susato wrote:Hoping that the extra detail is helpful.
Quite. Thanks!
6.24 only uses half of processor power?
Re: 6.24 only uses half of processor power?
Quite. Thanks!
-
jecrawford
- Posts: 97
- Joined: Sun Dec 09, 2007 8:02 am
- Location: Altrincham, UK
Re: 6.24 only uses half of processor power?
I am curious as to which, if any, of the new features of the Nehalem quad processors will accelerate folding in a major/minor way.
As I see it, these are the main new features:
1. Single die, quad core processor with 8MB of shared Level 3 cache.
2. Integrated memory controller, giving quicker access to data.
3. Fast DDR3 ECC SDRAM giving increased memory bandwidth.
4. Turbo Boost. One (or more?) cores can have their clock speeds increased if other cores are idle.
5. QuickPath. Quicker access to disk, I/O and other sub-systems, particularly processor-to-processor connection (no passing through I/O hub).
6. Graphics cards:
NVidia GeForce GT120 (can have up to 4 of these!).
ATI Radeon HD 4870
I imagine we will have to wait for Mac drivers and/or Snow Leopard with CUDA etc. to make use of the GPUs.
I would also like to know how 8 cores, each with 2 threads making 16 threads in all, would be deployed for Folding.
Might need some optimizing.
Simple yes/no answers to the above 6 points would be fine. Explanation would be even more welcome.
John
As I see it, these are the main new features:
1. Single die, quad core processor with 8MB of shared Level 3 cache.
2. Integrated memory controller, giving quicker access to data.
3. Fast DDR3 ECC SDRAM giving increased memory bandwidth.
4. Turbo Boost. One (or more?) cores can have their clock speeds increased if other cores are idle.
5. QuickPath. Quicker access to disk, I/O and other sub-systems, particularly processor-to-processor connection (no passing through I/O hub).
6. Graphics cards:
NVidia GeForce GT120 (can have up to 4 of these!).
ATI Radeon HD 4870
I imagine we will have to wait for Mac drivers and/or Snow Leopard with CUDA etc. to make use of the GPUs.
I would also like to know how 8 cores, each with 2 threads making 16 threads in all, would be deployed for Folding.
Might need some optimizing.
Simple yes/no answers to the above 6 points would be fine. Explanation would be even more welcome.
John
-
John Naylor
- Posts: 950
- Joined: Mon Dec 03, 2007 4:36 pm
- Hardware configuration: Q9450 OC @ 3.2GHz (Win7 Home Premium) - SMP2
E7500 OC @ 3.66GHz (Windows Home Server) - SMP2
i5-3750k @ 3.8GHz (Win7 Pro) - SMP2 - Location: University of Birmingham, UK
Re: 6.24 only uses half of processor power?
1) Probably yes. No longer having to use the FSB/QPI to send messages between two pairs of cores should make an appreciable difference.
2) Almost certainly yes. RAM is accessed a lot for the SMP client so speedups here should be noticable in folding performance.
3) Yes, same reason as 2)
4) No. The Folding@home client will load all four physical cores on each processor, thereby stopping any being powered down to allow others to be boosted.
5) Maybe, as the current client uses the localhost which invokes the networking hardware.
6) No difference to SMP, obviously. I don't know anything about the GT120 but the HD4870 will add a not-insignificant amount of points to production if it can run.
There are only 8 eight real cores, so for the purpose of Folding@home SMP it is an 8-core machine. The a2 core supports scalable threads so can work with the -smp 8 flag to utilise all the CPU power with one client. However the a1 core which is still in use is locked to four threads so to ensure 100% usage at all times it might be necessary to run two clients with -smp 4.
Nice to see another member from around my neck of the woods
2) Almost certainly yes. RAM is accessed a lot for the SMP client so speedups here should be noticable in folding performance.
3) Yes, same reason as 2)
4) No. The Folding@home client will load all four physical cores on each processor, thereby stopping any being powered down to allow others to be boosted.
5) Maybe, as the current client uses the localhost which invokes the networking hardware.
6) No difference to SMP, obviously. I don't know anything about the GT120 but the HD4870 will add a not-insignificant amount of points to production if it can run.
There are only 8 eight real cores, so for the purpose of Folding@home SMP it is an 8-core machine. The a2 core supports scalable threads so can work with the -smp 8 flag to utilise all the CPU power with one client. However the a1 core which is still in use is locked to four threads so to ensure 100% usage at all times it might be necessary to run two clients with -smp 4.
Nice to see another member from around my neck of the woods
Folding whatever I'm sent since March 2006 Beta testing since October 2006. www.FAH-Addict.net Administrator since August 2009.
Beta testing since October 2006. www.FAH-Addict.net Administrator since August 2009.Re: 6.24 only uses half of processor power?
CUDA has been supported on OSX for over a year. GPU on OSX could work fine if the PG decided to develop it.
We may speculate why they haven't done a GPU/OSX client. Possible reasons are:
- the popular Mac laptops are poor candidates for GPU folding
- an expectation that Windows virtualization software will someday support CUDA, at which point OSX users could fold GPU units on virtual machines
- the GPU/WINE community removing any incentive for the PG to develop an official GPU/Linux client which could be ported to OSX
- Lack of support from the card manufacturers whose staff have played a major role in development of the Windows GPU clients and are still actively improving the drivers, shader programming etc. for GPU's
If you're folding on one of the Nehalem quad processors (or I should say, two of them!) under OSX please be sure to submit your productivity results to fahinfo, to build the community's database on the performance of this new chip.
We may speculate why they haven't done a GPU/OSX client. Possible reasons are:
- the popular Mac laptops are poor candidates for GPU folding
- an expectation that Windows virtualization software will someday support CUDA, at which point OSX users could fold GPU units on virtual machines
- the GPU/WINE community removing any incentive for the PG to develop an official GPU/Linux client which could be ported to OSX
- Lack of support from the card manufacturers whose staff have played a major role in development of the Windows GPU clients and are still actively improving the drivers, shader programming etc. for GPU's
If you're folding on one of the Nehalem quad processors (or I should say, two of them!) under OSX please be sure to submit your productivity results to fahinfo, to build the community's database on the performance of this new chip.
-
jecrawford
- Posts: 97
- Joined: Sun Dec 09, 2007 8:02 am
- Location: Altrincham, UK
Re: 6.24 only uses half of processor power?
Thank you John & Susato.
Just what I needed as explanation.
John
Just what I needed as explanation.
John
-
jecrawford
- Posts: 97
- Joined: Sun Dec 09, 2007 8:02 am
- Location: Altrincham, UK
Re: 6.24 only uses half of processor power?
I see that Apple are going to offer the Quadro FX 4800 graphics card in May/June.
I wonder if this could be used for Folding by running Windows using Bootcamp on a Nehalem MacPro?
I wonder if this could be used for Folding by running Windows using Bootcamp on a Nehalem MacPro?
-
John Naylor
- Posts: 950
- Joined: Mon Dec 03, 2007 4:36 pm
- Hardware configuration: Q9450 OC @ 3.2GHz (Win7 Home Premium) - SMP2
E7500 OC @ 3.66GHz (Windows Home Server) - SMP2
i5-3750k @ 3.8GHz (Win7 Pro) - SMP2 - Location: University of Birmingham, UK
Re: 6.24 only uses half of processor power?
If it is supported by CUDA then it can run the folding@home client. If the client does not recognise it you can force it by using the "-forcegpu nvidia_g80" parameter, but only if CUDA supports it.
Folding whatever I'm sent since March 2006 Beta testing since October 2006. www.FAH-Addict.net Administrator since August 2009.
Beta testing since October 2006. www.FAH-Addict.net Administrator since August 2009.Re: 6.24 only uses half of processor power?
John Naylor wrote:If it is supported by CUDA then it can run the folding@home client. If the client does not recognise it you can force it by using the "-forcegpu nvidia_g80" parameter, but only if CUDA supports it.
. . . and only if you have a GPU client that will run on your OS.
Posting FAH's log:
How to provide enough info to get helpful support.
How to provide enough info to get helpful support.
Re: 6.24 only uses half of processor power?
The only mention of GPU on Mac hardware that I could find on this forum was here:
viewtopic.php?f=52&t=3193&p=32642#p32642
in which casw1000 runs the GPU client on a NVidia 8800 GT, a Windows SMP client, and a Linux SMP client in a VM all at once, on a 2.8 GHz dual quad Xeon Mac Pro running XP (32 bit) under Boot Camp. What a smorgasbord! Yet each client was folding away successfully. So John, I think your chances are excellent.
BTW, there is at least one person trying to run GPU on WINE on Mac hardware - Any reader interested in contributing to the effort is encouraged to visit that thread.
viewtopic.php?f=52&t=3193&p=32642#p32642
in which casw1000 runs the GPU client on a NVidia 8800 GT, a Windows SMP client, and a Linux SMP client in a VM all at once, on a 2.8 GHz dual quad Xeon Mac Pro running XP (32 bit) under Boot Camp. What a smorgasbord! Yet each client was folding away successfully. So John, I think your chances are excellent.
BTW, there is at least one person trying to run GPU on WINE on Mac hardware - Any reader interested in contributing to the effort is encouraged to visit that thread.