All of this supports my theory that dual core chips will not help SETI crunching.

Dual cores don't suffer from the same bottle necks as 2 logical CPUs.

---

Grant
Darwin NT

Reuben, would a $280 3.0Ghz Xeon 2MB chip out-crunch a $250 2.8GHz Pentium D 2MBx2 chip? Two Intel cores instead of one, each with the same cache and running at about the same speed, for a lower price?

From the World Wide SETI forums, we have discovered that a bottleneck with memory transfers and dual core chips running SETI. SETI is a memory intensive application, we cannot directly watch the memory transfer rates. However running SETI vs (CPDN or SuperPi benchmark) I find average memory temps for SETI to be 2 degrees higher!

Additionally, if I turn hyperthreading on with one chip installed in my 3.0ghz Irwindale I get WU averages of 29 minutes. However with two cpus and hyperthreading on my WU times jump to 90+ minutes. I have tested this combination a number of ways and cannot increase my WU times with 4 logical cpus. All of this supports my theory that dual core chips will not help SETI crunching.

I believe Cache on Intel chips do help. If I down clock my 2mb Xeon CPU to 2.4Ghz and overclock my 2.0ghz Xeon 512Kb, both running SSE2 applications, the 2mb Xeon wins (on average) by 15 minutes for a WU time of 28 minutes. I cannot directly prove this because of the core optimizations that occur with each new release of the Intel processor. I am eager to find out how fast the single core Xeon MP's with 4mb of cache process a WU, using opimized SETI applications.

How well would a 2.8GHz Celeron do, with a single core and only 256MB of cache?

Our team averages 65 minute processing times with a Celeron D's.

Also... why can't anyone use an AMD compiler to create more optimized apps for Athlon 64 and Opteron?

The AMD compiler costs about $4,000 last time I checked six months ago.
Two things upset me about the compiler optimizing:
1) Crunch3r still appears to have not been compensated by the community for paying for his Intel Compiler, IE: the link to paypal at the bottom of his webpage.

2) AMD charges $4,000 for their compiler. If they are so competitive with Intel... they should offer the compiler to one of the programmers in the SETI community. Why? Chip sales. :P

Hope that answered your questions...

---

Overclock with the MSI G31M3-L and Intel E8600 3.33Ghz
Intel D865GLC Socket 478 Motherboard
~How To Overclock The Eee ASUS 1005HA Netbook To 1.9Ghz~

Weren't the pentium m's based on the p3 core which is why they're much faster than a p4 clock per clock. If only there was a way back then intel could have extended the clock speed of the p3 to over 1300MHz there wouldn't have been a way amd could have competed clock per clock. Not forgetting the tualitins that are probably 20% faster than the coppermines clock per clock. Just a shame that the structure of the p4 meant to mislead buyers buying the clock speed and not the performance.

Ive got a dual 866 p3 that has a higher rac than an xp2700 amd (330/312 RAC)so 1700mhz of p3 beats 2170mhz of amd. It's strange that a design from 1999 can beat amd chip designed in 2001 yet the latest amd will beat intel in all now except seti.

---

[quote]Cache is not shared between cores.

Good point. That would explain why the SETI performance of the Pentium M laptops (single core, 2 MB x1 cache) seem to be at least as good as the Core Duo (dual core, 1 MB x2 cache) laptops.


No - I think that the Intel notebook CPU's are simply more efficient (more work per clock) than current P4. Just as AMD does more work per clock.
Intel has abandoned the P4 for future CPUs. P4 turned out to be dead-end because of power consumption (and resultant heat dissipation requirement).
Future Intel CPUs are being based in Intel Isralai Team design which I understand owes a lot to P3.

Because there isn't one.

There is no AMD compiler? Then why is SETI@home the only program in the world that is more optimized for Intel than AMD? Virtually everything else runs faster on AMD, so they must be better optimized for AMD somehow.

---

..... Also... why can't anyone use an AMD compiler to create more optimized apps for Athlon 64 and Opteron?

Because there isn't one.

---

AMD chips cannot compete with optimized SETI applications...
in other words, a $280 3.0ghz Xeon 2mb chip can crunch a WU in 15 minutes.
A $400 AMD 2.75 Ghz chip can do the same in 38 minutes.

Reuben, would a $280 3.0Ghz Xeon 2MB chip out-crunch a $250 2.8GHz Pentium D 2MBx2 chip? Two Intel cores instead of one, each with the same cache and running at about the same speed, for a lower price? How well would a 2.8GHz Celeron do, with a single core and only 256MB of cache?

Anybody have a S@h Celeron or two they can link to that are crunching pretty fast?

Also... why can't anyone use an AMD compiler to create more optimized apps for Athlon 64 and Opteron?

---

Just to point the obvious... AMD chips cannot compete with optimized SETI applications, because the Intel compiler used to create the optimized apps has specific machine code enhancements aimed at Intel chips.

In other words, a $280 3.0ghz Xeon 2mb chip can crunch a WU in 15 minutes.
A $400 AMD 2.75 Ghz chip can do the same in 38 minutes.

---

Overclock with the MSI G31M3-L and Intel E8600 3.33Ghz
Intel D865GLC Socket 478 Motherboard
~How To Overclock The Eee ASUS 1005HA Netbook To 1.9Ghz~

But now I understood why Pen-M is nice. (though I think dual-core is better :)) I would give Pen-D a go if I had a chance.

The new Pentium M based CPUs (now called Core) are Dual Core (Core Duo), although apparently there will be single core ones eventually available for very cheap systems.
See here for a comparative reveiw.

---

Grant
Darwin NT

Wow, I didn't know there was a 4mb cache Pentium D! That would explain why so many Pentium D's are at the top of the performance charts... I thought it was just because they had the same 2mb cache as Pentium M but faster clockspeeds than the mobiles. Pentium D 9x0 is thus what I would invest in, too!

Being dual core parts, they each have a 2 MB cache. Cache is not shared between cores.

But considering the clock speed of Pen-M, doesn't Pen-D 9x0 have much advantage? And that in Pen-D, 2 processes of S@H don't switch between 2 cores every time slice. So it should be much better than Pen-M in that Pen-D is dual-core and than AMD in cache size.

But now I understood why Pen-M is nice. (though I think dual-core is better :)) I would give Pen-D a go if I had a chance.

---

Luckiest in the world. WMD = Weapon of Mass Distraction.
Click this table.

Since yesterday I have a P IV D 930.

It rocks. 18-25 Minutes for 2 WUs.

Same Price as P IV D 830 but 2x2MB and not 2x1MB L2 Cache

There is one PIV D 9xx Extreme Edition with 1066 FSB
Dual Core AND HT.

But the price is 4 times higher as for the 930.

---

Cache is not shared between cores.

Good point. That would explain why the SETI performance of the Pentium M laptops (single core, 2 MB x1 cache) seem to be at least as good as the Core Duo (dual core, 1 MB x2 cache) laptops.

L2 Cache is shared on Core Duo

---

Join team BOINCstats

Cache is not shared between cores.

Good point. That would explain why the SETI performance of the Pentium M laptops (single core, 2 MB x1 cache) seem to be at least as good as the Core Duo (dual core, 1 MB x2 cache) laptops.

Cache is shared between cores sometimes, or at least it will be. Check out the new "Tulsa" Xeons for multiprocessor systems that have 16mb of shared cache per dual-core processor. In the standard config it will have 4 processors, 8 cores, and a whopping 64 MB (16 MB x 4) of cache! And to think that I have only a 1 MB cache on my AMD 64.

---

Pen-D 9x0 series must be very fast because they have 2 * 2mb cache, and that much cheaper than 840EE

Wow, I didn't know there was a 4mb cache Pentium D! That would explain why so many Pentium D's are at the top of the performance charts... I thought it was just because they had the same 2mb cache as Pentium M but faster clockspeeds than the mobiles. Pentium D 9x0 is thus what I would invest in, too!

Being dual core parts, they each have a 2 MB cache. Cache is not shared between cores.

---

Without love, breath is just a clock ... ticking.
Equilibrium

Pen-D 9x0 series must be very fast because they have 2 * 2mb cache, and that much cheaper than 840EE

Wow, I didn't know there was a 4mb cache Pentium D! That would explain why so many Pentium D's are at the top of the performance charts... I thought it was just because they had the same 2mb cache as Pentium M but faster clockspeeds than the mobiles. Pentium D 9x0 is thus what I would invest in, too!

Yes, but be careful not to buy 8x0 series :) Refer to Intel's site.

And another problem is in boinc often doesn't report AMD's clock speed, and cannot compare.

---

Luckiest in the world. WMD = Weapon of Mass Distraction.
Click this table.

Pen-D 9x0 series must be very fast because they have 2 * 2mb cache, and that much cheaper than 840EE

Wow, I didn't know there was a 4mb cache Pentium D! That would explain why so many Pentium D's are at the top of the performance charts... I thought it was just because they had the same 2mb cache as Pentium M but faster clockspeeds than the mobiles. Pentium D 9x0 is thus what I would invest in, too!

---

what is the fastest CPU for SETI and needs not much Power ??

Actually, there is an absolute win here! Please see this message I just posted in another thread to see why, but a 1600 MHz mobile Intel will outperform a 3500+ AMD on SETI@home, and it will also draw less power. Some versions of this chip use only 15 watts! Jim is right though about the desktop Intels, they will drink more power than a desktop AMD. So if you aren't in the market for a notebook computer, you'll sacrifice utility bills for the performance.

I've been wondering why Pen M is so fast. AFAIK, it's because of large L2 cache (2Mb). If it's the only reaosn, Pen-D 9x0 series must be very fast because they have 2 * 2mb cache, and that much cheaper than 840EE (of couse slower than EE). If not, what makes so much differences between Pen-M and P4 other than cache size?

If I'm to buy another one, I'd buy Pen-D 9xx (2.8G or 3.0G). It's fast enough for personal use or even for an usual server.

EDIT: as for comparison between Intel & AMD, optimized clients prefer Intel because SIMD (SSE/SSE2/SSE3) instructions consume nearly the same clock count between Intel & AMD; usually AMD runs at a lower clock rate (for ex AMD 3200+ runs at 2G+ or so) so as for SIMD instructions, Intel is clearly faster. But overall performance is another thing. Use at your own risk and credition.

what is the fastest CPU for SETI and needs not much Power ??

Actually, there is an absolute win here! Well, sort of. Please see this message I just posted in another thread to see why, but a 1600 MHz mobile Intel will outperform a 3500+ AMD on SETI@home, and it will also draw less power. Some versions of this chip use only 15 watts! Jim is right though about the desktop Intels, they will drink more power than a desktop AMD. So if you aren't in the market for a notebook computer, you'll sacrifice utility bills for the performance.

---

what is the fastest CPU for SETI and needs not much Power ?? AMD dualcore or Intel EE ?

cu

No absolute win there. AMDs suck up less juice but don't crunch as fast as the latest Intels and, of course, vice versa. If you can afford an EE is the extra $100 or so a year in power that big a deal?

---

Without love, breath is just a clock ... ticking.
Equilibrium

what is the fastest CPU for SETI and needs not much Power ?? AMD dualcore or Intel EE ?

cu