This may be the 1st conroe in seti.... 2,66 Ghz.... 4 logic units = 2core*2(HT) = 4? :>

nice times... 65 credit units in ~6000sec...

HERE

http://setiathome.berkeley.edu/show_host_detail.php?hostid=2458557

---

Pretty impressive.

I would guess this is a dual woodcrest xeon machine.

Regards Hans.

P.S: Still no sign of them in the online shops, grrmph.

---

That may be the fastest s@h client host currently crunching...

...and it is not even going at full speed yet since it does not look like it is running SSE optimized code.

---

That may be the fastest s@h client host currently crunching...

...and it is not even going at full speed yet since it does not look like it is running SSE optimized code.

If you look at the results, it is actually running at full speed with an optimized SSE3 setiapp.

---

_\|/_
U r s

Hmm, on average 34.3 CS/h per "core". :drool:

This should be 3292 CS/day for this 4-way system.

Also, for "normal" angle-range, averages 1.6h/wu, meaning 15 wu/day per core...
At these speeds, Seti_Enhanced-wu doesn't look long afterall...

That may be the fastest s@h client host currently crunching...

...and it is not even going at full speed yet since it does not look like it is running SSE optimized code.

If you look at the results, it is actually running at full speed with an optimized SSE3 setiapp.

*smacks self in forehead for being unobservant*
I stand corrected. Thanks.

In that case, I suspect this is =not= a 4 core Woodcrest system since a 2.66GHz 4C Woodcrest system rates to be even faster than this... (!)

---

With his Conroe he's doing just about as well (per thread) as my Pentium D950s with half-hour units (25 cobblestones per hour). However he's doing much better with the longer units, about 40 cobblestones per thread per hour conpared to my 14 per hour. And- he's got four threads to my two.

---

Conroes are FPU monsters compared to any previous CPUs - even Athlon64s, which for the past few years have been leading in this regard.

Also, their shared-cache approach instead of exclusive is getting them extra gains vs. other dual core architectures, even if the integrated memory controller isn't ready yet (next version, then...CSI).

All in all, on the Desktop front, it'll be the best for a while. It's a pity Intel won't be able to back it up with numbers (Core 2 will make up for less than 20% of their volume far into 2007).

But still - "precioussssssssssss", as Gollum would say. Damn, but I want one :o)

Regards

---

Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information

This may be the 1st conroe in seti.... 2,66 Ghz.... 4 logic units = 2core*2(HT) = 4? :>

nice times... 65 credit units in ~6000sec...

HERE

http://setiathome.berkeley.edu/show_host_detail.php?hostid=2458557

Conroe has 4MB of cache......2MB for each core. This computer says it only has 1MB of cache.

Thats a problem with BOINC all cpu's report 1 MB of Cache.

One reason why Gurra's computer is doing very well is that he's using Crunch3r's cruncher. I see, now, that his machine is número uno on the list of top computers with an RAC of over 2,000. But that doesn't explain nearly fully why BOTH of my P950s together are only doing about 1,100, even without a single erroneous unit in a week. They should be doing almost as well as his single Conroe, I should think. His machine eats up 60 to 64-cobblestone units whereas mine balk at them (4-1/2 hours), and especially the 56-cobblestone ones which take eight or nine hours. It could be that Corsair Value Select (667) memory which has a reported CAS of a huge 5. Or it might be the default cruncher.

---

One reason why Gurra's computer is doing very well is that he's using Crunch3r's cruncher. I see, now, that his machine is número uno on the list of top computers with an RAC of over 2,000. But that doesn't explain nearly fully why BOTH of my P950s together are only doing about 1,100, even without a single erroneous unit in a week. They should be doing almost as well as his single Conroe, I should think. His machine eats up 60 to 64-cobblestone units whereas mine balk at them (4-1/2 hours), and especially the 56-cobblestone ones which take eight or nine hours. It could be that Corsair Value Select (667) memory which has a reported CAS of a huge 5. Or it might be the default cruncher.

This machine has 2 "woodcrest" xeons, sporting a 1333 Mhz FSB :o)
(I'm guessing here, but that's the only plausible candidate...)

Intels pentium-M architecture always was pretty fast, but they improved it significantly with their new "core2" (conroe, woodcrest et al.)

Regards Hans

P.S: The optimized crunchers are significantly faster than the stock one.

---

Regards Hans

P.S: The optimized crunchers are significantly faster than the stock one.

I fully agree on that ;-)
(have a look at my 3 hosts - P4 Northwood & Prescott s.478)

Kiva

---

Greetings from Norway

Crunch3er & AK-V8 Inside

Gurra's host does not seem fast enough to be 2 2C 2.66GHz Woodcrests.
Especially since it is running SSE optimized apps.

If it is a 2S Woodcrest system, Gurra is not getting expected performance out of it.

---

On a related note: Dell is giving away xeons for free ATM.
(Offer ends Jun-27).

Methinks they're making room for the 51xx xeons :o)

Regards Hans

---

Gurra's host does not seem fast enough to be 2 2C 2.66GHz Woodcrests.
Especially since it is running SSE optimized apps.

If it is a 2S Woodcrest system, Gurra is not getting expected performance out of it.

I indeed think it could be a DP , 4 core Xeon system.
Crunch3r's ap is @ 25-40% faster than standard x86 S@H ap depending on the comparison.
Alex's G5 optimized ap is @ 80%-90% faster than standard S@H PPC ap.

If one were comparing apples to apples applications w/ similar efficiency improvements, Gurra's host would be blowing the G5 Quads into the weeds, RAC @ 5500-6000+. Plus, his RAC probably isn't done climbing yet as it is.

If a host running Crunch3r's 5.12 runs at 1/2 the RAC vs. Crunch3r's best non-enhanced ap, that would equate to each Woodcrest core at @ 1144 RAC under the old system. Don't recall seeing too many single CPU, single core, non HT hosts at stock clock speed w/ that high of RAC.

Can only imagine what that rig could do w/ an ap actually optimized for the new Core 2 architecture.
It would be awe-inspiring.

---

ICC 9.1 supports Core 2/Duo natively and will accelerate things even faster.

I'm still salivating at the prospect of getting one - I've been meaning to upgrade my box but have kept delaying first for X2s and then for Conroe.

They also overclock like mad on default voltage and air cooling.

BTW, Crunch3rs optimized app isn't the quickest Linux optimized code anymore ;)

My code runs between 2x to 4x faster on the same host than the default client does. Same as Alex', by the way. It's incredible what he did, my MacMini is a credit monster now.

---

Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information

ICC 9.1 supports Core 2/Duo natively and will accelerate things even faster.

I'm still salivating at the prospect of getting one - I've been meaning to upgrade my box but have kept delaying first for X2s and then for Conroe.

They also overclock like mad on default voltage and air cooling.

BTW, Crunch3rs optimized app isn't the quickest Linux optimized code anymore ;)

My code runs between 2x to 4x faster on the same host than the default client does. Same as Alex', by the way. It's incredible what he did, my MacMini is a credit monster now.

LOT's of Mac-fanatic forums have been speculating the last few weeks that the new PowerMac may be an Apple-exclusive 3GHZ Woodcrest. It would make perfect sense (which is exactly why the assumptions may be wrong in typical Apple counter-logic), since it would be the fullfillment of the 3GHZ promise, ala Intel instead of IBM, and a 1333 FSB would trump current 1.25GHZ. The marketing campaign would compare this event to the "2nd coming". A Quad model of this for $3500, and.....drool city!

Try to build a 4 core Woodcrest on a server board w/ similar features and other desktop multi-media hardware for same. New PowerMac may be well worth the wait : )

---

To put things into proper perspective, a 4C 2.66 Intel 5150 box running SSE 2/3 optimized apps (SSE4 optimizations would require the Core2 specific compiler support and libraries Simon mentioned) should be

2-4x faster per core than standard because it is running the optimized app

4x faster per core than a Core Duo due to having 2x as many FP/SSE units, each 128b wide rather than 64b wide (1cycle FP per FPU rather than 2cycles)

2.66/2.16= 16/13= ~1.23x faster per core than a T2600 due to clock rate.

2x faster aggregate speed than a Core Duo since it it has 4C rather than 2C

=> 2*(16/13)*4= 9.85x faster aggregate than the current fastest shipping Yonah.
...running a =non= optimized app.

=> 19.69x - 39.38x faster than above running optimized apps.
Memory bandwidth limitations are probably going to hold the performance increase down to closer to 19x rather than 39x under most circumstances.

A 2.16GHz T2600 Yonah has a stable 24x7x365 s@h RAC in excess of 900...

---

Yeah Intel is still using a fsb, Instead of something like Hyper Transport. Just think without the bottleneck how fast It would be. Oh well, what a shame.

Intel's new server chipset (i5000p et al) has 2 separate FSB interfaces, one per CPU, running at 1333Mhz.

Memory is FB-DIMM with 4 individual 666MHz channels.

This is a very clean setup with no artificial limitations coming from the chipset.
The memory bandwith is higher than for a similar opteron system.


Regards Hans

---

Yeah Intel is still using a fsb, Instead of something like Hyper Transport. Just think without the bottleneck how fast It would be. Oh well, what a shame.

Intel's new server chipset (i5000p et al) has 2 separate FSB interfaces, one per CPU, running at 1333Mhz.

Memory is FB-DIMM with 4 individual 666MHz channels.

This is a very clean setup with no artificial limitations coming from the chipset.
The memory bandwith is higher than for a similar opteron system.

Regards Hans

Only In Theory. In Practice, even the the mighty 5160 can only do ~5GBps memory bandwidth while the Opteron's weigh in at ~6GBps.

What Woodcrest =does= seem to have is enormous small footprint bandwidth.
L1 cache bandwidth is ~300-375GBps (!) {in contrast, a T2600 has ~45GBps of L1 cache bandwidth} and Woodcrest systems seem to have more RAM bandwidth than current similar Opteron systems up to ~48MB footprints.

All this may change once the AM2 based and especially the K8L based Opteron systems are available, but traditionally Intel does very well in the smaller footprint arena.

---