Nvidia rtx 3090 discussion

No shunt mod? Push that sucker to 450w :D.

Anyways with new 3960x, found out that turning off SMT makes a huge jump in performance in TimeSpy for the EVGA 3090 XC3 Ultra with Threadripper, this is with stock GPU settings. Also system memory speeds also has a big impact on performance. Anyways this is a comparison of before and after with no OC on the 3090, all stock settings and just changing system memory speeds from 3200mhz to 3733mhz and turning off SMT. Impact is huge!

https://www.3dmark.com/compare/spy/17329195/spy/17213622
This makes me REALLY interested to see how performance will be affected once I finally get my 3090 out of my old i7e 6950X system and into my Ryzen 9 5950X system.
 
  • Like
Reactions: noko
like this
This makes me REALLY interested to see how performance will be affected once I finally get my 3090 out of my old i7e 6950X system and into my Ryzen 9 5950X system.
TBH, I doubt you will notice much in terms of gaming performance (assuming you have a decent OC on that 6950x).
 
TBH, I doubt you will notice much in terms of gaming performance (assuming you have a decent OC on that 6950x).
It’s at 4GHz. Single threaded gaming performance is about equal to a 6700K. So there is probably some performance gains to be had in a few games even at higher resolutions. But I’m not expecting anything significant across the board.
 
This makes me REALLY interested to see how performance will be affected once I finally get my 3090 out of my old i7e 6950X system and into my Ryzen 9 5950X system.
With the EVGA 3090 FTWs going pop or with a bang becoming a paper weight. Maybe not so smart in shunt modding that XC3 :D. I will probably in the near term just keep my XC3 with original bios unless EVGA comes out with one like a 390w limit or via software allow a higher power target which may not be possible, seems like the newer GPUs are much more limited by the firmware. As for Threadripper and SMT, I really have to check out real games and if that makes that kind of obscene difference in performance.
 
With the EVGA 3090 FTWs going pop or with a bang becoming a paper weight. Maybe not so smart in shunt modding that XC3 :D. I will probably in the near term just keep my XC3 with original bios unless EVGA comes out with one like a 390w limit or via software allow a higher power target which may not be possible, seems like the newer GPUs are much more limited by the firmware. As for Threadripper and SMT, I really have to check out real games and if that makes that kind of obscene difference in performance.
390w still seems reasonable to me. But once you are pushing 400+ watts as a daily driver I just don’t think that’s a great idea. I thought my 1080 Ti FTW3 at 330-350w was already dumping a lot of heat into my loop. I can’t imagine another 100w on top of that.
 
With the EVGA 3090 FTWs going pop or with a bang becoming a paper weight. Maybe not so smart in shunt modding that XC3 :D. I will probably in the near term just keep my XC3 with original bios unless EVGA comes out with one like a 390w limit or via software allow a higher power target which may not be possible, seems like the newer GPUs are much more limited by the firmware. As for Threadripper and SMT, I really have to check out real games and if that makes that kind of obscene difference in performance.
Ha, what do you mean going "pop"...? Are people pushing a 400w+ BIOS on an XC3 (small cooler) and frying the damn things on air? My FTW3 has a 450W limit by default from factory, and under heavy load, my temps never top 70C. Should I be worried or something?
 
Ha, what do you mean going "pop"...? Are people pushing a 400w+ BIOS on an XC3 (small cooler) and frying the damn things on air? My FTW3 has a 450W limit by default from factory, and under heavy load, my temps never top 70C. Should I be worried or something?
It’s what I mentioned a few posts back. There are numerous threads/posts on the EVGA forum about FTW3 cards (mostly 3090s) that suffer from an audible pop then the screen goes black with fans running at max RPM. GPU is dead. Cause hasn’t been determined yet and there are even a few people who had the same issue occur with their replacement card.
 
As for Threadripper and SMT, I really have to check out real games and if that makes that kind of obscene difference in performance.

I've got a 3970x and had a 3960x (upgraded for productivity workloads). I noticed that some games seem to like SMT while others perform better without. Sounds weird (and a non-answer) but games like Master of Orion (3) do better with SMT enabled, FPS games get better FPS with SMT OFF as expected. Cyberpunk is a whole other bag of stupid.
 
  • Like
Reactions: noko
like this
some of you may have heard that certain cards thermal throttle due to heat of the gddr6x memory modules. While I didn't have thermal throttling exactly, the memory temps were causing my fan to go crazy while mining, doing deep learning, and some other intensive tasks, despite the core temp hanging out in the 40s. I heard the problem could be remedied by replacing the thermal pads, so I went ahead and did that, just on the rear under the backplate of my founders edition 3090.

I tested before and after by mining etherium. I mined for 1 hour to heatsoak the card, and then mined another hour while monitoring info, then took the average. Did this both before and after. This is on an open air test bench with no additional fans.

Before:64-67% fan, 65% 1700rpm avg; 43-44C core, 44* avg
After: 52-54%, 53% 1390rpm avg; 46-47C core, 47* avg

big improvement to the sound profile.
 
I've got a 3970x and had a 3960x (upgraded for productivity workloads). I noticed that some games seem to like SMT while others perform better without. Sounds weird (and a non-answer) but games like Master of Orion (3) do better with SMT enabled, FPS games get better FPS with SMT OFF as expected. Cyberpunk is a whole other bag of stupid.
I did Shadow Of The Tomb Raider, 1440p, Max, no RT, no motion blur . . . bla bla bla - With SMT 136 fps and without 145 fps. Not the same level of difference I saw in TimeSpy by any stretch of the imagination once again pointing the limitation of that benchmark. I will have to do Flight Simulator 2020, since a really mostly a CPU bound game/simulation.
 
SMT becomes inefficient when there are more threads than the game engine employs. For most games with 8 cores, you still get better performance with HT/SMT off. AFAIK its only with newer DICE games where SMT on is not a performance hit for 8+ core CPUs.
 
I did Shadow Of The Tomb Raider, 1440p, Max, no RT, no motion blur . . . bla bla bla - With SMT 136 fps and without 145 fps. Not the same level of difference I saw in TimeSpy by any stretch of the imagination once again pointing the limitation of that benchmark. I will have to do Flight Simulator 2020, since a really mostly a CPU bound game/simulation.

SMT off in FS2020 is about 5-10% gain with either a 3900x or a 5800x (the 2 cpus I have).
 
SMT off in FS2020 is about 5-10% gain with either a 3900x or a 5800x (the 2 cpus I have).
That is what I am finding, the 3090 GPU is not even being loaded 60% with the 3960x at 1440p Ultimate settings in San Francisco, 40-45fps unless flying really low, With SMT on it is also more erratic with frame times. Increasing game render resolution to 150% for 4K rendering, it does not affect the fps, GPU is around 80% loaded. Not the best CPU for FS2020. I could load Ryzen Master again and use game mode, maybe some other day.
 
That is what I am finding, the 3090 GPU is not even being loaded 60% with the 3960x at 1440p Ultimate settings in San Francisco, 40-45fps unless flying really low, With SMT on it is also more erratic with frame times. Increasing game render resolution to 150% for 4K rendering, it does not affect the fps, GPU is around 80% loaded. Not the best CPU for FS2020. I could load Ryzen Master again and use game mode, maybe some other day.

Just wondering, what speed ram are you running? I did notice a change in behavior when I replaced my 4x16 3200 with 4x16 3600 (B Die) ram across some games.
 
Just wondering, what speed ram are you running? I did notice a change in behavior when I replaced my 4x16 3200 with 4x16 3600 (B Die) ram across some games.
Using 3733mhz, 16,16,16,36 initially, went to 16,16,16,38 due to startup issues. Still fine tuning. Currently BDie 8gb sticks 4x, have 16gb 4x sticks on the shelf with Hynix which will do 3600mhz, slower timings. I think I am hitting platform limitations with the 3090 with SMT on and 24 cores/48 threads. Going from 1440p max settings in FS2020 to 4K and not seeing any fps change is a big indicator plus still the GPU is not fully loaded.
 
Is this considered a ghetto mod

20210114_143056.jpg
 
Is that actually functional? I’d imagine you’d want to install the heatsinks on the actual memory chips and VRMs to dissipate the heat effectively. The backplate will spread the heat before it gets absorbed into the heatsinks.
The backplate gets hot as 75c. The ram sinks are over the ram chips, sense chips, and power delivery. It will add more surface area at the least, but I do think it will help. It's an itx case with a rear rad directly over the backplate and a front rad blowing hot air into the case, so Im hoping this makes it a little less worst case scenario for the card, mainly to keep it from ramping the fan like crazy due to ram Temps.
 
The backplate gets hot as 75c. The ram sinks are over the ram chips, sense chips, and power delivery. It will add more surface area at the least, but I do think it will help. It's an itx case with a rear rad directly over the backplate and a front rad blowing hot air into the case, so Im hoping this makes it a little less worst case scenario for the card, mainly to keep it from ramping the fan like crazy due to ram Temps.
Is the backplate metal or plastic? If it’s plastic it’s probably acting more like an insulator than anything.
 
It's metal
Let me know how it works, I’m actually interested.

My water cooled 3090 will be mounted vertically directly over intake fans. That should hopefully alleviate any cooling concerns on the back of the card.
 
Is this considered a ghetto mod
Personally, I think it will help minimally and not worth the effort.

One, the backplate is metal and is already distributing the heat load and pulling heat away from the chips. The extra heatsinks my add a little more surface area, but without good thermal transference from one metal that is spreading the heat out to the other along with lack of sufficient airflow to cool down the heatsinks, you will be back to square one in heat.

The idea of heatsinks is to not only increase surface area to absorb and move away the heat, but also to give a larger surface area for a cooling mechanism (be it air or water) to cool them down so that they can continue to pull heat away from the source. Otherwise, you are just delaying the inevitable that they will become the same temp as everything else.

I think you would better served, if possible, to put a fan sideways and have it blow air directly over the top of the graphics card and its fins. I think that will give you better gains in dropping the temp down than adding these small heatsinks on top of what is pretty much a heatsink.
 
Personally, I think it will help minimally and not worth the effort.

One, the backplate is metal and is already distributing the heat load and pulling heat away from the chips. The extra heatsinks my add a little more surface area, but without good thermal transference from one metal that is spreading the heat out to the other along with lack of sufficient airflow to cool down the heatsinks, you will be back to square one in heat.

The idea of heatsinks is to not only increase surface area to absorb and move away the heat, but also to give a larger surface area for a cooling mechanism (be it air or water) to cool them down so that they can continue to pull heat away from the source. Otherwise, you are just delaying the inevitable that they will become the same temp as everything else.

I think you would better served, if possible, to put a fan sideways and have it blow air directly over the top of the graphics card and its fins. I think that will give you better gains in dropping the temp down than adding these small heatsinks on top of what is pretty much a heatsink.
I will report back, hopefully tonight. I don't really have room for a fan.

20210114_154042.jpg


Time waste or not, they're staying, because I've gone too far to turn back :D
 
As an Amazon Associate, HardForum may earn from qualifying purchases.
Yes that's a very ghetto mod.
If you're doing to do something like that, get this and throw a fan on it. You may have to dremel it to shape as I'm not sure if it will clear the 3090 edges.

https://www.amazon.com/gp/product/B089QJQY17/

*Edit* now that I saw your second picture...not sure it will fit anymore...
I looked at that one. I didn't have clearance for 15mm tall. The ones I have are 6mm.
Screenshot_20210114-214800_Gallery.jpg
 
As an Amazon Associate, HardForum may earn from qualifying purchases.
Personally, I think it will help minimally and not worth the effort.

One, the backplate is metal and is already distributing the heat load and pulling heat away from the chips. The extra heatsinks my add a little more surface area, but without good thermal transference from one metal that is spreading the heat out to the other along with lack of sufficient airflow to cool down the heatsinks, you will be back to square one in heat.

The idea of heatsinks is to not only increase surface area to absorb and move away the heat, but also to give a larger surface area for a cooling mechanism (be it air or water) to cool them down so that they can continue to pull heat away from the source. Otherwise, you are just delaying the inevitable that they will become the same temp as everything else.

I think you would better served, if possible, to put a fan sideways and have it blow air directly over the top of the graphics card and its fins. I think that will give you better gains in dropping the temp down than adding these small heatsinks on top of what is pretty much a heatsink.
I have my Zotac 3090 ghettoed with a heatsink and fan on the backplate. I bought it for gaming but am mining with it also to make up some of the money I spent.

Anyway, it reduced the backplate temp by 10 to 20C. Heatsink with no fan didn't help much. I put a 60mm Noctua on it. Even at 100% power, no thermal throttling.

Ghetto but works until I go to water at some point. Although the backplate might still be an issue depending on the design of the waterblock I suppose.
 
Using 3733mhz, 16,16,16,36 initially, went to 16,16,16,38 due to startup issues. Still fine tuning. Currently BDie 8gb sticks 4x, have 16gb 4x sticks on the shelf with Hynix which will do 3600mhz, slower timings. I think I am hitting platform limitations with the 3090 with SMT on and 24 cores/48 threads. Going from 1440p max settings in FS2020 to 4K and not seeing any fps change is a big indicator plus still the GPU is not fully loaded.

Well that's just the worst; having a non-fully loaded CPU AND GPU while not achieving framerates that you want.


I have my Zotac 3090 ghettoed with a heatsink and fan on the backplate. I bought it for gaming but am mining with it also to make up some of the money I spent.

Anyway, it reduced the backplate temp by 10 to 20C. Heatsink with no fan didn't help much. I put a 60mm Noctua on it. Even at 100% power, no thermal throttling.

Ghetto but works until I go to water at some point. Although the backplate might still be an issue depending on the design of the waterblock I suppose.

Have you thought about moving the GPU to the next x16 slot and using a PCI slot fan cooler pointing down? Amazon Link

I always wondered why we don't use more dimensions for cooling.
 
As an Amazon Associate, HardForum may earn from qualifying purchases.
Well that's just the worst; having a non-fully loaded CPU AND GPU while not achieving framerates that you want.




Have you thought about moving the GPU to the next x16 slot and using a PCI slot fan cooler pointing down? Amazon Link

I always wondered why we don't use more dimensions for cooling.
I don't think I have room. I also have an Elgato 4k60 MK2 in there that needs at least a x4 slot. I have that exact cooler in my saved items but decided it wouldn't fit. I might give it another look though as I wasn't considering moving down the vid card. Thanks for the idea.
 
As an Amazon Associate, HardForum may earn from qualifying purchases.
Well that's just the worst; having a non-fully loaded CPU AND GPU while not achieving framerates that you want.




Have you thought about moving the GPU to the next x16 slot and using a PCI slot fan cooler pointing down? Amazon Link

I always wondered why we don't use more dimensions for cooling.
Games limiting threads to only a few cores and probably how the memory is allocated per core and who knows what else. Anyways the CPU is like 9% loaded :ROFLMAO:, not really doing much of anything but definitely not loading the GPU. I should be getting in the 50s, not 40s. There maybe better ways for this system as in like game mode using Ryzen Master, I don't know. FS2020 and Threadripper, at least for me has issues on performance. CPU really is for other type of applications which is fine and needed. I just need to explore more.
 
As an Amazon Associate, HardForum may earn from qualifying purchases.
Let me know how it works, I’m actually interested.

My water cooled 3090 will be mounted vertically directly over intake fans. That should hopefully alleviate any cooling concerns on the back of the card.

Mine is mounted horizontally, watercooled and the back reaches upwards of 124F. So far it hasn't been a problem.
 
Mine is mounted horizontally, watercooled and the back reaches upwards of 124F. So far it hasn't been a problem.
124f = ~51c. I was seeing 70-75C on the founders backplate on an open air test bench. Not an issue per se.
 
Well...my waterblock that was supposed to ship today was delayed until early February. :banghead:

I have an Amazon order for the 3090 FTW3 for the original, pre-tariff MSRP that’s expected to be delivered at the end of this month. I think I’m going to try it out and compare it against my XC3 to see if it’s worth dropping down the extra $200.
 
Well...my waterblock that was supposed to ship today was delayed until early February. :banghead:

I have an Amazon order for the 3090 FTW3 for the original, pre-tariff MSRP that’s expected to be delivered at the end of this month. I think I’m going to try it out and compare it against my XC3 to see if it’s worth dropping down the extra $200.
Not sure how the XC3 does, but my FTW3 stays under 70C at all times gaming and the boosts are about 1985Mhz~2025Mhz average, with peaks 2055Mhz~2100Mhz under very heavy loads.

I've typically bought the FTW3 models simply because I like the larger PCB so I can just use the massive stock coolers on them without needing a water loop.

I "heard" they use better power delivery components on the FTW3 cards too...
 
Not sure how the XC3 does, but my FTW3 stays under 70C at all times gaming and the boosts are about 1985Mhz~2025Mhz average, with peaks 2055Mhz~2100Mhz under very heavy loads.

I've typically bought the FTW3 models simply because I like the larger PCB so I can just use the massive stock coolers on them without needing a water loop.

I "heard" they use better power delivery components on the FTW3 cards too...
I’m more interested in seeing the clocks that it reaches compared to my XC3 than temps since it’ll be under water either way.

I also just got notification that the FTW3 shipped and is supposed to be here Tuesday. Gives me a bit of time to test it out as I have until Feb 7th to return the XC3 if I decide to stay with the FTW3.
 
Back
Top