So I'm bored and don't have much to do the next couple of days. I was searching around lately to find some info about HT on vs off on the i7 processors. Lots of the info I found was right after release and I was curious how everything looked now, after updates and everything...
I'm looking to do tests on how all cores (physical, and logical) perform in different scenarios to find out out for myself and others what is better. HT on, or HT off and slightly higher clock.
Right now I am performing Fritz Chess Benchmarks on individual cores, then I will mix it up with combinations. Later I plan on doing some Vantage runs.
I use task manager to set affinity to cores of my choosing and make sure logical cores aren't being used when physical ones should be, etc.
I'm using Windows 7 as an os.
Any other tests or suggestions anyone thinks of I will be happy to do.
TESTS
Fritz Chess
Some quick results there.
For physical cores when HT is on or off it doesn't make much of a difference. It also makes no difference if a core has the PHYSICAL or LOGICAL core stressed.
However, if BOTH physical and logical cores are stressed neither performs nearly as well when the SAME data is being calculated. (This may change if each core runs a different type of program and different instruction sets)
Still, with HT on you see a TOTAL improvement of 25-30% on each core.
Started working on Crysis CPU test with HT OFF clocked at 4.33GHz.
I did the benchmark on 1,2,3 and all 4 cores and the FPS made a HUGE jump from when just one core could be used. (HUGE bottleneck) Then very small improvements after more than one core could be accessed.
Something to note here:
I used Orthos stress test to *simulate* an intensive program running on a core. I chose this because it uses a core to 100%, but it doesn't COMPLETELY steal all of its resources.
Note how when I had three instances of it running the FPS went UP. I was watching core usage during the test and noticed all 4 processors were utilized 100%. (the three Orthos and the one without a *dedicated* process)
BUT! Even though FPS went up there was a noticeable degradation in *playing experience*. Crysis was stealing cycles from cores with 100% utilization already, this made things slightly choppy. Most of all objects would jump to where they were supposed to end up without actually moving smoothly all the way.
This just proves that sometimes even though you see number improvements or shortfalls they don't always matter, playability went down even though FPS went up.
Starting with some game graphs...
L4D for starters.
First is the graph for FPS improvements for 1,2,3,4 cores used.
NOTE: Often people ran into big problems because as seen above, when a physical core and its logical core are used together the performance would be poor on both cores, but in TOTAL 25-30% higher. Its like this:
If [0 - 1] [2 - 3] [4 - 5] [6 - 7] is used it will use FOUR physical cores to full potential.
If [0 -1] [2 - 3] [ 4 - 5] [6 - 7] is used, performance will be only ~64% of what the above can achieve.
ANYWAY: L4D solves this problem! It WILL NOT use a logical core. It just won't. It will ALWAYS pick the physical cores to use. (this is why my graph only shows 1-4 cores even though HT is on, I couldn't use logical ones.)
Well here is the picture, look for yourself and see what you get...
Next we have a graph showing the effects of having other programs running while gaming. This could be converting files, encrypting, scanning for viruses, spyware, burning a DVD, playing a HD movie on another screen, WHATEVER. Check out how well an i7 with HT can handle MANY HEAVY processes running at once...
Note: I switched to using IBT for *other processes* because it was a little easier to use and provided a heavier processes not letting my tests interfere as much...
Edited by Finny - 8/25/09 at 9:51pm
I'm looking to do tests on how all cores (physical, and logical) perform in different scenarios to find out out for myself and others what is better. HT on, or HT off and slightly higher clock.
Right now I am performing Fritz Chess Benchmarks on individual cores, then I will mix it up with combinations. Later I plan on doing some Vantage runs.
I use task manager to set affinity to cores of my choosing and make sure logical cores aren't being used when physical ones should be, etc.
I'm using Windows 7 as an os.
Any other tests or suggestions anyone thinks of I will be happy to do.
TESTS
Fritz Chess
Some quick results there.
For physical cores when HT is on or off it doesn't make much of a difference. It also makes no difference if a core has the PHYSICAL or LOGICAL core stressed.
However, if BOTH physical and logical cores are stressed neither performs nearly as well when the SAME data is being calculated. (This may change if each core runs a different type of program and different instruction sets)
Still, with HT on you see a TOTAL improvement of 25-30% on each core.
Started working on Crysis CPU test with HT OFF clocked at 4.33GHz.
I did the benchmark on 1,2,3 and all 4 cores and the FPS made a HUGE jump from when just one core could be used. (HUGE bottleneck) Then very small improvements after more than one core could be accessed.
Something to note here:
I used Orthos stress test to *simulate* an intensive program running on a core. I chose this because it uses a core to 100%, but it doesn't COMPLETELY steal all of its resources.
Note how when I had three instances of it running the FPS went UP. I was watching core usage during the test and noticed all 4 processors were utilized 100%. (the three Orthos and the one without a *dedicated* process)
BUT! Even though FPS went up there was a noticeable degradation in *playing experience*. Crysis was stealing cycles from cores with 100% utilization already, this made things slightly choppy. Most of all objects would jump to where they were supposed to end up without actually moving smoothly all the way.
This just proves that sometimes even though you see number improvements or shortfalls they don't always matter, playability went down even though FPS went up.
Starting with some game graphs...
L4D for starters.
First is the graph for FPS improvements for 1,2,3,4 cores used.
NOTE: Often people ran into big problems because as seen above, when a physical core and its logical core are used together the performance would be poor on both cores, but in TOTAL 25-30% higher. Its like this:
If [0 - 1] [2 - 3] [4 - 5] [6 - 7] is used it will use FOUR physical cores to full potential.
If [0 -1] [2 - 3] [ 4 - 5] [6 - 7] is used, performance will be only ~64% of what the above can achieve.
ANYWAY: L4D solves this problem! It WILL NOT use a logical core. It just won't. It will ALWAYS pick the physical cores to use. (this is why my graph only shows 1-4 cores even though HT is on, I couldn't use logical ones.)
Well here is the picture, look for yourself and see what you get...
Next we have a graph showing the effects of having other programs running while gaming. This could be converting files, encrypting, scanning for viruses, spyware, burning a DVD, playing a HD movie on another screen, WHATEVER. Check out how well an i7 with HT can handle MANY HEAVY processes running at once...
Note: I switched to using IBT for *other processes* because it was a little easier to use and provided a heavier processes not letting my tests interfere as much...
Edited by Finny - 8/25/09 at 9:51pm
The acrylic beast
(13 items) |
|
The acrylic beast
(13 items) |
