Is there a more power efficient GPU than the Titan V?
Not quite sure how to quantify this.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
the 40-series cards were pretty close or a little more efficient from what i saw in previous testing. at least with the 4090 rentals I tried out and reports from other users. but of course those are a lot more expensive too.
If you are using the MPS server under Linux I have a few notes for you.
Titan V and a GTX 1060 will not play together on the same system under MPS.
Rtx 3090 and Titan Xp will not play together on the same system under MPS.
Each GPU will run under MPS.
This also likely applies to the GTX 1080 ti.
Yet to be tested are a rtx 2000 series or rtx 3000 series GPU with a Titan V. Based on some of our guesses these might work.
I think we can be fairly confident that if you don't mix GPU's under the MPS they all work.
If you don't run the MPS server you can mix and match the above GPU's.
And under Windows you can mix and match.
Respectfully,
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
I am reporting the results of both myself and San Fernando Valley with Ian&StevC's instigation for testing.
I almost ordered a Titan Xp until the results with the Gtx 1060 indicated it probably wouldn't be useful on my; Titan V box.
Looking for a low cost rtx 2080 ti right now.
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
The issue is with mixing pre and post Volta GPUs in combination with MPS. you dont need to test every possible combination to find what does or doesnt work. MPS has different implementations for "Pascal and older" and "Volta and newer".
From MPS docs wrote:
The Volta architecture introduced new MPS capabilities. Compared to MPS on pre-Volta GPUs, Volta MPS provides a few key improvements:
Volta MPS clients submit work directly to the GPU without passing through the MPS server.
Each Volta MPS client owns its own GPU address space instead of sharing GPU address space with all other MPS clients.
Volta MPS supports limited execution resource provisioning for Quality of Service (QoS).
Pascal + Maxwell (both pre-Volta) = works
Volta + Ampere (both post-Volta) = works
Pascal + Ampere (pre and post Volta) = wont work (the Pascal card will not be detected)
just about any combination will work together when you are not using MPS if they are supported by the same driver.
Pascal + Maxwell (both pre-Volta) = works
Volta + Ampere (both post-Volta) = works
Pascal + Ampere (pre and post Volta) = wont work (the Pascal card will not be detected)
just about any combination will work together when you are not using MPS if they are supported by the same driver.
Pascal + Maxwell (both pre-Volta) + Volta + Ampere (both post-Volta) + ????? for the 50xx series cards?
50-series cards are Blackwell. nothing changes with MPS. the examples I gave are examples to show the three basic categories of configurations in regards to MPS, they are not meant to be all inclusive. every possible combination will fit in these three categories.
everything after Volta (Titan V is Volta architecture) 2017/2018-ish is "post-Volta" so you really don't need to speculate. every new generation is going to be after every previous one.
If you didn't see Nvidia's keynote, here are some highlights:
They "CLAIM" the 5070 will have the performance of the 4090 (not sure what they are measuring) but it will be $549 retail (this is going to be so much more expensive/impossible to find if that is true).
5080- $999
5070 Ti -$749
5090- $1,999. What I can't believe is the size of the GPU- it's so thin. It looks dual slot?? It will need an empty slot next to it because of the fans, but that was a huge surprise. Keep in mind this is the reference card and other designs will probably be thicker. But it's so thin. I paused the keynote to make sure what he was holding was the 5090, and it was.
This REALLY is impressive- we thought they would be more expensive (MSRP). I will add to this post as I watch more.
Rtx 5070 crunching as fast as the Rtx 4090? That is impressive if true.
They may have to ration them to help keep the scalping under control.
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
Is there a more power
)
Is there a more power efficient GPU than the Titan V?
Not quite sure how to quantify this.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
the 40-series cards were
)
the 40-series cards were pretty close or a little more efficient from what i saw in previous testing. at least with the 4090 rentals I tried out and reports from other users. but of course those are a lot more expensive too.
_________________________________________________________________________
If you are using the MPS
)
If you are using the MPS server under Linux I have a few notes for you.
Titan V and a GTX 1060 will not play together on the same system under MPS.
Rtx 3090 and Titan Xp will not play together on the same system under MPS.
Each GPU will run under MPS.
This also likely applies to the GTX 1080 ti.
Yet to be tested are a rtx 2000 series or rtx 3000 series GPU with a Titan V. Based on some of our guesses these might work.
I think we can be fairly confident that if you don't mix GPU's under the MPS they all work.
If you don't run the MPS server you can mix and match the above GPU's.
And under Windows you can mix and match.
Respectfully,
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
Have you tried your GTX 1060
)
Have you tried your GTX 1060 & Titan XP?
Have you tried your RTX 3090 & Titan V?
Just a thought...
Proud member of the Old Farts Association
GWGeorge007 wrote: Have you
)
I don't own a Titan Xp. I don't own an Rtx 3090.
I am reporting the results of both myself and San Fernando Valley with Ian&StevC's instigation for testing.
I almost ordered a Titan Xp until the results with the Gtx 1060 indicated it probably wouldn't be useful on my; Titan V box.
Looking for a low cost rtx 2080 ti right now.
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
The issue is with mixing pre
)
The issue is with mixing pre and post Volta GPUs in combination with MPS. you dont need to test every possible combination to find what does or doesnt work. MPS has different implementations for "Pascal and older" and "Volta and newer".
Pascal + Maxwell (both pre-Volta) = works
Volta + Ampere (both post-Volta) = works
Pascal + Ampere (pre and post Volta) = wont work (the Pascal card will not be detected)
just about any combination will work together when you are not using MPS if they are supported by the same driver.
_________________________________________________________________________
Ian&Steve C. wrote: Pascal +
)
Pascal + Maxwell (both pre-Volta) + Volta + Ampere (both post-Volta) + ????? for the 50xx series cards?
Do we know what the new cards will be?
Proud member of the Old Farts Association
50-series cards are
)
50-series cards are Blackwell. nothing changes with MPS. the examples I gave are examples to show the three basic categories of configurations in regards to MPS, they are not meant to be all inclusive. every possible combination will fit in these three categories.
everything after Volta (Titan V is Volta architecture) 2017/2018-ish is "post-Volta" so you really don't need to speculate. every new generation is going to be after every previous one.
_________________________________________________________________________
If you didn't see Nvidia's
)
If you didn't see Nvidia's keynote, here are some highlights:
They "CLAIM" the 5070 will have the performance of the 4090 (not sure what they are measuring) but it will be $549 retail (this is going to be so much more expensive/impossible to find if that is true).
5080- $999
5070 Ti -$749
5090- $1,999. What I can't believe is the size of the GPU- it's so thin. It looks dual slot?? It will need an empty slot next to it because of the fans, but that was a huge surprise. Keep in mind this is the reference card and other designs will probably be thicker. But it's so thin. I paused the keynote to make sure what he was holding was the 5090, and it was.
This REALLY is impressive- we thought they would be more expensive (MSRP). I will add to this post as I watch more.
Rtx 5070 crunching as fast as
)
Rtx 5070 crunching as fast as the Rtx 4090? That is impressive if true.
They may have to ration them to help keep the scalping under control.
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!