Hi all: I have gone through hundreds of comments related to Radeon RX 7900 XT or similar...... The more I read more confused I become and are, in general, far above my skill set....... Soooo...Let me ask my question in a simple way.
I have the opportunity to to obtain an RX7900XT. It would run on a Windows 10 (maybe Win 11, but prefer 10). This dedicate system would be used ONLY for EIN work. No fancy stuff, just EIN. Again, EIN, GPU, with enough CPU and RAM for it to work.
If this type set-up works for you, please let me know.
Hi all: I have gone through hundreds of comments related to Radeon RX 7900 XT or similar...... The more I read more confused I become and are, in general, far above my skill set....... Soooo...Let me ask my question in a simple way.
I have the opportunity to to obtain an RX7900XT. It would run on a Windows 10 (maybe Win 11, but prefer 10). This dedicate system would be used ONLY for EIN work. No fancy stuff, just EIN. Again, EIN, GPU, with enough CPU and RAM for it to work.
If this type set-up works for you, please let me know.
Thanks, folks..... JB
According to versus.com the 7900xt is a better buy than a Radeon VII:
Techpowerup lists the 32 bit floating point as being 3x a Titan V. Since a Titan V can crank out north of 2M RAC on the All-Sky GW tasks it should allow your proposed system to join the top 50 list.
The Radian VII lists a 32 bit floating point about 1/4 of the 7900 xt.
It looks like a good bet from here.
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!
try not to look at performance based single metrics only. reality is a lot more complicated. here at einstein, memory bandwidth in general is pretty important too. and if you're running O3AS, the CPU speed is important to the overall runtime. it will never do "3x" a Titan V as your comment suggests. a well tuned 4090 can't even do that.
other users here have gotten the RadeonVII up to like 3M+ ppd on O3AS with a fast CPU (5-6 tasks per gpu). I've not seen any AMD RX7000 approach anything close to that.
check the leaderboards for reference (not absolute). there's one host running at least one 7900XTX, but not a lot of insight into how this system is being run or what the other GPUs in the host are to make a determination on what the performance is really like at peak output (running multiple apps, etc).
i think a 7900XT will be a fine performer, it's a decently powerful card and if that's what you want to do and you're stuck with Windows and Linux is not an option it will be fine.
(Linux/Nvidia is generally better/faster/more tuning options/etc)
Hi all: I have gone through hundreds of comments related to Radeon RX 7900 XT or similar...... The more I read more confused I become and are, in general, far above my skill set....... Soooo...Let me ask my question in a simple way.
I have the opportunity to to obtain an RX7900XT. It would run on a Windows 10 (maybe Win 11, but prefer 10). This dedicate system would be used ONLY for EIN work. No fancy stuff, just EIN. Again, EIN, GPU, with enough CPU and RAM for it to work.
If this type set-up works for you, please let me know.
Thanks, folks..... JB
If you are using the system for E@H work only, then I highly suggest switching to Linux- even a user-friendly version like Mint works great. It is super easy to install and get up and running.
Hi all: I have gone through hundreds of comments related to Radeon RX 7900 XT or similar...... The more I read more confused I become and are, in general, far above my skill set....... Soooo...Let me ask my question in a simple way.
I have the opportunity to to obtain an RX7900XT. It would run on a Windows 10 (maybe Win 11, but prefer 10). This dedicate system would be used ONLY for EIN work. No fancy stuff, just EIN. Again, EIN, GPU, with enough CPU and RAM for it to work.
If this type set-up works for you, please let me know.
Thanks, folks..... JB
I did run O3AS before but my guess is that it would be about 3.5X more output than RX 570 just based on memory bandwidth speed as O3AS relies heavily on this and to some extent fast cpu speed. RX 570 has 224GB/s memory bandwidth while 7900XT has 800GB/s or just about 3.5 * 700K PPD or 2.45M PPD would be my expected estimate for 7900 XT. My Radeon VII has 1020GB/s, so about 4.5X faster than RX 570. Currently that Linux RX 570 host above is doing like 700K PPD in O3AS, so 700K x 4.5 = 3.15M PPD which is close to what I got with Radeon VII without memory overclock. BTW, when I say " fast CPU", I'm looking at 4GHz speed. An overclock Ryzen 2600 can get up to that speed. Not 100% sure about BRP7 tasks but most of the top 50 hosts are running O3AS tasks. There is other factor like running multiple tasks per gpu to take advantage of the available memory bandwidth. Just don't run one O3AS task per gpu :)
For Windows based system, there is a host with two RX6600XT cards, each card producing about 800K PPD in O3AS with fast cpu. 6600XT has memory bandwidth of 256GB/s, so relative to my Radeon VII, it is about 1020/256 = 4x slower. So I guess the math works out ok for ball park estimate of the 7900XT performance.
I think you might be able to run on Windows and still get close to 2.45M PPD in O3AS. Anyway as others as pointed out here, linux distro generally have an advantage over windows in crunching. There is an old thread discussing the "best linux distro for crunching" but honestly it is up to each user preference.
I have a 7900xtx on my bench, so I can run some tasks later as a sample.
The XTX in addition to the increased clock speed over the XT, also has more shaders, TMUs, L2 and a wider memory bus. It won't be a direct comparison, but my guess is the XTX is probably 15-20% faster than the XT on O3AS.
Thanks to all for your input. I guess I will try it--replacing a 1070. If it works, I can get a second one to replace my other 1070. I have only done NVIDIA in the past and saw many comments on other threads about the difficulty in loading and using a 7900xt, but it appears all you experts are not having issues. My source for the device is from a long time trusted PC engineer who is not familiar with our particular application. Bottom line: I am trying to keep my antiques up to date. Thanks folks.....
Re suggestions to go LINUX ... Don't I wish,,,,, That is too much for me to tackle now.
The performance data here is what gives me hope. Smile. JB
But that's really an apples to oranges comparison with respect to gpgpu processing on scientific applications versus synthetic benchmark numbers.
The only real proof is to run the Einstein gpu apps on both devices at their maximum throughput capabilities and then compare the RAC credit over a month to reach baseline for both devices.
Hi all: I have gone through
)
Hi all: I have gone through hundreds of comments related to Radeon RX 7900 XT or similar...... The more I read more confused I become and are, in general, far above my skill set....... Soooo...Let me ask my question in a simple way.
I have the opportunity to to obtain an RX7900XT. It would run on a Windows 10 (maybe Win 11, but prefer 10). This dedicate system would be used ONLY for EIN work. No fancy stuff, just EIN. Again, EIN, GPU, with enough CPU and RAM for it to work.
If this type set-up works for you, please let me know.
Thanks, folks..... JB
James Bradshaw wrote: Hi
)
According to versus.com the 7900xt is a better buy than a Radeon VII:
https://versus.com/en/amd-radeon-rx-7900-xt-vs-amd-radeon-vii
BUT there is no crunching stats included in that data. You can play with the choices but I didn't see an Nvidia M100 to compare it too.
Rtx 7900 xtTechpowerup
)
Rtx 7900 xt
Techpowerup lists the 32 bit floating point as being 3x a Titan V. Since a Titan V can crank out north of 2M RAC on the All-Sky GW tasks it should allow your proposed system to join the top 50 list.
The Radian VII lists a 32 bit floating point about 1/4 of the 7900 xt.
It looks like a good bet from here.
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!
try not to look at
)
try not to look at performance based single metrics only. reality is a lot more complicated. here at einstein, memory bandwidth in general is pretty important too. and if you're running O3AS, the CPU speed is important to the overall runtime. it will never do "3x" a Titan V as your comment suggests. a well tuned 4090 can't even do that.
other users here have gotten the RadeonVII up to like 3M+ ppd on O3AS with a fast CPU (5-6 tasks per gpu). I've not seen any AMD RX7000 approach anything close to that.
check the leaderboards for reference (not absolute). there's one host running at least one 7900XTX, but not a lot of insight into how this system is being run or what the other GPUs in the host are to make a determination on what the performance is really like at peak output (running multiple apps, etc).
i think a 7900XT will be a fine performer, it's a decently powerful card and if that's what you want to do and you're stuck with Windows and Linux is not an option it will be fine.
(Linux/Nvidia is generally better/faster/more tuning options/etc)
_________________________________________________________________________
James Bradshaw wrote: Hi
)
If you are using the system for E@H work only, then I highly suggest switching to Linux- even a user-friendly version like Mint works great. It is super easy to install and get up and running.
James Bradshaw wrote: Hi
)
I did run O3AS before but my guess is that it would be about 3.5X more output than RX 570 just based on memory bandwidth speed as O3AS relies heavily on this and to some extent fast cpu speed. RX 570 has 224GB/s memory bandwidth while 7900XT has 800GB/s or just about 3.5 * 700K PPD or 2.45M PPD would be my expected estimate for 7900 XT. My Radeon VII has 1020GB/s, so about 4.5X faster than RX 570. Currently that Linux RX 570 host above is doing like 700K PPD in O3AS, so 700K x 4.5 = 3.15M PPD which is close to what I got with Radeon VII without memory overclock. BTW, when I say " fast CPU", I'm looking at 4GHz speed. An overclock Ryzen 2600 can get up to that speed. Not 100% sure about BRP7 tasks but most of the top 50 hosts are running O3AS tasks. There is other factor like running multiple tasks per gpu to take advantage of the available memory bandwidth. Just don't run one O3AS task per gpu :)
For Windows based system, there is a host with two RX6600XT cards, each card producing about 800K PPD in O3AS with fast cpu. 6600XT has memory bandwidth of 256GB/s, so relative to my Radeon VII, it is about 1020/256 = 4x slower. So I guess the math works out ok for ball park estimate of the 7900XT performance.
I think you might be able to run on Windows and still get close to 2.45M PPD in O3AS. Anyway as others as pointed out here, linux distro generally have an advantage over windows in crunching. There is an old thread discussing the "best linux distro for crunching" but honestly it is up to each user preference.
I have a 7900xtx on my bench,
)
I have a 7900xtx on my bench, so I can run some tasks later as a sample.
The XTX in addition to the increased clock speed over the XT, also has more shaders, TMUs, L2 and a wider memory bus. It won't be a direct comparison, but my guess is the XTX is probably 15-20% faster than the XT on O3AS.
Thanks to all for your
)
Thanks to all for your input. I guess I will try it--replacing a 1070. If it works, I can get a second one to replace my other 1070. I have only done NVIDIA in the past and saw many comments on other threads about the difficulty in loading and using a 7900xt, but it appears all you experts are not having issues. My source for the device is from a long time trusted PC engineer who is not familiar with our particular application. Bottom line: I am trying to keep my antiques up to date. Thanks folks.....
Re suggestions to go LINUX ... Don't I wish,,,,, That is too much for me to tackle now.
The performance data here is what gives me hope. Smile. JB
https://technical.city/en/video/GeForce-GTX-1070-vs-Radeon-RX-7900-XT
But that's really an apples
)
But that's really an apples to oranges comparison with respect to gpgpu processing on scientific applications versus synthetic benchmark numbers.
The only real proof is to run the Einstein gpu apps on both devices at their maximum throughput capabilities and then compare the RAC credit over a month to reach baseline for both devices.
Thanks Keith. I never
)
Thanks Keith. I never thought of it in that way. Appreciate it.