AMD only does X86 ISA chips . . . . for now . . . . they're rumored to be revisiting their K12 ARM chip project from a decade ago abandonment and starting that up again. Or the modern version of that project.
Sorry, I don't get this ARM desktop thing. If you could offer an explanation to go along what you post, it may help. But for me, I use AMD's CPUs. And I'm not really sure if Intel uses an ARM chip or not.
To me, I'll let this pass...
George,
I have been curious about the possibility of crunching boinc projects at a competitive level (more than raspberry Pi) per system for a while. I was very interested in turnkey level systems because I am not a very good Linux technician.
The idea would be to get x86 level performance while spending less money on electricity. This henges on a user friendly (and price friendly) system and boinc application support.
Yes I could rent a cloud instance but for what we do it is out of my budget range.
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!
Sorry, I don't get this ARM desktop thing. If you could offer an explanation to go along what you post, it may help. But for me, I use AMD's CPUs. And I'm not really sure if Intel uses an ARM chip or not.
To me, I'll let this pass...
George,
I have been curious about the possibility of crunching boinc projects at a competitive level (more than raspberry Pi) per system for a while. I was very interested in turnkey level systems because I am not a very good Linux technician.
The idea would be to get x86 level performance while spending less money on electricity. This henges on a user friendly (and price friendly) system and boinc application support.
Yes I could rent a cloud instance but for what we do it is out of my budget range.
Tom M
Right now the Arm cpu's, ie R-Pi, Android etc, applications aren't nearly as profilic as the AMD and Intel 'standard' cpu tasks for very task, ie they are app specific like cpu's versus gpu's. Now in a Server setting for most things each task is usually small and doesn't require sustained intensive computing power in Boinc for example it's all allowing a connection, point to the correct directories, close the connection and move to the next one. For the backend it's similar in that it's about moving tasks to the right directory once they are received or change the settings to reflect that the tasks were sent to or received from a user. For work it's similar in that alot of computing power is not needed to provide x number of users access to this or that database, the key is the more cpu's the more users can acess the database etc at one time. So LOTS of tiny cpu's means things happen faster for the employees and others accessing the Server. That's the theory anyway, with those requiring more intensive Server access seeing a slowdown due to the limitations of the current Arm cpu as opposed to the normally Intel but also Epyc and AMD cpu's. Now IF they can increase the capabilities of the new ARM cpu's then the limitations can be reduced or eliminated and then the cost price points of the various other cpu's could drop dramatically.
Sorry, I don't get this ARM desktop thing. If you could offer an explanation to go along what you post, it may help. But for me, I use AMD's CPUs. And I'm not really sure if Intel uses an ARM chip or not.
To me, I'll let this pass...
George,
I have been curious about the possibility of crunching boinc projects at a competitive level (more than raspberry Pi) per system for a while. I was very interested in turnkey level systems because I am not a very good Linux technician.
The idea would be to get x86 level performance while spending less money on electricity. This henges on a user friendly (and price friendly) system and boinc application support.
Yes I could rent a cloud instance but for what we do it is out of my budget range.
Tom M
If you dig into the charts that Phoronix always publishes for cpus, the $ per flops and $ per watt metrics always have the new Genoa cpus at the highest performance per calculation costs and wattage.
Beats anything that's come before and likely into the future till the next AMD Zen 5 processors show up..
I wonder how that Japanese supercomputer that was on top of that list compared to those numbers? It was some kind of Arm system. ;)
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!
That would have been the Fugaku system I believe. Don't know whether anyone has analyzed the wattage/per flop metric for the TOP500 supercomputers.
Might be they were the top position for that metric. Think I saw that they have been toppled down the chart pretty good by the latest supercomputers that have been built.
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!
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!
One concern being raised about the M3 is the memory bandwidth is lower on the regular and Pro models than on the M1 and M2. Yet the M3 benchmarks faster. Once a bunch get in use we'll find out more about this.
An M3 Max 16" configured with the same RAM and SSD as my M1 Max 16" costs about $500 more. Not cool...
Intel only does X86 ISA
)
Intel only does X86 ISA chips.
AMD only does X86 ISA chips . . . . for now . . . . they're rumored to be revisiting their K12 ARM chip project from a decade ago abandonment and starting that up again. Or the modern version of that project.
GWGeorge007 wrote: Tom M
)
George,
I have been curious about the possibility of crunching boinc projects at a competitive level (more than raspberry Pi) per system for a while. I was very interested in turnkey level systems because I am not a very good Linux technician.
The idea would be to get x86 level performance while spending less money on electricity. This henges on a user friendly (and price friendly) system and boinc application support.
Yes I could rent a cloud instance but for what we do it is out of my budget range.
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!
Tom M wrote: GWGeorge007
)
Right now the Arm cpu's, ie R-Pi, Android etc, applications aren't nearly as profilic as the AMD and Intel 'standard' cpu tasks for very task, ie they are app specific like cpu's versus gpu's. Now in a Server setting for most things each task is usually small and doesn't require sustained intensive computing power in Boinc for example it's all allowing a connection, point to the correct directories, close the connection and move to the next one. For the backend it's similar in that it's about moving tasks to the right directory once they are received or change the settings to reflect that the tasks were sent to or received from a user. For work it's similar in that alot of computing power is not needed to provide x number of users access to this or that database, the key is the more cpu's the more users can acess the database etc at one time. So LOTS of tiny cpu's means things happen faster for the employees and others accessing the Server. That's the theory anyway, with those requiring more intensive Server access seeing a slowdown due to the limitations of the current Arm cpu as opposed to the normally Intel but also Epyc and AMD cpu's. Now IF they can increase the capabilities of the new ARM cpu's then the limitations can be reduced or eliminated and then the cost price points of the various other cpu's could drop dramatically.
Tom M wrote: GWGeorge007
)
If you dig into the charts that Phoronix always publishes for cpus, the $ per flops and $ per watt metrics always have the new Genoa cpus at the highest performance per calculation costs and wattage.
Beats anything that's come before and likely into the future till the next AMD Zen 5 processors show up..
I wonder how that Japanese
)
I wonder how that Japanese supercomputer that was on top of that list compared to those numbers? It was some kind of Arm system. ;)
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!
That would have been the
)
That would have been the Fugaku system I believe. Don't know whether anyone has analyzed the wattage/per flop metric for the TOP500 supercomputers.
Might be they were the top position for that metric. Think I saw that they have been toppled down the chart pretty good by the latest supercomputers that have been built.
https://www.notebookcheck.net
)
https://www.notebookcheck.net/First-Snapdragon-X-Elite-benchmarks-Impressive-gains-over-M2-Max-Ryzen-9-7940HS-Intel-13th-gen-H-and-14th-gen-desktop-CPUs.763149.0.html
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!
Wonder where the Apple M3
)
Wonder where the Apple M3 will fall out tonight?
Keith Myers wrote: Wonder
)
Apples always drop in the Fall? ;)
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!
One concern being raised
)
One concern being raised about the M3 is the memory bandwidth is lower on the regular and Pro models than on the M1 and M2. Yet the M3 benchmarks faster. Once a bunch get in use we'll find out more about this.
An M3 Max 16" configured with the same RAM and SSD as my M1 Max 16" costs about $500 more. Not cool...