Hyperthreading

Blank Reg
Blank Reg
Joined: 18 Jan 05
Posts: 228
Credit: 40,599
RAC: 0

CPU type GenuineIntel

CPU type GenuineIntel Intel(R) Pentium(R) 4 CPU 3.00GHz
Number of CPUs 2
Operating System Microsoft Windows XP Professional Edition, Service Pack 2, (05.01.2600.00)
Memory 1022.66 MB
Cache 976.56 KB
Swap space 2464.75 MB
Total disk space 61.15 GB
Free Disk Space 47.75 GB
Measured floating point speed 1306.15 million ops/sec
Measured integer speed 1134.7 million ops/sec
Average upload rate 388.99 KB/sec
Average download rate 341.91 KB/sec
Average turnaround time 3.4 days

Time to complete 1 WU 38,000 t0 44,000

1206236 351668 18 Feb 2005 18:17:20 UTC 20 Feb 2005 4:34:55 UTC Over Success Done 37,955.00 53.97 91.03
1205840 351574 18 Feb 2005 15:29:56 UTC 19 Feb 2005 21:03:12 UTC Over Success Done 38,086.39 54.15 83.34
1180070 347373 17 Feb 2005 4:26:55 UTC 18 Feb 2005 18:17:20 UTC Over Success Done 38,891.92 56.48 82.49
1169305 345113 16 Feb 2005 4:54:39 UTC 18 Feb 2005 15:29:56 UTC Over Success Done 39,591.56 57.49 69.37
1164117 344166 15 Feb 2005 13:47:40 UTC 17 Feb 2005 16:28:26 UTC Over Success Done 39,314.23 55.55 86.15
1164112 344165 14 Feb 2005 22:48:12 UTC 17 Feb 2005 4:26:55 UTC Over Success Done 44,129.28 62.35 78.51

Jordan look at your results it takes you 84000 per CPU or 84000 x 2

1414471 399727 25 Feb 2005 22:20:07 UTC 27 Feb 2005 3:00:18 UTC Over Success Done 86,760.58 95.34 pending
1375850 390741 25 Feb 2005 19:26:24 UTC 27 Feb 2005 2:47:05 UTC Over Success Done 86,479.50 95.03 pending
1375838 390738 24 Feb 2005 21:27:38 UTC 26 Feb 2005 0:57:53 UTC Over Success Done 84,240.54 92.57 pending
1369564 389266 24 Feb 2005 17:11:25 UTC 26 Feb 2005 0:51:56 UTC Over Success Done 84,029.23 92.34 pending
1341214 382731 23 Feb 2005 20:14:12 UTC 25 Feb 2005 0:16:42 UTC Over Success Done 84,500.85 92.85 pending
1336678 381678 23 Feb 2005 16:16:33 UTC 25 Feb 2005 0:17:44 UTC Over Success Done 84,775.61 93.16 pending

Jordan Wilberding
Jordan Wilberding
Joined: 19 Feb 05
Posts: 162
Credit: 715,454
RAC: 0

> Jordan look at your results

Message 5206 in response to message 5205


> Jordan look at your results it takes you 84000 per CPU or 84000 x 2

It wouldnt be 84000 x 2, it would just be 84000 since they are both being run at the same time. Right?

such things just should not be writ so please destroy this if you wish to live 'tis better in ignorance to dwell than to go screaming into the abyss worse than hell

Jordan Wilberding
Jordan Wilberding
Joined: 19 Feb 05
Posts: 162
Credit: 715,454
RAC: 0

> > > Woah, I'd no idea how

Message 5207 in response to message 5202

> > > Woah, I'd no idea how slow the P4's were at this! A 2400+ Athlon
> I'm
> > running
> > > in Windows does ~22k secs per WU
> > > http://einsteinathome.org/host/33405
> >
> > I am just curious. I can only get around 40K secs per WU on my Athlon XP
> > 2400+. What would cause this?
>
> Motherboard quality
> Dual channel memory vs. single channel
> Memory speed
> Cache size
> Overclocking/aggressive timings vs. normal timings
> Bad luck
> Evil gremlins, Gnolls & Trolls

Everything I am running is fairly standard, I think I will try running a few other benchmark CPU tests to see what I find.

such things just should not be writ so please destroy this if you wish to live 'tis better in ignorance to dwell than to go screaming into the abyss worse than hell

Jordan Wilberding
Jordan Wilberding
Joined: 19 Feb 05
Posts: 162
Credit: 715,454
RAC: 0

> Jordan, > > It seems that

Message 5208 in response to message 5203


> Jordan,
>
> It seems that there is a serious configuration problem with your box. Here is
> an example of a similar box and its performance (take note of the Measured
> floating point speed
):

I downloaded the optimized boinc client, and now my benchmarks are similar to the norm. Only problem is, I doubt this will translate much for actually einstein computations, since the einstein app will not be optimized.

such things just should not be writ so please destroy this if you wish to live 'tis better in ignorance to dwell than to go screaming into the abyss worse than hell

Jordan Wilberding
Jordan Wilberding
Joined: 19 Feb 05
Posts: 162
Credit: 715,454
RAC: 0

> My AMD 1.3 GHz Athlon from

Message 5209 in response to message 5204


> My AMD 1.3 GHz Athlon from 2000 can almost keep pace with this monster CPU
> (Time to complete 1 WU: about 36,000)!

This is true for me as well!

such things just should not be writ so please destroy this if you wish to live 'tis better in ignorance to dwell than to go screaming into the abyss worse than hell

john.mac
john.mac
Joined: 9 Feb 05
Posts: 85
Credit: 167,393
RAC: 0

This cheapo celeron is on par

This cheapo celeron is on par with this P4 monster, doing about 34K per WU.
Al with standard client !!! and XPhome SP2.

CPU type GenuineIntel Intel(R) Celeron(R) CPU 2.66GHz
Operating System Microsoft Windows XP Home Edition, Service Pack 2,
Measured floating point speed 1264.2 million ops/sec
Measured integer speed 3972.4 million ops/sec

Greetings,

John,

[SETI.USA]Tank_Master
[SETI.USA]Tank_...
Joined: 22 Jan 05
Posts: 4
Credit: 15,212,681
RAC: 0

im running a P4 prescott 2.8

im running a P4 prescott 2.8 @ 3-3.2 (depending on what day and what i feel like :Þ) Server 2003 and 1 GB dual channel DDR 400 memory. I have Boinc running on a differnt HD than my OS is. HT IS enabled and I participate in 5 other projects.

(memory/fsb always 1:1)
when running @ 3.36GHz:
2005-02-22 12:14:40 [---] Running CPU benchmarks
2005-02-22 12:14:40 [---] Suspending computation and network activity - running CPU benchmarks
2005-02-22 12:15:39 [---] Benchmark results:
2005-02-22 12:15:39 [---] Number of CPUs: 2
2005-02-22 12:15:39 [---] 1419 double precision MIPS (Whetstone) per CPU
2005-02-22 12:15:39 [---] 1763 integer MIPS (Dhrystone) per CPU
2005-02-22 12:15:39 [---] Finished CPU benchmarks
2005-02-22 12:15:39 [---] Resuming computation and network activity

When running@ default 2.8
2/27/2005 6:25:06 PM||Running CPU benchmarks
2/27/2005 6:25:06 PM||Suspending computation and network activity - running CPU benchmarks
2/27/2005 6:26:05 PM||Benchmark results:
2/27/2005 6:26:05 PM|| Number of CPUs: 2
2/27/2005 6:26:05 PM|| 1160 double precision MIPS (Whetstone) per CPU
2/27/2005 6:26:05 PM|| 1482 integer MIPS (Dhrystone) per CPU
2/27/2005 6:26:05 PM||Finished CPU benchmarks

1356804 386317 24 Feb 2005 8:23:11 UTC 27 Feb 2005 15:57:21 UTC Over Success Done 61,323.92 92.76 92.76
1314918 376687 22 Feb 2005 23:44:31 UTC 25 Feb 2005 1:38:58 UTC Over Success Done 39,104.89 69.87 82.40
1206489 351729 18 Feb 2005 12:30:49 UTC 19 Feb 2005 16:56:38 UTC Over Success Done 44,070.17 75.08 80.53
1201776 339966 19 Feb 2005 21:31:41 UTC 20 Feb 2005 16:22:48 UTC Over Success Done 39,664.78 67.57 67.57
1169493 345154 14 Feb 2005 18:52:36 UTC 17 Feb 2005 4:09:17 UTC Over Success Done 42,251.75 72.28 78.87
1146144 340506 11 Feb 2005 21:50:32 UTC 13 Feb 2005 8:41:12 UTC Over Success Done 41,934.52 87.75 85.26
1143879 340008 18 Feb 2005 1:17:55 UTC 19 Feb 2005 0:03:03 UTC Over Success Done 40,767.88 69.74 80.98
1127077 335887 16 Feb 2005 5:14:16 UTC 18 Feb 2005 1:17:55 UTC Over Success Done 43,029.91 73.61 76.66
1123247 334943 14 Feb 2005 0:54:36 UTC 15 Feb 2005 15:24:51 UTC Over Success Done 59,409.44 101.63 85.83
1119258 333957 12 Feb 2005 17:58:43 UTC 14 Feb 2005 0:31:16 UTC Over Success Done 40,730.30 85.23 86.61
1115162 333038 10 Feb 2005 20:36:41 UTC 12 Feb 2005 17:58:43 UTC Over Success Done 40,291.41 84.31 94.33

Paul D. Buck
Paul D. Buck
Joined: 17 Jan 05
Posts: 754
Credit: 5,385,205
RAC: 0

We are getting apples and

We are getting apples and oranges mixed here.

The speeds detected and reported for the BOINC system have nothing to do (sorry, but it is the truth) with the actual processing speed of a system for a project.

There is a significant difference between platforms/operating systems having more to do with the compilers and the actual benchmarks running on the same physical hardware.

This is an issue with the reality of synthetic benchmarks. This is discussed at length in the lecture notes on performance.

Actual performance processing work is affected by the factors that I listed.

john.mac
john.mac
Joined: 9 Feb 05
Posts: 85
Credit: 167,393
RAC: 0

I know that benchmarking is

I know that benchmarking is very subjective, was wondering though that topline Intel pentium proc. do not appear to be faster, or is WU selection so much fine grained that such processors are loaded with the bigger WU's compared to say a Celeron D330 ??
If thats true apples and oranges are mixed indeed even on processing times.

Thanks for your nice doc's about the subject Paul,

Regards,

John,

Paul D. Buck
Paul D. Buck
Joined: 17 Jan 05
Posts: 754
Credit: 5,385,205
RAC: 0

> I know that benchmarking is

Message 5214 in response to message 5213

> I know that benchmarking is very subjective, was wondering though that topline
> Intel pentium proc. do not appear to be faster, or is WU selection so much
> fine grained that such processors are loaded with the bigger WU's compared to
> say a Celeron D330 ??
> If thats true apples and oranges are mixed indeed even on processing times.

There are no simple answers. To any of this. If we look at two processors that are "identical" in all respects with the execption of the cache size and cache line size (don't sweat it, just bear with me). When we run EAH only on the two CPUs we will see a typical performance. If we start to run two projects we would expecct that the one with the larger cache would be more effective, and most of the time this is true, but it is also possible that the smaller cache with the smaller line size has a lower "miss-penalty" and so, in effect, it could give better performance ...

Intel has been extending the length of the pipeline. *IF* you do not have a pipeline "stall" this increases the effective speed of the processor. If you do have a pipeline stall, well, it is not pretty ... :(

The resaon that the AMD chips may be outperforming the INtel chips at the moment is from these kinds of interactions ...

Celerons have too small of a cache and their performance in BOINC projects suffers from this.

Engineering is the "science" of choices. Make bad ones, and you get bad outcomes, make poor ones and you get poor outcomes. The problem is that you don't know which are which until the road meets the rubber ...

One of the reasons that the Xeons may have better performance than some of the P4s is that (at least the ones I know of, this may not be true for the lastest generations) they are using a P3 core that seems to be more effecient in many respects than the P4's core.

>
> Thanks for your nice doc's about the subject Paul,

You are welcome. While I was still able to teach it took me from 32 to 64 hours per contact hour to write my notes, exclusive of the research time to find the source material and the time to read it when found. That is the reason I wanted to get them back out there ... what the heck, they are already written ...

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.