Just as there is a price/performance sweet spot for Nvidia (rtx 3080 ti) what is your best guess for the Radeon series?
Maybe still the Radeon VII with the price decline? Or something else?
No I am not considering jumping ship. But I had a good run with my Rx 5700's.
So?
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 was wondering about why the Radeon VIi still gets a lot of use in crunching.
It turns out it's fp64 Tflops are quite robust at 3+.
A Nvidia Titan V has about the same fp32 but about twice fp64.
.Apparently both the Radeon VIi and Titan V are pro cards.
Are there any current generation Amd pro cards that exceed Radeon VIi performance and what kind of prices are they going for? Or even previous gen but later than Vega64.
Calculating volunteers want to know :)
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 Radeon Pro VII has twice the FP64 as the normal Radeon VII.
you won’t get more FP64 than that on AMD without going to the Instinct cards. MI100 comes to mind, it’s the same card that valterc is using on that 8-GPU system in 2nd place on the hosts leaderboard.
FP64 isn’t majorly important here at Einstein, but it does seem to help a little for BRP7.
The Radeon Pro VII has twice the FP64 as the normal Radeon VII.
you won’t get more FP64 than that on AMD without going to the Instinct cards. MI100 comes to mind, it’s the same card that valterc is using on that 8-GPU system in 2nd place on the hosts leaderboard.
FP64 isn’t majorly important here at Einstein, but it does seem to help a little for BRP7.
It looks as if the Nvidia Titan V and the Radeon Pro VIi have about the same theoretical fp32 and fp64 performance?
Instinct GPUs would apparently need an after market fan for cooling in non-server case environments. Standard server cases presumably howl.
One of the after market fans looks like it would hinder installing 4 PCIe Instinct GPU's side by side on an AsRock Epycd8 motherboard.
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 Pro VII is about 14% behind the Titan V in peak theoretical computational Flops (FP16/FP32/FP64).
the Pro VII lacks CUDA or Tensor cores, which somewhat limits what you can do with it in regards to other use cases/projects/apps. Milkyway Separation used to be the go-to app for high performing FP64 cards, but that's gone now. GPUGRID recently released a QChem app which greatly benefits from high performing FP64 cards also, but it's CUDA only so the Pro VII can't do it.
the Pro VII has 16GB VRAM. and more overall memory bandwidth which might be more important for Einstein's apps. the Tesla V100 or Quadro GV100 Nvidia cards remedy the VRAM/bandwidth situation, but comes at a hefty price.
every choice has compromises. there is no single best solution. just have to decide what's right for you and your budget/goals.
As far as I remember, E@H don’t require FP64 at all, FP32 is sufficient. So FP64 might only of interest for other projects than E@H. And only AMD has a FP32 ALU, which is requiring less electrical power, because of smaller registers. NVIDIA is crunching FP32 on his FP64 ALU. (There was a thread, I don’t find anymore, from Bernd (???) several years ago about this.)
As far as I remember, E@H don’t require FP64 at all, FP32 is sufficient. So FP64 might only of interest for other projects than E@H. And only AMD has a FP32 ALU, which is requiring less electrical power, because of smaller registers. NVIDIA is crunching FP32 on his FP64 ALU. (There was a thread, I don’t find anymore, from Bernd (???) several years ago about this.)
Kind regards and happy crunching
Martin
the relatively new BRP7 project does require FP64 support from the GPU.
every choice has compromises. there is no single best solution. just have to decide what's right for you and your budget/goals.
The joys of "optimization" :)
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!
As you know most of but not all top performing systems on the project are Nvidia based cpus.
Now this shows up. So possiblly with its assistance we could run Nvidia petri optimized GPU apps on Amd GPU's?
It's claiming binary compatible.
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!
Just as there is a
)
Just as there is a price/performance sweet spot for Nvidia (rtx 3080 ti) what is your best guess for the Radeon series?
Maybe still the Radeon VII with the price decline? Or something else?
No I am not considering jumping ship. But I had a good run with my Rx 5700's.
So?
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!
with BRP7 having more benefit
)
with BRP7 having more benefit from FP64, and the rather heavy price decrease recently, best bang for buck is probably the Radeon VII.
_________________________________________________________________________
I was wondering about why the
)
I was wondering about why the Radeon VIi still gets a lot of use in crunching.
It turns out it's fp64 Tflops are quite robust at 3+.
A Nvidia Titan V has about the same fp32 but about twice fp64.
.Apparently both the Radeon VIi and Titan V are pro cards.
Are there any current generation Amd pro cards that exceed Radeon VIi performance and what kind of prices are they going for? Or even previous gen but later than Vega64.
Calculating volunteers want to know :)
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 Radeon Pro VII has twice
)
The Radeon Pro VII has twice the FP64 as the normal Radeon VII.
you won’t get more FP64 than that on AMD without going to the Instinct cards. MI100 comes to mind, it’s the same card that valterc is using on that 8-GPU system in 2nd place on the hosts leaderboard.
FP64 isn’t majorly important here at Einstein, but it does seem to help a little for BRP7.
_________________________________________________________________________
Ian&Steve C. wrote: The
)
It looks as if the Nvidia Titan V and the Radeon Pro VIi have about the same theoretical fp32 and fp64 performance?
Instinct GPUs would apparently need an after market fan for cooling in non-server case environments. Standard server cases presumably howl.
One of the after market fans looks like it would hinder installing 4 PCIe Instinct GPU's side by side on an AsRock Epycd8 motherboard.
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 Pro VII is about 14%
)
the Pro VII is about 14% behind the Titan V in peak theoretical computational Flops (FP16/FP32/FP64).
the Pro VII lacks CUDA or Tensor cores, which somewhat limits what you can do with it in regards to other use cases/projects/apps. Milkyway Separation used to be the go-to app for high performing FP64 cards, but that's gone now. GPUGRID recently released a QChem app which greatly benefits from high performing FP64 cards also, but it's CUDA only so the Pro VII can't do it.
the Pro VII has 16GB VRAM. and more overall memory bandwidth which might be more important for Einstein's apps. the Tesla V100 or Quadro GV100 Nvidia cards remedy the VRAM/bandwidth situation, but comes at a hefty price.
every choice has compromises. there is no single best solution. just have to decide what's right for you and your budget/goals.
_________________________________________________________________________
Hallo!As far as I
)
Hallo!
Why worrying about FP64 performance for E@H???
As far as I remember, E@H don’t require FP64 at all, FP32 is sufficient. So FP64 might only of interest for other projects than E@H. And only AMD has a FP32 ALU, which is requiring less electrical power, because of smaller registers. NVIDIA is crunching FP32 on his FP64 ALU. (There was a thread, I don’t find anymore, from Bernd (???) several years ago about this.)
Kind regards and happy crunching
Martin
astro-marwil
)
the relatively new BRP7 project does require FP64 support from the GPU.
_________________________________________________________________________
Ian&Steve C. wrote: every
)
The joys of "optimization" :)
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!
https://videocardz.com/newz/u
)
https://videocardz.com/newz/unmodified-nvidia-cuda-apps-can-now-run-on-amd-gpus-thanks-to-zluda
As you know most of but not all top performing systems on the project are Nvidia based cpus.
Now this shows up. So possiblly with its assistance we could run Nvidia petri optimized GPU apps on Amd GPU's?
It's claiming binary compatible.
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!