• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

FYI: SETI VLARs - HT P4 suffers more than dual Barton

I

IsOs

Diamond Member
It seems that the HyperThreading 2.4c P4 is more affected by very low AR than the dual Barton 2500+

The

dual Barton took only 30 minutes more to crunch a low angle workunit while the HyperThreading 2.4c P4 took as much as 1 hour 15 minutes more to crunch a very low angle workunit.

Both were running Windows 2000 Pro.

I don't know if this is an unusual case or if it's the expected outcome. Anyway, I thought I share it with you all.🙂
 
IsOs,

What speed are you running the P4 2.4C? Are you running Dual Channel Ram?

I'll check my WU times when I get home...I know that the best is around 2:20 on the P4 3.2GHz. I'm not sure what the worst...but I think it is also around 1 hour + for the low angle WU's. Probably due to the fact that the Barton has "Physical" hardware x 2 and the P4 has "logical" hardware in that it can only route the processes to certain parts of the chip and to whether they are being used at that time or not....i.e. the low angle WU's, when running with another WU (since it's 2 instances of SETI running), may need the same part of the processor as the other WU running. Confused? so am I! :Q
 
Originally posted by: Engineer
I'll check my WU times when I get home...I know that the best is around 2:20 on the P4 3.2GHz.
Click to expand...

Well, FWIW, my humble AMD 2100+ Tbred -- modestly overclocked to 2.1 Ghz, but running on an FSB of 192 -- has a median time of about 2:30 to 2:40 for normal AR's, and times of 2:00 to 2:10 for wu's with AR > 1. In the first three full days of production since getting back to crunching, she's produced 33 wu's, 11 a day, for an average time of 2:11 per WU.

I'm anxious to compare these results to the stock P4 2.4 Ghz's we are running at school when I get in later today.

I think seti@home really likes fast ram and high fsb's as much as raw CPU power.

-baz
 
Yep it certainly does ,though the XPs better IPC make them faster crunches at the same clock speed ,course the P4 can clock much higher😉

Btw Baz ,P4's are much faster if they use the Dual channel DDR RAM ,next fastest is RDRAM followed closely by Single channel DDR.
 
Originally posted by: blcjr
Originally posted by: Engineer
I'll check my WU times when I get home...I know that the best is around 2:20 on the P4 3.2GHz.
Click to expand...

Well, FWIW, my humble AMD 2100+ Tbred -- modestly overclocked to 2.1 Ghz, but running on an FSB of 192 -- has a median time of about 2:30 to 2:40 for normal AR's, and times of 2:00 to 2:10 for wu's with AR > 1. In the first three full days of production since getting back to crunching, she's produced 33 wu's, 11 a day, for an average time of 2:11 per WU.

I'm anxious to compare these results to the stock P4 2.4 Ghz's we are running at school when I get in later today.

I think seti@home really likes fast ram and high fsb's as much as raw CPU power.

-baz
Click to expand...

Ah...but remember, the HT enabled P4 3.2 is running two WU's at a time. It averages 2:56 per WU x 2 on a single processor = approximately 1:28 when you considering 2 WU's. It is averaging around 16-17 WU's per day. Hit 18 one time! 🙂

 
Originally posted by: Engineer
IsOs,

What speed are you running the P4 2.4C? Are you running Dual Channel Ram?

I'll check my WU times when I get home...I know that the best is around 2:20 on the P4 3.2GHz. I'm not sure what the worst...but I think it is also around 1 hour + for the low angle WU's. Probably due to the fact that the Barton has "Physical" hardware x 2 and the P4 has "logical" hardware in that it can only route the processes to certain parts of the chip and to whether they are being used at that time or not....i.e. the low angle WU's, when running with another WU (since it's 2 instances of SETI running), may need the same part of the processor as the other WU running. Confused? so am I! :Q
Click to expand...

The 2.4c Pr is running at 205 FSB dual channel RAM. I'm using the lower priced Kingston PC3200 non registered value ram. I lowered the FSB to get it running stable in a non-airconditioned room whose temp goes as high as 98F. This one is housed in an inexpensive case with one case fan in the rear blowing out, one case fan in the front sucking in, one fan in the drive bay blowing to the ram and a power supply with 2 fans. CPU is using the retail cooler.

The dual Bartons are set to run at 150 FSB x 14 = 2100MHz. I'll probably have to lower the FSB once I put this system inside a case.
I'm using Kingston Registered PC2700 value ram.
 
Originally posted by: Engineer
Ah...but remember, the HT enabled P4 3.2 is running two WU's at a time. It averages 2:56 per WU x 2 on a single processor = approximately 1:28 when you considering 2 WU's. It is averaging around 16-17 WU's per day. Hit 18 one time! 🙂
Click to expand...

Duh, I should have caught that, given the comparison to a dual Barton.

Since I stopped teaching pc hardware a couple of years ago, I've had a hard time keeping pace with the changing technology. Every 18 months or so, I get the urge to build a new system, and that's usually when I catch up on the latest technology.

-baz

 
Seti loves large cache, fast fsb & memory-speed, but also low multiplier...

The low multiplier is the reason my old p2 is the fastest cruncher per MHz, even it's only running at 66 MHz memory-bus.
In the other end, I've got 2 amd-systems, there the fastest is according to p-rating 50% faster, but in seti only 18% faster. :frown:

As for VLAR's, with a little bit intelligent setup, these shouldn't be a problem. 😉
Since the new v3.08 is much faster than v3.03 on VLAR's, it's just to use this to crunch them. 🙂

I'm steering all VLAR's to the old p2-box, and coupled with NT4 this makes v3-08-VLAR's 8% faster than normal wu crunched with v3.03... 😀

Oh, and it looks like I've found the cut-off also, since AR 0.210 is fast and 0.235 is very slow.
 
Originally posted by: Rattledagger
Seti loves large cache, fast fsb & memory-speed, but also low multiplier...

The low multiplier is the reason my old p2 is the fastest cruncher per MHz, even it's only running at 66 MHz memory-bus.
In the other end, I've got 2 amd-systems, there the fastest is according to p-rating 50% faster, but in seti only 18% faster. :frown:

As for VLAR's, with a little bit intelligent setup, these shouldn't be a problem. 😉
Since the new v3.08 is much faster than v3.03 on VLAR's, it's just to use this to crunch them. 🙂

I'm steering all VLAR's to the old p2-box, and coupled with NT4 this makes v3-08-VLAR's 8% faster than normal wu crunched with v3.03... 😀

Oh, and it looks like I've found the cut-off also, since AR 0.210 is fast and 0.235 is very slow.
Click to expand...

I thought the new v3.08 is slower in processing workunits than v3.03. I'm still using v3.03.
 
V3.08 seems to like the VLAR's better than v3.03. Anymore i try to process any WU's with an AR less than 0.0xx with the v3.08 client. You can see how they compare running on the same computer.

v3.03
Average time for 1520 WU's- 3:24:46

v3.08
Average time for 134 WU's- 3:34:31



Here's some more testing i did, running the same WU with both clients on different computers. The 0.417 AR WU is the Ars benchmark WU.

Windows 98SE
3.413AR
v3.03 - 3:03:55
v3.08 - 3:05:26

0.834 AR
v3.03 - 3:21:33
v3.08 - 3:38:24

0.417AR
v3.03 - 3:35:24
v3.08 - 4:00:33

-----------
Windows XP Home
0.417AR
v3.03 - 3:30:23
v3.08 - 3:51:43

0.380
v3.03 - 3:39:37
v3.08 - 4:06:36

0.235
v3.03 - 3:37:59
v3.08 - 4:06:36

0.072AR
v3.03 - 3:45:54
v3.08 - 3:37:31
 
Just adding a little to Spacehead's times:
BTW, actually the most interesting thing to know he hasn't benchmarked, and that is how win9x performs VLAR's with the new client. 😉


The new v3.08 is slower on "normal" wu, or angle-range 0.2 to 1+
For "high" angle ranges, I've personally not seen any marked difference either faster or slower.
For VLAR, or more specifically, angle-range below 0.2, v3.08 is faster than v3.03. Under win2k, a v3.08-VLAR will use comparable time to a "normal" wu under v3.03, or a couple of % slower... It's atleast much faster than using v3.03 to crunch them. 🙂

And as already stated, it seems the cut-off is around 0.2.

Lastly, after I last year was forced to downgrade to NT4 to at all use a new hd in the old p2-machine, I found out NT4 was much faster on crunching VLAR's than win2k. For other wu, there wasn't any difference. Together with the new v3.08-boost, VLAR's is now 8% faster than normal v3-03-wu. 😀

 
Originally posted by: IsOs
Originally posted by: Rattledagger
Seti loves large cache, fast fsb & memory-speed, but also low multiplier...

The low multiplier is the reason my old p2 is the fastest cruncher per MHz, even it's only running at 66 MHz memory-bus.
In the other end, I've got 2 amd-systems, there the fastest is according to p-rating 50% faster, but in seti only 18% faster. :frown:

As for VLAR's, with a little bit intelligent setup, these shouldn't be a problem. 😉
Since the new v3.08 is much faster than v3.03 on VLAR's, it's just to use this to crunch them. 🙂

I'm steering all VLAR's to the old p2-box, and coupled with NT4 this makes v3-08-VLAR's 8% faster than normal wu crunched with v3.03... 😀

Oh, and it looks like I've found the cut-off also, since AR 0.210 is fast and 0.235 is very slow.
Click to expand...

I thought the new v3.08 is slower in processing workunits than v3.03. I'm still using v3.03.
Click to expand...

It is ,except at VLAR WUs which are the minority of WUs, best setup is just like RD has ,have a rig dedicated to crunching VLAR WUs ,using SETI you can direct VLAR WUs to a v3.08 client ,though a batch file method will ensure 3.08 gets only VLAR WUs.

Rattle dagger & Isos
Good info ,thanks🙂

 
Originally posted by: Assimilator1
That's wierd ,I'm sure a page I linked too the other day at MS said Win2k does support HT😕

Yes it does! ,look here! 😕 😕
Click to expand...


It's not that Win2k won't run HT...I don't think it's OPTIMIZED for it...so it may run slower than XP or XP-Pro! 🙂
 
Originally posted by: Rattledagger
Just adding a little to Spacehead's times:
BTW, actually the most interesting thing to know he hasn't benchmarked, and that is how win9x performs VLAR's with the new client. 😉


The new v3.08 is slower on "normal" wu, or angle-range 0.2 to 1+
For "high" angle ranges, I've personally not seen any marked difference either faster or slower.
For VLAR, or more specifically, angle-range below 0.2, v3.08 is faster than v3.03. Under win2k, a v3.08-VLAR will use comparable time to a "normal" wu under v3.03, or a couple of % slower... It's atleast much faster than using v3.03 to crunch them. 🙂

And as already stated, it seems the cut-off is around 0.2.

Lastly, after I last year was forced to downgrade to NT4 to at all use a new hd in the old p2-machine, I found out NT4 was much faster on crunching VLAR's than win2k. For other wu, there wasn't any difference. Together with the new v3.08-boost, VLAR's is now 8% faster than normal v3-03-wu. 😀
Click to expand...

You're right, i haven't benchmarked my Win 98SE computer with a VLAR WU. I've just assumed that Win XP would handle them better with either client. Win 9x just doesn't like the VLAR's in my experience.

On my particular computers, most WU's with an AR in the 0.1xx range process about the same with both clients. Once i get an AR of less the 0.100 is where i see the v3.08 client running noticabley faster.
Both my computer have an AMD XP1800@ stock speed, Abit KX7 mobo & Corsair PC 2700 RAM.

 
You must log in or register to reply here.

TRENDING THREADS

Back
Top