Overclock.net banner
2441 - 2460 of 2532 Posts

·
PC Evangelist
Joined
·
47,600 Posts

·
Invalid Media
Joined
·
4,400 Posts
Did you read my comment? I said that it would be strange if it was indeed a reference card on air. But the pic you just posted. It looks like they adjusted the FCLK on that card. If they made the right tweaks, and they are using the latest driver, I could see it. 3DMark doesn't always report the correct clocks. This is especially true if there is a large variation in clock speed during the test.
...and in addition, he has a decent PL and apparently is undervolting a bit which also helps, subject to having a good chip and decent cooling
 

·
Registered
Joined
·
92 Posts
...and in addition, he has a decent PL and apparently is undervolting a bit which also helps, subject to having a good chip and decent cooling
Yeah his GPU is highly tuned. He optimized his clocks, PL,TDC, SOC, FCLK, adjusted his min and max SOC voltage, his min GFX voltage was increased...etc. If that was with the latest driver and a cold enough ambient temp, I could see it being legit.
 

·
Registered
Joined
·
393 Posts
Clearly you guys aren't paying attention to the screens.

First of all, he has deepsleep disabled (all the DS check marks) which means nothing on the card will downclock to idle (clockspeed, infinity fabric, etc).

Second of all, power limit numbers are meaningless. You can set them all to 1000 and watch your scores be the same as if you set them at 450 because your card will not hit them.

Third of all, it's on water so nothing special. Many cards on water here and none do that score at 2650 Mhz average.
There's no variation because DS is disabled so the card will keep itself at the min value, at least, which is low.

The undervolt does nothing because the card does not hit 1.175V at all during high loads. It has too much Vdroop so the offset does not help specially if you're on water and have decent temps.

fCLK changes do nothing at all, try them yourself. Infinity fabric is related to SAM and SAM does almost nothing for Time Spy.

I've tested million of settings, I also observed HWinfo during various runs running various settings, the reported clockspeed from 3dmark is pretty consistent with what the cards do.
 

·
Registered
Joined
·
393 Posts
Cold temps do nothing. His core is at 60ºC so clearly nothing exotic.
I've done runs with a portable AC blowing onto the card, GPU at 10C and it's exactly the same with ambient of 25C.
If you do not hit HS throttle the card doesn't care.
 

·
Registered
Joined
·
92 Posts
Dude. SAM makes a HUGE difference for me in Timespy with the latest driver. It's like an automatic +700 points when I enable it. With a higher FCLK, maybe it could add another 100 points. I haven't tested it yet. Also, for my card specifically Power limit numbers make a difference. If I go too high then I get instability. +35% PL/TDC/SOC is the sweet spot for me. +40% PL/TDC/SOC is less stable at 2770MHz than +35% for my GPU.
 

·
Invalid Media
Joined
·
4,400 Posts
...in my experience and on my setup, undervolting also makes a difference. These modern GPUs adjust almost continuously given their key input parameter readings rather than stay at a locked voltage, and if you have a decent chip, undervolting does help, depending also on your MPT PL etc.

...as to FCLK, that too can make a difference, depending on your setup. For example, when I first got my 3090 Strix OC earlier in the year, I had it running on an older 5960X (Intel 8c/16t HEDT) before moving it to an AMD 3950x and then AMD 5950X; bumping bus speeds from 100 to 103 instantly yielded around 300 points in PR with everything else equal (ie GPU settings) on the 5960X.

...just because some people outscore others that have a 'higher MHz' doesn't mean they're cheating, they may just know more about overall system efficiency, or have a better chip sample.
 

·
Registered
Joined
·
393 Posts
...in my experience and on my setup, undervolting also makes a difference. These modern GPUs adjust almost continuously given their key input parameter readings rather than stay at a locked voltage, and if you have a decent chip, undervolting does help, depending also on your MPT PL etc.

...as to FCLK, that too can make a difference, depending on your setup. For example, when I first got my 3090 Strix OC earlier in the year, I had it running on an older 5960X (Intel 8c/16t HEDT) before moving it to an AMD 3950x and then AMD 5950X; bumping bus speeds from 100 to 103 instantly yielded around 300 points in PR with everything else equal (ie GPU settings) on the 5960X.

...just because some people outscore others that have a 'higher MHz' doesn't mean they're cheating, they may just know more about overall system efficiency, or have a better chip sample.
I'm talking GPU score here not overall.

I have the fastest 5800X on Time Spy without exotic cooling (LN2/DICE) so I kinda know what I'm talking about.

I have a 23600 run myself so it's not like I started overclocking today. I was also #1 on my setup previous to 7.1 drivers.
Even with 7.1 drivers and despite having only managed a 23600 Graphic Score run, I'm still #4 (techinically #2because all the 3 first runs are from the same guy).

If you don't find it odd that reference cards are doing 24000 Graphic Score at 2650 Mhz average or you're not curious about it at least, then I'm sorry to ask but what are we doing here?

Anyway, I'll keep investigating and testing on my own setup since I'm doing covid isolation so got time.
 

·
Registered
Joined
·
92 Posts
I'm talking GPU score here not overall.

I have the fastest 5800X on Time Spy without exotic cooling (LN2/DICE) so I kinda know what I'm talking about.

I have a 23600 run myself so it's not like I started overclocking today. I was also #1 on my setup previous to 7.1 drivers.
Even with 7.1 drivers and despite having only managed a 23600 Graphic Score run, I'm still #4 (techinically #2because all the 3 first runs are from the same guy).

If you don't find it odd that reference cards are doing 24000 Graphic Score at 2650 Mhz average or you're not curious about it at least, then I'm sorry to ask but what are we doing here?

Anyway, I'll keep investigating and testing on my own setup since I'm doing covid isolation so got time.
It's definitely an outlier, but he has shown a lot of the tweaks he's done. He probably did some other tweaks as well. We'll never know unless someone contacts him and asks.
 

·
Registered
Joined
·
393 Posts
It's definitely an outlier, but he has shown a lot of the tweaks he's done. He probably did some other tweaks as well. We'll never know unless someone contacts him and asks.
There are a couple more runs of other reference cards doing the same.
I'm not exactly sure what they are doing because I have tested all the new features on MPT and nothing does anything special and worth mentioning.

Look at this run:


2518136


Reference card doing 2800Mhz average.
The guy even says "no EVC" which I 99% doubt unless he binned a bunch of reference cards to get one that could do those clocks stable.

There's certainly something going on with these runs and I'm not sure what it is, maybe some OS tweaks, maybe LOD tweaks, I'm not sure.

There's a few runs on TS with reference cards doing insane scores with pretty low clockspeed.
 

·
Invalid Media
Joined
·
4,400 Posts
(...)
If you don't find it odd that reference cards are doing 24000 Graphic Score at 2650 Mhz average or you're not curious about it at least, then I'm sorry to ask but what are we doing here?
(...)
...there's a difference between being curious and jumping to conclusions, as I have learned in my decade plus of oc'ing which includes HWBot (below) many moons ago. The poster of the score you referenced seemed to have done a decent job of sharing the number of tweaks. Perhaps there's funny stuff going on, perhaps he just really good to produce such an outlier...I certainly won't jump to any automatic conclusion with what I have seen so far

2518135
 

·
PC Evangelist
Joined
·
47,600 Posts
There are a couple more runs of other reference cards doing the same.
I'm not exactly sure what they are doing because I have tested all the new features on MPT and nothing does anything special and worth mentioning.

Look at this run:


View attachment 2518136

Reference card doing 2800Mhz average.
The guy even says "no EVC" which I 99% doubt unless he binned a bunch of reference cards to get one that could do those clocks stable.

There's certainly something going on with these runs and I'm not sure what it is, maybe some OS tweaks, maybe LOD tweaks, I'm not sure.

There's a few runs on TS with reference cards doing insane scores with pretty low clockspeed.

That is pretty fast for TS at 2850 for 1.175mv which is not even XTXH.
 

·
Registered
Joined
·
393 Posts
...there's a difference between being curious and jumping to conclusions, as I have learned in my decade plus of oc'ing which includes HWBot (below) many moons ago. The poster of the score you referenced seemed to have done a decent job of sharing the number of tweaks. Perhaps there's funny stuff going on, perhaps he just really good to produce such an outlier...I certainly won't jump to any automatic conclusion with what I have seen so far

No idea where this jumping to conclusions comes from.
Read my posts. I question the scores not the integrity of the overclockers or whatever you're trying to insinuate.
 

·
Registered
Joined
·
92 Posts
New here, finally upgraded to a 6900xt. First run and just playing with settings got me this score.
Card is a 6900xt toxic.
Max power curve, 1175 volts, 2800mhz max and 2700 min, memory at 2150 normal timings.

Very nice! With MPT you should be able to get around 24k GPU score.
 

·
Registered
Joined
·
64 Posts
Need advice from the watercooling experts here. I've got yet again the same problem with my Red Devil 6900xt: GPU Edge and Junction temp delta has gotten out of hand, it's 40C at worst scenarios and I've got a relatively cool apartment with aircon on, maybe 23C ambient or so. Any advice why this might happen?

Rig
-Alphacool Red Devil 6800xt/6900xt waterblock using their default thermal paste and pads that came with the block. I put a very thin layer of paste. I installed the block strictly following the guide, however one of the vram thermal pads got a bit diagonal as it didn't somehow fit in the right place
-Using 21.7.1 drivers, 2550mhz min 2650mhz max core, 2110mhz mem fast timings, MPT: 350W/375TDP, +15% PL in Adrenaline. I think the drivers upped the temps, but they might have gotten higher before and I just didn't test them before driver update
-Edge/Junction Delta goes above 30C in TS once power load goes above 360W or so, at 380W it is 35C, and at 405W 40C.
-I've got Corsair XD3 Gen.2 pump at 3500rpm, two radiators 240 (top) and 360 (front) in Lianli Lancool II Performance Mesh, both rads are corsair XR5 30mm. 240 rad has 2x best noctuas at 1650rpm push config and the 360 has push pull 6 arctic p12 at 1720rpm (I learned later that that's too many fans for the front rad, live and learn)
-With the above I see GPU temp edge max out at 60C and spikes up to 100C junction in Timespy GT2, junction hovers above 80C constantly in non-vsynced 4K games, it used to be 10C-15C lower. CPU goes up to 63C in TS test (5600X, given slight voltage boost where its absolute high spikes are at 1.31v, mostly cores stays under 1.27v)
-No warranty on the GPU, can keep using it, but I don't like that high temps, and by no means it should be that high and the delta is concerning
-Corsair Softtubes, no leaks, pump well under highest points, flow from pump to GPU to 240 rad to cpu waterblock (also corsair) to 360 rad to pump

My hypothesis is that the thermal paste quality is bad, or I applied too little of it, or this one thermal pad is loosening the mount, or I have a clogged loop. Or alternatively, this is exactly the cooling performance I can expect from the above parts, and that the drivers have upped the thermals. I still think the Delta is too large, but as it stays under 110C and does not affect perfromance, I can live with it.

IMPORTANTLY: This is what I did with it when I had the loop installed: (I know this was a mistake) I put lukewarm tap water in it first to flush out any crap from the parts (not distilled and not too hot, I made a mistake here), then I realized my mistake, and used distilled water to flush out the tap water, and after that I used clear coolant 1000ml flushing out the distilled water as well, and finally leaving the loop very much filled (1.5cm space on top of the pump.)

What would you suggest given the above, should I do a full maintenance on all the parts, or just leave as it as is?

I have four weeks left on my (paid) time off before I start working again, so I have time to flush & drain the loop, remount gpu/cpu blocks, and do a full-fledged maintenance. I'm pretty new at this so now that the system is working at least, I'm a bit wary of doing this as I might actually break something (any experience in remounting waterblocks would be appreciated, especially how to deal with thermal pads that came with the block).
 

·
Invalid Media
Joined
·
4,400 Posts
A quick fyi: I noticed that MPT beta '4' was out. It still needs work, but I could oc my VRAM well past 2150. In fact, when I let the auto VRAM overclock in Wattman do it, it came up with 2260 MHz for the VRAM on its own :p ...I even 'seemed' to have 1.200v going on my non-H (Wattman accepted the 'lock') but could only run light rendering - crasheritis after...further, when you slide the VRAM above 2150, it will take out the 'max GPU MHz' setting, sadly. Still, some progress by the good folks behind MPT ! Clearly, they're on the right track.

2518184



@ptt1982

While you had an 'unusual method' to flush (while already installed ?), I wouldn't panic about that if you subsequently flushed it thoroughly with distilled water and appropriate cooling liquid.

The Hotspot delta is very high, though it will increase significantly with the PL (ie 380 W) you mentioned anyhow. Re. flow, you might want to invest in a flow meter but even a visual inspection re. flow into the reservoir should tell you about blockage. I've only used D5 pumps for the last 10 years or so (usually two in series) rather than DDCs so I can't really comment directly on that setup as to what is a good flow for it. I have read though that that pump can be a bit restrictive.

Finally, is your GPU mounted vertically ? I ask because if you use thin (even highly-rated) thermal paste AND a block that has some stand-off tolerance issues, you might get an uneven mount and subsequently uneven die coverage / thermal paste displacement. I am not sure if that is the case with your setup, but the last year or so has seen a host of quality and tolerance problems with many GPU blocks (including my EKWB / 3090)...here is another OCN link which > alludes to that ...using washers (w/ plastic on the bottom re. avoidance of conductivity) can help, btw, but make sure not to over-tighten the block and also strive for even mounting pressure

If you still have a few weeks to re-do things, you might want to just give yourself some peace of mind by taking the loop apart, flushing each piece properly (as much as I don't think it is the problem) and also remount the block - perhaps after using some imprint paper and cheap thermal paste first for an imprint to see about die coverage.

I suggest you might also try some 'thicker' thermal paste such as Gelid GC Extreme (it is highly rated). Finally, if you are not sure about thermal pad size (or your block just has odd tolerances), you might want to try thermal putty (below, 10 W/mk) for the VRAM and power stages though obviously not the GPU die itself. Thermal putty is good stuff and it conforms to the available space w/o running out, even in a vertical GPU mount.

But it is worth noting again that a high PL such as 380 W will increase the delta hotspot temp no matter what...

2518188
 
2441 - 2460 of 2532 Posts
Top