Coil whine because I can't stand it. I'd rather have a worse clocker with no coil whine. It's been a long time though since I got a card with coil whine.... Pascal era.

 

[Space Lynx]

No coil whine anywhere on my current rig. It's lovely hearing all the quiet

That gigabyte am5 mobo i tried earlier this year had coil whine so loud it still gives me nightmares.

 

[izy]

Does memory OC without increasing any V ( i think it auto adds more V when mem OC? ) adds W that counts to the power limit?

 

[yzonker]

It's just lottery. I game all the time at +1600 on my Asus TUF 4090. Best benchmark scores usually come from the +1700 to +1800 range.

 

[TheoneandonlyMrK]

I wouldn't recommend people try though.

The 4090 is the number one card I see being fixed in videos. So without a wrap around water block I wouldn't personally.

Memory Vrm, zones are notoriously minimalist, YmmV.

 

[JustBenching]

reviews, videos, forums... on average i see between +1200 and +1800 on the memory. some people need the extended memory clock support in gpu tweak up to run +2200 fully 24/7 stable. (with evidence). I have three 4090s and these are my results. (and it does not matter which software or system i use)

Gainward Phantom GS from December: +450 stable, +500 flickering textures and broken geometry, +650 artifacting imediately after applying across the whole desktop, +700 = imediately locking up.

Zotac AMP Extreme Airo from December: +500 stable, +600 flickering textures, +650 desktop full with black squares and full system lockup.

MSI Gaming X Trio from August: +550 barely stable, +650 flickering textures, +700 broken geometry and artifacting on desktop, +800 full system lockup when applying.

how is every review and so many people are often far beyond +1200? three cards in a row perform basically the same is no longer coincidence.

for example the TPU reviews... the highest i can achieve on all cards is~1370MHz + 1390MHz which is lightyears behind everything else.

View attachment 311053

My palit, which is basically the exact same model as Gainward (they are the same company) does 2200-2400 for benchmarks (sometimes with artifacts for example on timespy) and around 1800 for gaming. For 24/7 I run it at 1500mhz with a 70% power limit and +100 on the core.

 

[Belfaborac]

I've got a Phantom GS myself and while I can't remember exactly how far I got during testing, I believe I got it to +1900 and +275 on the core without artifacting or any other indications of instability. Watercooled btw. As far as coil whine goes there is none, at any speed, just as with all other cards I've owned.

I just run it stock though, since nothing I play needs the extra oomph so far in 1440pUW. The highest preset I've saved in Afterburner is +1500/+100, which is as far as I'll ever want to push it in regular usage - and that's if it will still do it when I get that far.

 

[GerKNG]

exactly what i meant. +1200 to +1500 seems to be "average" except of my three cards it seems.

 

[Belfaborac]

Luck of the draw I guess. As long as you get what Nvidia promises the only choice is between being happy with what you've got and flipping the card and trying again.

 

[Bobaganoosh]

I have a 4090FE on a Heatkiller waterblock. I haven't played with it a lot for stability (just got it up and running on water a week ago), but I was pushing memory for benchmarks and got to the point where Afterburner maxed out around +2000 on the memory and it still benchmarked just fine (for timespy, port royal, and speedway). Some of the benchmarks I can run 3060MHz on the core, but others will crash if I go over 3030MHz (even if it just poke 3045 for a second in Port Royal for example). I didn't push past +1300 on the memory while it was air cooled, so I don't know if this is a result of the water-block or if it just always can do this. ECC is off (default) btw. I don't think most games would benefit too much from pushing it that far though, so I'd probably just set it at +1000 or even +800 for games. Stability is more important to me while gaming and I haven't gotten to anything that even pushes this card yet at stock settings.

 

[neatfeatguy]

Not sure what your problem is - not all cards OC the same. Sometimes you get cards that OC well and sometimes you don't.

I had two identical 8800 GTS 512MB cards I ran in SLI. According to GPU-Z, both ran the same BIOS, too. One card I could OC really well, the other card would just crash/BSOD if I OC'ed it at all.

I had two 980Ti AMP! Omega cards. I could crank on the memory for one of the cards and not so much on the other. One card (I forget which OC'ed better) had Hynix for the memory and the other had Samsung. The one that OC'ed really high I could push the memory up from stock 7020 (1755) to 7800 (1950). I never bothered to run the card with that heavy of an OC, I actually kept the memory at 7400 (1850).

Got to love that silicon lottery.