Intel Atom Hyperthreading test
I did a simple test to try to see how much benefit one really gets from having two logical CPUs on the single core Intel Atom 270 (in the Dell Mini 9). It seemed, after all, like it would be basically a useless addition to a processor that is aggressively low power -- would there really be enough functional units to sustain a reasonable amount of multithreading?
One interesting note is that, like genuine imitation naugahyde, this isn't even real hyperthreading. Rather than simultaneous multithreading on a out of order core, it is (as near as I can tell) an in-order superscalar with two pipelines presented as two logical processors.
To test, I did some video encoding, which generally is a good way to load up a CPU using a theoretically real application. Though this isn't exactly the load that was intended for a netbook (browsing, email, and streaming porn), but perhaps at least it could show if there's any theoretical possible benefit at all.
Before starting encoding I played the whole source file through, to increase the likelyhood that most if not all of the file would be in the cache (looking at the system cache rise by about the file size as it played indicates it was, though I didn't get a kernel debugger out to try to make sure :)
The test:
Input file 196.00MB XVID, 10776 frames
Output: H264 (x264 vfw codec) 800kbit single pass, all codec settings default. 1 thread and 2 threads, no thread queue.
Two threads: 4-6 FPS, 37 minutes
One thread: 3-4 FPS, 52 minutes
Certainly not a 2:1 speedup, but nearly 30% was far more than I expected.
One interesting note is that, like genuine imitation naugahyde, this isn't even real hyperthreading. Rather than simultaneous multithreading on a out of order core, it is (as near as I can tell) an in-order superscalar with two pipelines presented as two logical processors.
To test, I did some video encoding, which generally is a good way to load up a CPU using a theoretically real application. Though this isn't exactly the load that was intended for a netbook (browsing, email, and streaming porn), but perhaps at least it could show if there's any theoretical possible benefit at all.
Before starting encoding I played the whole source file through, to increase the likelyhood that most if not all of the file would be in the cache (looking at the system cache rise by about the file size as it played indicates it was, though I didn't get a kernel debugger out to try to make sure :)
The test:
Input file 196.00MB XVID, 10776 frames
Output: H264 (x264 vfw codec) 800kbit single pass, all codec settings default. 1 thread and 2 threads, no thread queue.
Two threads: 4-6 FPS, 37 minutes
One thread: 3-4 FPS, 52 minutes
Certainly not a 2:1 speedup, but nearly 30% was far more than I expected.