New Centrino - Benchmark Seems Too Low - Page 2

AP · 2003-06-10, 18:23

Hi all,

any news about Pentium-M´s PRIME95-performance? Will there be a special version of the programme which will run faster on Pentium-M than it does now?

Regards
Achim

Prime95 · 2003-06-10, 18:38

Quote:

any news about Pentium-M´s PRIME95-performance? Will there be a special version of the programme which will run faster on Pentium-M than it does now?

Pentium-M's poor performance remains a mystery. Without access to one it is impossible to investigate any further.

patrik · 2003-07-18, 13:46

I also bought one now, Inspiron 500m.
Below the speed is given as about 600 MHz, but when Prime95 is running it increases to 1494 MHz, i.e. the processor was in a slower state when Prime95 read its speed, but stepped up to full speed when the iterations started.
---------------------
Intel(R) Pentium(R) M processor 1500MHz
CPU speed: 597.54 MHz
CPU features: RDTSC, CMOV, PREFETCH, MMX, SSE, SSE2
L1 cache size: 32 KB
L2 cache size: 1024 KB
L1 cache line size: 64 bytes
L2 cache line size: 64 bytes
TLBS: 128
Prime95 version 23.5, RdtscTiming=1
Best time for 384K FFT length: 42.585 ms.
Best time for 448K FFT length: 51.091 ms.
Best time for 512K FFT length: 57.896 ms.
Best time for 640K FFT length: 71.241 ms.
Best time for 768K FFT length: 87.245 ms.
Best time for 896K FFT length: 104.281 ms.
Best time for 1024K FFT length: 117.193 ms.
Best time for 1280K FFT length: 149.718 ms.
Best time for 1536K FFT length: 183.071 ms.
Best time for 1792K FFT length: 217.146 ms.
Best time for 2048K FFT length: 244.745 ms.

Dresdenboy · 2003-08-15, 16:57

Maybe here we see a reason...

The full thread can be found here:
http://www.aceshardware.com/forum?read=105030620

some snippet:

Code:

BLAS 
Banias' implementation of SSE2 seems really really crummy for some reason. Here's the results: 

Pentium-M 1300 mhz: 
DGEMM 
795.63459 SSE2 scalar 
963.83745 SSE2 packed 
853.25131 compiler 
1060.59689 x87 
SGEMM 
2669.249 SSE 
795.5295 compiler

The values seem to be MFLOPs for matrix multiply (D=double precision, S=single precision) in ScienceMark 2.0. And I think "compiler" means pure C-code.

Who could test Prime95 on Pentium M with disabled SSE2?

DDB

patrik · 2003-08-15, 20:22

With SSE2:Intel(R) Pentium(R) M processor 1500MHz
CPU speed: 597.56 MHz
CPU features: RDTSC, CMOV, PREFETCH, MMX, SSE, SSE2
L1 cache size: 32 KB
L2 cache size: 1024 KB
L1 cache line size: 64 bytes
L2 cache line size: 64 bytes
TLBS: 128
Prime95 version 23.5, RdtscTiming=1
Best time for 384K FFT length: 43.602 ms.
Best time for 448K FFT length: 51.810 ms.
Best time for 512K FFT length: 58.217 ms.
Best time for 640K FFT length: 71.767 ms.
Best time for 768K FFT length: 88.046 ms.
Best time for 896K FFT length: 104.792 ms.
Best time for 1024K FFT length: 118.280 ms.
Best time for 1280K FFT length: 151.527 ms.
Best time for 1536K FFT length: 184.617 ms.
Best time for 1792K FFT length: 218.575 ms.
Best time for 2048K FFT length: 247.353 ms.

With CpuSupportsSSE2=0 in local.ini:Intel(R) Pentium(R) M processor 1500MHz
CPU speed: 597.55 MHz
CPU features: RDTSC, CMOV, PREFETCH, MMX, SSE
L1 cache size: 32 KB
L2 cache size: 1024 KB
L1 cache line size: 64 bytes
L2 cache line size: 64 bytes
TLBS: 128
Prime95 version 23.5, RdtscTiming=1
Best time for 384K FFT length: 40.883 ms.
Best time for 448K FFT length: 51.315 ms.
Best time for 512K FFT length: 57.026 ms.
Best time for 640K FFT length: 74.742 ms.
Best time for 768K FFT length: 93.959 ms.
Best time for 896K FFT length: 110.191 ms.
Best time for 1024K FFT length: 128.112 ms.
Best time for 1280K FFT length: 168.165 ms.
Best time for 1536K FFT length: 199.251 ms.
Best time for 1792K FFT length: 245.195 ms.
Best time for 2048K FFT length: 267.028 ms.

Dresdenboy · 2003-08-16, 18:45

Thank you.

Seems, that Banias is just hindered by it's clock. For FFTs up to 512K the non-SSE2 code is even a little bit faster. It's like for Opteron where SSE2 uses the same fast fmul/fadd units while on P4 SSE2 uses faster units than normal x87 code.

trif · 2003-08-16, 19:24

Seems like more than just the clock. How do the timings compare to a Northwood 1.6 GHz with and without SSE2?

sdbardwick · 2003-08-17, 02:20

P4 1600 Northwood with SSE2:
[code:1]
Intel(R) Pentium(R) 4 CPU 1.60GHz
CPU speed: 1614.45 MHz
CPU features: RDTSC, CMOV, PREFETCH, MMX, SSE, SSE2
L1 cache size: 8 KB
L2 cache size: 512 KB
L1 cache line size: 64 bytes
L2 cache line size: 128 bytes
TLBS: 64
Prime95 version 23.5, RdtscTiming=1
Best time for 384K FFT length: 23.685 ms.
Best time for 448K FFT length: 28.112 ms.
Best time for 512K FFT length: 32.148 ms.
Best time for 640K FFT length: 39.006 ms.
Best time for 768K FFT length: 47.399 ms.
Best time for 896K FFT length: 55.532 ms.
Best time for 1024K FFT length: 62.595 ms.
Best time for 1280K FFT length: 83.227 ms.
Best time for 1536K FFT length: 101.208 ms.
Best time for 1792K FFT length: 120.354 ms.
Best time for 2048K FFT length: 136.787 ms.
[/code:1]

With CpuSupportsSSE2=0 in local.ini:
[code:1]
Intel(R) Pentium(R) 4 CPU 1.60GHz
CPU speed: 1614.39 MHz
CPU features: RDTSC, CMOV, PREFETCH, MMX, SSE
L1 cache size: 8 KB
L2 cache size: 512 KB
L1 cache line size: 64 bytes
L2 cache line size: 128 bytes
TLBS: 64
Prime95 version 23.5, RdtscTiming=1
Best time for 384K FFT length: 70.493 ms.
Best time for 448K FFT length: 84.616 ms.
Best time for 512K FFT length: 94.171 ms.
Best time for 640K FFT length: 121.662 ms.
Best time for 768K FFT length: 152.774 ms.
Best time for 896K FFT length: 180.725 ms.
Best time for 1024K FFT length: 206.301 ms.
Best time for 1280K FFT length: 269.297 ms.
Best time for 1536K FFT length: 322.897 ms.
Best time for 1792K FFT length: 383.031 ms.
Best time for 2048K FFT length: 438.059 ms.[/code:1]

Dresdenboy · 2003-08-17, 18:45

And here a comparable Athlon system (found in the v23.5 benchmarks thread):

Quote:

Originally Posted by lycorn

AMD Athlon(TM) XP1800+
CPU speed: 1544.73 MHz
CPU features: RDTSC, CMOV, PREFETCH, MMX, SSE
L1 cache size: 64 KB
L2 cache size: 256 KB
L1 cache line size: 64 bytes
L2 cache line size: 64 bytes
L1 TLBS: 32
L2 TLBS: 256
Prime95 version 23.5, RdtscTiming=1
Best time for 384K FFT length: 43.030 ms.
Best time for 448K FFT length: 49.304 ms.
Best time for 512K FFT length: 53.280 ms.
Best time for 640K FFT length: 70.483 ms.
Best time for 768K FFT length: 84.218 ms.
Best time for 896K FFT length: 100.923 ms.
Best time for 1024K FFT length: 113.367 ms.
Best time for 1280K FFT length: 150.287 ms.
Best time for 1536K FFT length: 179.212 ms.
Best time for 1792K FFT length: 223.835 ms.
Best time for 2048K FFT length: 257.452 ms.

Here it looks like Banias is Intel's Athlon (that idea came up in some other board) ;) Clock for clock they are at equal speed using x87 code with a small gain for Banias running SSE2 code (thanks to easier register adressing and code density).

QuintLeo · 2003-08-18, 03:07

More like the Banias uses the P-III core with some enhancements, like SSE2 support and a larger cache.

The P-III was VERY close in most cases in IPC to the Athlon - small add one way on some stuff, small add the other on other stuff, typically less than a 5% speed diff overall.

Intel just had MAJOR problems getting the P-III archetecture to ramp up in speed (the original 1 Ghz P-III got recalled 'cause it had too many problems), so they decided to completely change the archetecture for the P-IV.

Dresdenboy · 2003-08-18, 06:47

Quote:

Originally Posted by QuintLeo

More like the Banias uses the P-III core with some enhancements, like SSE2 support and a larger cache.

I know. That was described in detail in some articles before the release of Banias to the public. It's better to say, it is based on the P-III core. They also modified the decoders (I think this was called "µOP Fusion") and changed everything so, that every unit can be dynamically switched on/off for power saving. So besides the faster FSB and larger cache these better decoder structures caused the higher SPEC scores clock for clock compared to P-III. It seems, they reached the same IPC level like the Athlon Palomino and newer cores.

DDB

2003-08-15, 16:57	#15
Dresdenboy Apr 2003 Berlin, Germany 19² Posts	Maybe here we see a reason... The full thread can be found here: http://www.aceshardware.com/forum?read=105030620 some snippet: Code: BLAS Banias' implementation of SSE2 seems really really crummy for some reason. Here's the results: Pentium-M 1300 mhz: DGEMM 795.63459 SSE2 scalar 963.83745 SSE2 packed 853.25131 compiler 1060.59689 x87 SGEMM 2669.249 SSE 795.5295 compiler The values seem to be MFLOPs for matrix multiply (D=double precision, S=single precision) in ScienceMark 2.0. And I think "compiler" means pure C-code. Who could test Prime95 on Pentium M with disabled SSE2? DDB

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
29.2 benchmark help	Prime95	Software	69	2017-05-23 23:49
Benchmark Estimate	Primeinator	Information & Answers	8	2009-06-11 23:39
Does anyone have i7 920? for Benchmark?	cipher	Twin Prime Search	2	2009-04-14 20:16
Not happy on Centrino	delta_t	NFSNET Discussion	7	2004-01-09 16:03
Centrino has problems with Prime95...	magicfan241	Software	1	2003-11-03 20:46

2003-06-10, 18:23	#12
AP Sep 2002 2×7 Posts	Hi all, any news about Pentium-M´s PRIME95-performance? Will there be a special version of the programme which will run faster on Pentium-M than it does now? Regards Achim

2003-07-18, 13:46	#14
patrik "Patrik Johansson" Aug 2002 Uppsala, Sweden 5²·17 Posts	I also bought one now, Inspiron 500m. Below the speed is given as about 600 MHz, but when Prime95 is running it increases to 1494 MHz, i.e. the processor was in a slower state when Prime95 read its speed, but stepped up to full speed when the iterations started. --------------------- Intel(R) Pentium(R) M processor 1500MHz CPU speed: 597.54 MHz CPU features: RDTSC, CMOV, PREFETCH, MMX, SSE, SSE2 L1 cache size: 32 KB L2 cache size: 1024 KB L1 cache line size: 64 bytes L2 cache line size: 64 bytes TLBS: 128 Prime95 version 23.5, RdtscTiming=1 Best time for 384K FFT length: 42.585 ms. Best time for 448K FFT length: 51.091 ms. Best time for 512K FFT length: 57.896 ms. Best time for 640K FFT length: 71.241 ms. Best time for 768K FFT length: 87.245 ms. Best time for 896K FFT length: 104.281 ms. Best time for 1024K FFT length: 117.193 ms. Best time for 1280K FFT length: 149.718 ms. Best time for 1536K FFT length: 183.071 ms. Best time for 1792K FFT length: 217.146 ms. Best time for 2048K FFT length: 244.745 ms.

2003-08-15, 20:22	#16
patrik "Patrik Johansson" Aug 2002 Uppsala, Sweden 425₁₀ Posts	With SSE2:Intel(R) Pentium(R) M processor 1500MHz CPU speed: 597.56 MHz CPU features: RDTSC, CMOV, PREFETCH, MMX, SSE, SSE2 L1 cache size: 32 KB L2 cache size: 1024 KB L1 cache line size: 64 bytes L2 cache line size: 64 bytes TLBS: 128 Prime95 version 23.5, RdtscTiming=1 Best time for 384K FFT length: 43.602 ms. Best time for 448K FFT length: 51.810 ms. Best time for 512K FFT length: 58.217 ms. Best time for 640K FFT length: 71.767 ms. Best time for 768K FFT length: 88.046 ms. Best time for 896K FFT length: 104.792 ms. Best time for 1024K FFT length: 118.280 ms. Best time for 1280K FFT length: 151.527 ms. Best time for 1536K FFT length: 184.617 ms. Best time for 1792K FFT length: 218.575 ms. Best time for 2048K FFT length: 247.353 ms. With CpuSupportsSSE2=0 in local.ini:Intel(R) Pentium(R) M processor 1500MHz CPU speed: 597.55 MHz CPU features: RDTSC, CMOV, PREFETCH, MMX, SSE L1 cache size: 32 KB L2 cache size: 1024 KB L1 cache line size: 64 bytes L2 cache line size: 64 bytes TLBS: 128 Prime95 version 23.5, RdtscTiming=1 Best time for 384K FFT length: 40.883 ms. Best time for 448K FFT length: 51.315 ms. Best time for 512K FFT length: 57.026 ms. Best time for 640K FFT length: 74.742 ms. Best time for 768K FFT length: 93.959 ms. Best time for 896K FFT length: 110.191 ms. Best time for 1024K FFT length: 128.112 ms. Best time for 1280K FFT length: 168.165 ms. Best time for 1536K FFT length: 199.251 ms. Best time for 1792K FFT length: 245.195 ms. Best time for 2048K FFT length: 267.028 ms.

2003-08-16, 18:45	#17
Dresdenboy Apr 2003 Berlin, Germany 19² Posts	Thank you. Seems, that Banias is just hindered by it's clock. For FFTs up to 512K the non-SSE2 code is even a little bit faster. It's like for Opteron where SSE2 uses the same fast fmul/fadd units while on P4 SSE2 uses faster units than normal x87 code.

2003-08-16, 19:24	#18
trif Aug 2002 2·101 Posts	Seems like more than just the clock. How do the timings compare to a Northwood 1.6 GHz with and without SSE2?

2003-08-17, 02:20	#19
sdbardwick Aug 2002 North San Diego Coun 821 Posts	P4 1600 Northwood with SSE2: [code:1] Intel(R) Pentium(R) 4 CPU 1.60GHz CPU speed: 1614.45 MHz CPU features: RDTSC, CMOV, PREFETCH, MMX, SSE, SSE2 L1 cache size: 8 KB L2 cache size: 512 KB L1 cache line size: 64 bytes L2 cache line size: 128 bytes TLBS: 64 Prime95 version 23.5, RdtscTiming=1 Best time for 384K FFT length: 23.685 ms. Best time for 448K FFT length: 28.112 ms. Best time for 512K FFT length: 32.148 ms. Best time for 640K FFT length: 39.006 ms. Best time for 768K FFT length: 47.399 ms. Best time for 896K FFT length: 55.532 ms. Best time for 1024K FFT length: 62.595 ms. Best time for 1280K FFT length: 83.227 ms. Best time for 1536K FFT length: 101.208 ms. Best time for 1792K FFT length: 120.354 ms. Best time for 2048K FFT length: 136.787 ms. [/code:1] With CpuSupportsSSE2=0 in local.ini: [code:1] Intel(R) Pentium(R) 4 CPU 1.60GHz CPU speed: 1614.39 MHz CPU features: RDTSC, CMOV, PREFETCH, MMX, SSE L1 cache size: 8 KB L2 cache size: 512 KB L1 cache line size: 64 bytes L2 cache line size: 128 bytes TLBS: 64 Prime95 version 23.5, RdtscTiming=1 Best time for 384K FFT length: 70.493 ms. Best time for 448K FFT length: 84.616 ms. Best time for 512K FFT length: 94.171 ms. Best time for 640K FFT length: 121.662 ms. Best time for 768K FFT length: 152.774 ms. Best time for 896K FFT length: 180.725 ms. Best time for 1024K FFT length: 206.301 ms. Best time for 1280K FFT length: 269.297 ms. Best time for 1536K FFT length: 322.897 ms. Best time for 1792K FFT length: 383.031 ms. Best time for 2048K FFT length: 438.059 ms.[/code:1]

2003-08-18, 03:07	#21
QuintLeo Oct 2002 Lost in the hills of Iowa 2⁶·7 Posts	More like the Banias uses the P-III core with some enhancements, like SSE2 support and a larger cache. The P-III was VERY close in most cases in IPC to the Athlon - small add one way on some stuff, small add the other on other stuff, typically less than a 5% speed diff overall. Intel just had MAJOR problems getting the P-III archetecture to ramp up in speed (the original 1 Ghz P-III got recalled 'cause it had too many problems), so they decided to completely change the archetecture for the P-IV.