Akos' C40 thread
26 Mar 2006 7:06:17 UTC
Topic 190962
(moderation:
I have a new WU starting in 2 minutes - let you know how it goes. Athlon XP_M Barton core (512L2 cache) WinXP.
Michael R.
microcraft
"The arc of history is long, but it bends toward justice" - MLK
Akos' C40 thread
)
C40 is good for 386, 486, Pentium-I, Pentium-II, K5, K6...
The SSE optimised S39L is the best for Barton at the moment.
Akos F.
Yes, I just suspended the
)
Yes, I just suspended the work on that one. With C40, it was 59.01% completed an 44:46. Projected out, that makes it 33% slower than C39.
microcraft
"The arc of history is long, but it bends toward justice" - MLK
RE: Yes, I just suspended
)
Thanks. It will be good for a speed test. :-)
Ok I made 2 small checks to
)
Ok I made 2 small checks to :)
A P3 is about 21% slower with C40 than with C39L.
A P4 is about 23% slower with C40 than with C39L.
Anders n
I have measured these ratios
)
I have measured these ratios on the same system with the same wu:
(Duron 2,03GHz, Win2000)
REF: 1.00
C01: 2.00 (first attempt)
A36: 2.37 (PPro+)
C37: 2.44
C40: 2.76
S39L:3.74 (SSE, accuracy is a bit worse than previous)
D40: 3.89 (3DNow, accuracy is a bit better than SSE)
Early report: C40 about 13%
)
Early report: C40 about 13% faster compared to C37 on my T-Bird 1200 (crunched two comparable WUs to 25% each)
Is D40 available for public download?
Cheers!
Tau
RE: Is D40 available for
)
Not, yet.
The validation test is running now.
RE: I have measured these
)
Results for Workunits 1044,5 Hz
Host: AMD Duron 892 MHz = Duron 1200 without SSE
C37 = 14973 s
C37/C40 mixed = 14686 s
C40 = 13110 s
C40 maybe nearly 13 % faster
Chris
*Die Signatur befindet sich aus technischen Gründen auf der Rückseite dieses Beitrages!*
Hi Akosf, I have had
)
Hi Akosf,
I have had trouble with C40 on my AMD K7 CPU computer. It runs fine with 387 but gives a validate error with C40. The computer can be seen here
I have run WCPUID on this cpu and here is most of the information:
AMD Athlon (Model 4)
Socket A (Socket 462)
Family 6 Model 4 Stepping ID 4 Standard
7 Model 4 Stepping ID 4 Extended
L1 I-Cache 64K Byte L2 Cache 256K Byte
L1 D-Cache 64K Byte L2 Speed Full 1470 MHz
MMX Supported MMX+ Supported
SSE Not Supported 3D Now! Supported
SSE2 Not Supported 3DNow!+ Supported
SSE3 Not Supported
Memory 128 M Byte
This computer has been running well with 387 and I will return it to 387.
Is there any possibility you will put out a revised version of C40 to support the above family of CPUs?
BTW I have been using S39L on 2 other computers (Athlon XPs basically) and they are working very well on those. Thank you for your efforts!
Joe B
RE: Hi Akosf, Hi
)
Hi JoeB,
I know that you had trouble with C37 too, but both version have to be good for your Athlon. Only bigger difference between 387 and C37/C40 is the method of a sinus/cosinus calculation. Other changes don't affect the results. I'm a bit perplexed. So I knocked up a code to your canary. Could you do a test? to JoeB
This code uses the original sin/cos algorithm.
I hope you will like the new 3DNow! code. (If it will work...)