- Apr 27, 2000
- 21,632
- 10,845
- 136
Formulav8 approached me about putting together some benchmark numbers to see what effects NB overclocking would have on a Propus chip (he is testing an L3-less dual-core K10 . . . Rana I think?). Here are the results at four different NB speeds:
Here are some newer numbers to show the effects of NB clocks on applications that tax the video card (somewhat, more on this below):
I was going to create graphs and such, but found it somewhat tedious compared to just cutting and pasting the data in a post. I'm so lazy.
edit: formulav8 made some graphs using my data. Not 100% complete but I'm not complaining.
http://www.vbcodesource.info/graphs/drmrlordx/
(did not embed jpgs with img tags to avoid slow loads).
Thanks for the graphis, formulav8!
edit edit: I have added the above extra benchmark results to show what happens when you raise NB speed when running applications that tax the video card. Unfortunately, my monitor will not support resolutions higher than 1280x1024, so that's the best I can do; however, that was only really a factor in the Trackmania Nations Forever benching.
It seems that the benchmark showing the greatest benefit from NB speed is Cinebench R10 OpenGL. R11.5, not so much.
edit edit edit: Added 3DMark Vantage scores. Highly inconsistent. I couldn't generate enough trial keys to do enough runs to normalize data, but it seems like there's a lot of noise involved.
Anyway, as you can see, there are gains to be had, though it should be noted that they are not amazing; still, I did not expect NB speed to play much of a role in Propus' performance (and I certainly did not expect memory read and write numbers to be affected by it). Also, my Cinebench R10 single CPU results are a bit odd in that they seemed to go up as NB speed decreased after a certain point. How odd. The highest NB speed still gave me the best single CPU and multi CPU scores so that was as expected.
Blender scores were generated using an optimized 32-bit build of version 2.5 alpha 2 while rendering a test file known as test.blend: http://www.eofw.org/bench/test.blend?PHPSESSID=819e1c2b501ae28ff68c17f67b1a8a69
Another bizarre finding was that Cinebench R11.5 had to be run at a LOWER vcore than other benchmarks because the heat from high vcore settings made it unstable at speeds of 3.7 ghz or higher. I am planning to improve my cooling on this chip to see what positive results can be had from dropping a few more degrees @ load.
Also, this should demonstrate, without a doubt, that Propus just can't hang with Deneb at the same clock speed. With a better IMC, maybe it could, but as it stands, it is simply inferior. Propus MIGHT be competitive in benchmarks/applications with an enormous working set, but I haven't found anything to demonstrate that possibility (yet).
Here are some newer numbers to show the effects of NB clocks on applications that tax the video card (somewhat, more on this below):
I was going to create graphs and such, but found it somewhat tedious compared to just cutting and pasting the data in a post. I'm so lazy.
edit: formulav8 made some graphs using my data. Not 100% complete but I'm not complaining.
http://www.vbcodesource.info/graphs/drmrlordx/
(did not embed jpgs with img tags to avoid slow loads).
Thanks for the graphis, formulav8!
edit edit: I have added the above extra benchmark results to show what happens when you raise NB speed when running applications that tax the video card. Unfortunately, my monitor will not support resolutions higher than 1280x1024, so that's the best I can do; however, that was only really a factor in the Trackmania Nations Forever benching.
It seems that the benchmark showing the greatest benefit from NB speed is Cinebench R10 OpenGL. R11.5, not so much.
edit edit edit: Added 3DMark Vantage scores. Highly inconsistent. I couldn't generate enough trial keys to do enough runs to normalize data, but it seems like there's a lot of noise involved.
Anyway, as you can see, there are gains to be had, though it should be noted that they are not amazing; still, I did not expect NB speed to play much of a role in Propus' performance (and I certainly did not expect memory read and write numbers to be affected by it). Also, my Cinebench R10 single CPU results are a bit odd in that they seemed to go up as NB speed decreased after a certain point. How odd. The highest NB speed still gave me the best single CPU and multi CPU scores so that was as expected.
Blender scores were generated using an optimized 32-bit build of version 2.5 alpha 2 while rendering a test file known as test.blend: http://www.eofw.org/bench/test.blend?PHPSESSID=819e1c2b501ae28ff68c17f67b1a8a69
Another bizarre finding was that Cinebench R11.5 had to be run at a LOWER vcore than other benchmarks because the heat from high vcore settings made it unstable at speeds of 3.7 ghz or higher. I am planning to improve my cooling on this chip to see what positive results can be had from dropping a few more degrees @ load.
Also, this should demonstrate, without a doubt, that Propus just can't hang with Deneb at the same clock speed. With a better IMC, maybe it could, but as it stands, it is simply inferior. Propus MIGHT be competitive in benchmarks/applications with an enormous working set, but I haven't found anything to demonstrate that possibility (yet).
Last edited:
Facebook
Twitter