"The RTX 5090 is projected to include 204 Streaming Multiprocessors (SMs), leading to a total core count of 26,112. It is expected to be equipped with 32Gbps GDDR7 memory, operating over a 384-bit bus. "
but the 4x system that is slowly getting upgraded is not watercooled. The stock Titan V cooler is a 2-slot blower cooler.
Today, after re-pasting one of my blower rtx 3080 ti's I noticed that they are not rattling at all. Simply blowing noisily. Then I put my new-to-me Dell rtx 3080 ti on the same test bed. And it is slightly quieter. But not all that quiet.
It seems like the Evga rtx 3080 ti FTW3 is the quietest of the bunch. Can't really test it since it is hiding inside the Super Micro dual cpu system. Which has its own howling VRM cooling fan issues.
So exactly how noisy is the 4 x Titan V blower/air cooled system? Ian&SteveC?
I am still looking at sidegrading from 4 rtx 3080 ti's (two blowers, 1 air cooled, 1 liquid cooled) to 4 Titan V's but if they are as noisy or nearly so as my currently blowers its "less inviting".
Tom M
Just my 5c worth: as I am working only & exclusively with blower GPUs recently.
Yes, blower GPUs are noisier, while also similarly (or more) efficient (as standard GPU coolers). But the reason why I use them:
- mostly in my systems I use Quadro cards, as those are more robust for calculations
- all Quadro(s) come with blowers & what I found out is that blowers take hot air & blow it out. Compared to other cards, which "heat up my cases". So I also switched to all cards being blower ones & now my CPU temps. have dropped by 5~10°C.
Now, that is much, especially if you run calculations also on CPUs (which I do, mostly on Universe@home, WCG, Rosetta@home & DENIS). So this gives me some advantage, so it is an easy trade of.
Part of my constraints is wanting a rtx 3080 ti class blower type GPU. Apparently rtx 3080 blower type GPU's are more common. But 3080's loose production over 3080 ti's.
I ran across a large supply of rtx 3090 blower type GPU'S. The prices are quite high. Beyond my budget unless I am going to settle for just one GPU. And it specifically says they are from servers. So they easily could be quite noisy.
Or sell 4 rtx 3080 ti's to get 2 3090's.
I expect I will simply put off the decision for now. My budgetary constraints are limiting me to running one rtx 3080 ti even though I have four more.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
Here is more discussion to add to the Titan V vs rtx 5000 comments.
you're confusing the "RTX 5000 Ada" (professional "Quadro" level card) with 32GB (capacity) of GDDR6 VRAM. this card is in between a 4080/4080Ti and a 4090 in specs, with double the VRAM from the 4080.
with
hypothetical specs for the next generation of Geforce RTX 5090 which will likely have 24GB of "32Gbps" GDDR7 (the speed of the VRAM, not the capacity)
Do GPU's hyperthread the same way hyperthreading CPU's do?
This also maybe related to my experience that running 2x tasks on a GPU that is already running at 100 percent utilization results in no net gain in production.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
Do GPU's hyperthread the same way hyperthreading CPU's do?
This also maybe related to my experience that running 2x tasks on a GPU that is already running at 100 percent utilization results in no net gain in production.
I ran across this. And I remember discussion about using Nvidia GPU's in a specific mode that both allowed more performance and did not allow multiple different GPU applications.
I am wondering if the underlying technology that is described in the above article is what Ian&Steve discussed in that high performance mode for Nvidia GPU's?
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
Possibly a simpler solution.
)
Possibly a simpler solution. Run one EVGA rtx 3080 ti ftw3 in the next room.
Sell all other temptations.
:)
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
Tom M wrote: Possibly a
)
Nah just get one of these and you be more than at the bottom of the list!!
https://www.guru3d.com/story/rumored-nvidia-rtx-5090-specs-cores-gddr7-memory-and-bandwidth/
"The RTX 5090 is projected to include 204 Streaming Multiprocessors (SMs), leading to a total core count of 26,112. It is expected to be equipped with 32Gbps GDDR7 memory, operating over a 384-bit bus. "
Tom M wrote: Ian&Steve C.
)
Just my 5c worth: as I am working only & exclusively with blower GPUs recently.
Yes, blower GPUs are noisier, while also similarly (or more) efficient (as standard GPU coolers). But the reason why I use them:
- mostly in my systems I use Quadro cards, as those are more robust for calculations
- all Quadro(s) come with blowers & what I found out is that blowers take hot air & blow it out. Compared to other cards, which "heat up my cases". So I also switched to all cards being blower ones & now my CPU temps. have dropped by 5~10°C.
Now, that is much, especially if you run calculations also on CPUs (which I do, mostly on Universe@home, WCG, Rosetta@home & DENIS). So this gives me some advantage, so it is an easy trade of.
non-profit org. Play4Life in Zagreb, Croatia, EU
Klik, I see your
)
Klik,
I see your logic.
Part of my constraints is wanting a rtx 3080 ti class blower type GPU. Apparently rtx 3080 blower type GPU's are more common. But 3080's loose production over 3080 ti's.
I ran across a large supply of rtx 3090 blower type GPU'S. The prices are quite high. Beyond my budget unless I am going to settle for just one GPU. And it specifically says they are from servers. So they easily could be quite noisy.
Or sell 4 rtx 3080 ti's to get 2 3090's.
I expect I will simply put off the decision for now. My budgetary constraints are limiting me to running one rtx 3080 ti even though I have four more.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
https://www.servethehome.com/
)
https://www.servethehome.com/nvidia-rtx-5000-ada-32gb-workstation-gpu-review/
Here is more discussion to add to the Titan V vs rtx 5000 comments.
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
Tom M
)
you're confusing the "RTX 5000 Ada" (professional "Quadro" level card) with 32GB (capacity) of GDDR6 VRAM. this card is in between a 4080/4080Ti and a 4090 in specs, with double the VRAM from the 4080.
with
hypothetical specs for the next generation of Geforce RTX 5090 which will likely have 24GB of "32Gbps" GDDR7 (the speed of the VRAM, not the capacity)
these are not the same thing.
_________________________________________________________________________
Do GPU's hyperthread the same
)
Do GPU's hyperthread the same way hyperthreading CPU's do?
This also maybe related to my experience that running 2x tasks on a GPU that is already running at 100 percent utilization results in no net gain in production.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
Tom M wrote: Do GPU's
)
no
_________________________________________________________________________
https://towardsdatascience.co
)
https://towardsdatascience.com/parallelizing-heavy-gpu-workloads-via-multi-queue-operations-50a38b15a1dc
I ran across this. And I remember discussion about using Nvidia GPU's in a specific mode that both allowed more performance and did not allow multiple different GPU applications.
I am wondering if the underlying technology that is described in the above article is what Ian&Steve discussed in that high performance mode for Nvidia GPU's?
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
No, The article is about
)
No, The article is about Vulkan technology. What Ian uses is for CUDA technology.
Not the same.