I also mentioned my 3070 Ti in the post. Those have GDDR6X, to save you the time of looking it up. And the point about the 3060 was to highlight the volatility of VRAM overclocking in general. But anyway, I'm glad to have cleared up your misconception.
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
3080 Undervolting
- Thread starter Psycho Sonny
- Start date
- Joined
- 21 Jun 2006
- Posts
- 38,367
Which is why I specifically said 3060 and I left out the 3070ti.
Also the top 5 scores on port royal off this forum all have memory overclocks of between 0 and 400. Yes someone is in the top 5 on this forum for 3080 score with memory on zero overclock.
https://www.3dmark.com/3dm/57056441?
https://www.3dmark.com/3dm/63342871
https://www.3dmark.com/3dm/60360747?
https://www.3dmark.com/3dm/57256876?
https://www.3dmark.com/pr/442274
Would be interesting to see what this +800 overclock does to his port royal score with a similar core clock but I'm guessing it wouldn't be beneficial at all and possibly detrimental.
The 3080 specifically doesn't gain much from memory clocks if anything judging by the fact someone got into the top 5 with zero memory overclock.
Soldato
- Joined
- 14 Oct 2007
- Posts
- 9,942
- Location
- newcastle
I run my 3080 Zotac trinity OC @900mv 1920Mhz
Soldato
- Joined
- 21 Jul 2005
- Posts
- 21,107
- Location
- Officially least sunny location -Ronskistats
A while ago that would have cost you about £40 so people aren't going to rush out and do that lol! Might have more punters now its dropped in price.
RT does stress test an overclock though so that's good advice!
A while ago that would have cost you about £40 so people aren't going to rush out and do that lol! Might have more punters now its dropped in price.
There are not many PC gamers going to complain over £40. There are 14,288 playing it right now with 21,483 peak over the last 30 days. Certainly wasn't as good as the hype train, but at the same time there was nothing better to play when it released. Graphically, it's one of those titles every PC enthusiast should have.
Associate
- Joined
- 11 Jun 2021
- Posts
- 101
- Location
- UK
No to be fair, first im hearing about it. I always ran mem around 500-1000 and it was fine. Ill adjust my curves and adjust mem later and see if it has an improvement.
I dont play benchmarks or even benchmark any more. Its primary use is Warzone / Modern Warfare.
My goal was a solid 97fps (100Hz gsync monitor) all the time with no crashes, slowdowns or anything else.
3400 x 1440 res, all settings maxed with the dof rubbish all off.
When I played with undervolting and overclocking, benchmarks and some games were fine but Warzone for some reason sometimes in action would either ditch fps then climb back up or crash completely, ruining games. One time was with a slight undervolt, we got to final circle in quads and a vehicle came in, I thermite stuck it and we all opened fire.
There must have not been enough watts to power the gpu at that specific point and my computer flat powered off. I thought it was a power cut at first but it wasnt. It cost us the game and we came 2nd, i almost threw my keyboard out the window.
I immediately benched everything and all seemed fine only for it to happen a few days later. I went to stock settings and its been fine since. Nvidia dont have ‘stock’ settings for no reason, they’re the most stable. Sure overclocking/undervolting is fun but when you play team games you want stability not to care about a slight bit of extra heat or power usage.
Also to add, Quake II RTX was a great tool for benchmarking when i was benching a year or so ago. It really hammered the GPU
Last edited:
Its the way 3D Mark reads the baseclock speed vs the offset people apply in Afterburner. When people talk about +1000 etc, they are going into MSI AB and pushing the memory slider up by +1000. That will not translate in 3D mark as showing an extra +1000 on the memory. Here is a good example of my 3090 I just ran at baseclock and before and after +1000 in MSI AB:
https://www.3dmark.com/compare/pr/1202314/pr/1202308
The +1000 in AB shows up as 1,332 Mhz in MSI AB while base shows memory as 1,219 MHz. My score however does go up very slightly with the memory overclock.
Now not to say you will not get in top 5 without touching the memory as the bulk of the gains come from the core, as above adds less then half a percent for me adding +1000 as the one chap that did in your links for the 3080 had a high core speed, but the rest did overclock the memory and looing at the actual memory speed vs what's in the bracket, they defo added more then +250 in MSI AB.
On my 3080FE I found it tailed off for me around +700. Nothing but gains (albeit small) below that, except higher temps of course. So clearly differs card to card and environment.
For ref, mine is +500 for balance and 20gbps mark.
Last edited:
- Joined
- 21 Jun 2006
- Posts
- 38,367
Gains in benchmarks? or did you specifically test fps? I think the issue might be his ram was thermal throttling then. It would have been nice if he showed his temps. He likely had a card with very bad contact on the memory
Benchmarks, inc Farcry 5 and BL3 in game. Around 700 it consistently gives lower, though. Admittedly I haven’t benchmarked RTX stuff.
Im sceptical of those that run 1000+, I suspect many don’t test it properly and just assume it’s good because as it doesn’t crash. Unless they’re mining of course, just need them hashes I suppose.
Last edited:
- Joined
- 21 Jun 2006
- Posts
- 38,367
Ampere seems vary a lot with silicon quality, mine does not get near stable at some of the values people are using here
That's obvious. But the setting I have posted was taken from the video and it's been tested across both 3080 and 3090 by the YouTube channel.
Watch the video and there is a graph with various settings and voltages you can try as well as showing you what the stock values are.
You can always just run MSI oc scanner. Then use that as a base for undervolting. Whatever speed it finds stable for 887mV just use that.
When I undervolted my 3090 it was constantly hitting the power limit and occasionally crashing unless I supplied it enough volts to make the underclock pretty pointless. Overclocking and raising the power limit made little real world difference apart from more heat and fan noise. As you say, stock works no matter the scenario. It's almost like they have a bunch of quite bright people who figure this stuff out.
Soldato
- Joined
- 27 Jan 2003
- Posts
- 6,108
- Location
- Chesterfield, UK
1830 @ 800mv here on my FTW3. Rock solid and runs silent.
Is everyong using a custom curve or just alterting core voltage?
My original curve -
And after reading the sucess of lower voltages I will start nudging the lower nodes up -
What do your curves look like?
I forgot that I also have a +450 on the memory. I'll need to prove to myself again that it does as intended.
My original curve -
And after reading the sucess of lower voltages I will start nudging the lower nodes up -
What do your curves look like?
I forgot that I also have a +450 on the memory. I'll need to prove to myself again that it does as intended.