Did anybody get an AMD Phenom II X6 1055T so far?

[odin]

Originally Posted by joblack

Here we go:

http://mersenneforum.org/showthread.php?t=13185

How do I get mPrime to process this one exponent on all 6 cores? I only know how to pass it to one core. What do I put in the worktodo.add file? Any special changes to prime.txt and local.txt?

Thanks.

[Mini-Geek]

Originally Posted by odin

How do I get mPrime to process this one exponent on all 6 cores? I only know how to pass it to one core. What do I put in the worktodo.add file? Any special changes to prime.txt and local.txt?

Thanks.

You don't do anything special to any of the files. Tell mprime to run the worker that is running (or will run) the number to use all 6 cores, and make it the only worker (or pause all the other workers while doing this number - you don't want mprime to be fighting itself for cores when a simple setting change could fix that). Do this from mprime's menu. Run "./mprime -m" and go to Test > Worker Windows, and follow the prompts (specifically, look for "Number of worker windows to run" and "CPUs to use (multithreading)").

[joblack]

Originally Posted by Mini-Geek

You don't do anything special to any of the files. Tell mprime to run the worker that is running (or will run) the number to use all 6 cores, and make it the only worker (or pause all the other workers while doing this number - you don't want mprime to be fighting itself for cores when a simple setting change could fix that). Do this from mprime's menu. Run "./mprime -m" and go to Test > Worker Windows, and follow the prompts (specifically, look for "Number of worker windows to run" and "CPUs to use (multithreading)").

For the benchmark you just have to use the 'Time' option. It will start with one thread and goes up to MAX_CORES (in your case 6). It only needs around 4 minutes.

After that you can copy/paste the benchmark results.

For real mprime searching you can setup up the worker/helper-thread distribution in the 'Worker' menu option.

[moebius]

Originally Posted by Batalov

Just as I expected, for the GA-MA790GP-UD4H the F7-beta BIOS is not doing anything yet. I've prodded GA's web support, no answer

As long as the X6-processors are not in the CPU Support List......
But, the GA-MA790GPT UD3H supports now X6 1035T and 1055T.
http://www.gigabyte.com.tw/Support/M...uctID=3143&ver=

[joblack]

Originally Posted by odin

I just bought the AMD Phenom II X6 1090T Black Edition on ebay for $278 OEM version. I'm going to use the Gigabyte GA-MA785GMT-UD2H motherboard with it.

When I get it an set everything up I'll post benchmarks. Give me about a week or 2.

Odin.

Did you get it?

[stars10250]

How important is the L2 and L3 cache with this processor compared to that of the Intel 980X...assuming one wants to run 6 LL simultaneously? It looks like the L2 is larger for the AMD cpu (512 Kb per core vs 256 Kb), but the L3 is smaller (6Mb vs 12 Mb) than that of the Intel cpu. Since these cache sizes are too small for prime95 to fit completely into, does it matter? The price difference between these two cpus is considerable (ok, ridiculous).

[joblack]

Originally Posted by stars10250

How important is the L2 and L3 cache with this processor compared to that of the Intel 980X...assuming one wants to run 6 LL simultaneously? It looks like the L2 is larger for the AMD cpu (512 Kb per core vs 256 Kb), but the L3 is smaller (6Mb vs 12 Mb) than that of the Intel cpu. Since these cache sizes are too small for prime95 to fit completely into, does it matter? The price difference between these two cpus is considerable (ok, ridiculous).

Prime95 fits into it but the attached 100m number isn't. It's a matter of clever cache management (pipelining, ...).

[stars10250]

Originally Posted by joblack

Prime95 fits into it but the attached 100m number isn't. It's a matter of clever cache management (pipelining, ...).

Yes, I meant that the calculation can't be done entirely in cache. So if it has to access main memory anyway, will it matter all that much (considering the differences between the AMD and Intel CPU)?

[mdettweiler]

Originally Posted by stars10250

Yes, I meant that the calculation can't be done entirely in cache. So if it has to access main memory anyway, will it matter all that much (considering the differences between the AMD and Intel CPU)?

Not to drag this off-topic or anything, but could someone please explain how CPU cache size relates to FFT sizes and whether a given number can be calculated in-cache? I've heard various talk here and there relating the two, but never was able to quite figure it out.

What I do know is that if a given number does NOT fit within cache, then the speed of the test depends primarily on the speed of the memory. I'm not entirely positive what it depends on when the number DOES fit within cache--is it the clock speed? The FSB?

The other thing I'm not clear on is exactly which cache (L1, L2, or L3) that is in question here, and just what "counts" as far as determining what fits within it. Is it just the size of the FFT, or is it the FFT + the size of the Prime95 executable? Or is it something else entirely?

Sorry to go a little off-topic, but since everyone's on the subject of LL tests and cache sizes I figured this was as good a time as any to ask.

[TheJudger]

Hi,

the FFT sizes are not in bytes, they are in doubles (double precision floating point variables). Each double needs 8 bytes.
So an 1280k FFT (current doublechecks) needs at least 10MiB memory plus some additional stuff.

Oliver

[mdettweiler]

Originally Posted by TheJudger

Hi,

the FFT sizes are not in bytes, they are in doubles (double precision floating point variables). Each double needs 8 bytes.
So an 1280k FFT (current doublechecks) needs at least 10MiB memory plus some additional stuff.

Oliver

Ah, that makes sense. So, I have a Core 2 Duo E4500 with 2 MB of L2 cache (which I presume is shared between the two cores). Therefore, any FFT less than about 256K should be able to fit in cache. Did I get that right?