@Mikey: sorry, what the hell are you talking about? Pentium 4? Integer results have to be exact anyway and for floating point there is an IEEE standard which both, Intel and AMD, have to adhere to. Otherwise they could just outright stop producing those CPUs. (there is a market for non-IEEE conform fp, but that's not the x86 world)
Actually Intel has had the advantage of "doing more work per clock" ever since the Core 2. Phenom II has caught up regarding price and performance/watt, but not performance/clock. Core i7 and brothers are in a league of their own.
Regarding the 24 core AMD machine: nice to see cores count, after all. However, really impressed I am not: we're at 25k - 26k seconds per GW WU per core at 2.4 GHz. My stock i7 920 is doing them at 24k - 25k seconds with HT on. So 3 of these quads (=24 threads) would be more productive than the 4 x 6-core Opterons. That's 2.2 billion Intel transistors and 789 mm² worth of precious wafer area versus 3.6 billion AMD transistors and 1384 mm² area. That doesn't get me celebrating ;)
@Matt: not sure what you're talking about. With GPUs the answer is almost always "it totally depends on the code and problem". Using the GPU as a co-processor is problematic, unless the chunks of work the CPU is sending to it are small, but compute-heavy, i.e. take a rather long time. In that case a realtively cheap GTX460 will be an order of magnitude faster than your cheap GPU, and so will be overall performance. And there's protein folding running entirely on the GPU. Here raw horse power counts, and not much else. However, if you're talking about the current Einstein app.. well, it hardly counts as "using the GPU", does it?
And to comment on FUD spread in other posts: no sane consumer measures
performance solely with design characteristics (read: number of cores,
die size, etc.). Car analogy: got a V8? Really? Well pitch it against
my turbocharged flat-4 in lightweight chassis :)
Question's always the same: how good it performs desired task
and for what money (and how much of one user is willing to sacrifice
for the other).
On a side note: acquisition cost/system power draw (TCO), performance
per Watt and _space_ (yes!) _may_ also be important (or even esssential)
to some.
Question's always the same: how good it performs desired task
and for what money (and how much of one user is willing to sacrifice
for the other).
There is wisdom in that. For those who have a glimmer how markets actually work it should not be surprising to find competitive offerings to be in rather close balance as seen by some appropriate summary function on the customer base. Otherwise the overpriced one will rapidly lose market share.
The trick (as I think you imply) is that one's own circumstance may differ rather greatly in the importance of various attributes in a way that makes one offering rather dramatically more attractive than the other. People deciding to build a system with BOINC computational contribution as a high-ranking priority differ rather considerably from the center of the aggregate consumer market, the personal use for business market, the server market, the high-end game-playing market, and most other major identifiable markets. Clusters such as the Atlas AEI Hannover probably come the closest in some ways, but even they differ greatly from a typical home Einstein participant in the way things should matter to them.
So I'd not be surprised on simple market theory grounds to find at one time or another the products of one manufacturer or the other to be found greatly preferable for SETI or Einstein participation by people with this interest who review more evidence than just the manufacturer's name.
I'm sorry but I don't know what exactly you want to say. Am I right in guessing that in your quote you're showing some pearl script (or something as cryptic) showing that your machine got only GC tasks for the last 24 WUs?
Do you know that under "your profile" / "Einstein@Home settings" / "Run only the selected applications" you can just opt out of ABP tasks? Doing so would increase the efficiency of your machine and wouldn't hurt, as we're currently handling these tasks just fine.
Quote:
no sane consumer measures performance solely with design characteristics
Did anyone say so?
Of course the hardware has to be judged by the performance in the given task. That's why we're discussing "is an amd cpu better performing than intel at einstein?" here. Judged by Performance along with TCO and maybe other special factors.
Quote:
Now, which one is better, again?
Why would we have to say it again? Did anything change based on what you said? I still see a host using 4 expensive 115W TDP CPUs producing about as much GC work as a dual socket Westmere with HT at ~2.7 GHz (see current machine 2 and 3 in top hosts list).
And to actually answer the question we'd have to know at least the TCO of those systems, going by what you just said ;)
And for personal crunching one would actually buy single socket systems instead. But then that's what we've been comparing previously, haven't we?
Just for the record my Opteron 1210 running Linux at 1.8 GHz completes a GC task in 36k s CPU time against 29k s of ATLAS AEI Hannover with its Xeon also on Linux.
Tullio
Those are old 65 nm Core 2 Quad with FSB 1066. Just the same as the venerable Q6600. Nice but many years old. If you compare to Pentium 4 based Xeons your Opti starts to shine even more ;)
I have used a PII Deschutes from 2000 to 2008 and the Opteron is faster. The Pentiums I have avoided, but the Cores are faster than the Opteron. But it never errors, except for some CPDN models, but it's not its fault.
Tullio
Well now i am crunching with Samsung i5 laptop, with Geforce GPU.. and i have to say: Prosessor is quite good, but GForce GPU cant do Collaz at all it seems.
No matter, quite strong regular work done. And i gave my old Acer to my best friend, i could have keeped it and got 10 - 20k credit from collaz, but that is not important. More important is my friends, it seems that i5 crunches happily normal tasks, i got now 1k credit income from einsten compared to 200 - 400.
>> Bargained for ye, by Ferengi Scavenger Ink. <<<
<<< Only thriving and employees still living, small business company in Nightcity >>>
@Mikey: sorry, what the hell
)
@Mikey: sorry, what the hell are you talking about? Pentium 4? Integer results have to be exact anyway and for floating point there is an IEEE standard which both, Intel and AMD, have to adhere to. Otherwise they could just outright stop producing those CPUs. (there is a market for non-IEEE conform fp, but that's not the x86 world)
Actually Intel has had the advantage of "doing more work per clock" ever since the Core 2. Phenom II has caught up regarding price and performance/watt, but not performance/clock. Core i7 and brothers are in a league of their own.
Regarding the 24 core AMD machine: nice to see cores count, after all. However, really impressed I am not: we're at 25k - 26k seconds per GW WU per core at 2.4 GHz. My stock i7 920 is doing them at 24k - 25k seconds with HT on. So 3 of these quads (=24 threads) would be more productive than the 4 x 6-core Opterons. That's 2.2 billion Intel transistors and 789 mm² worth of precious wafer area versus 3.6 billion AMD transistors and 1384 mm² area. That doesn't get me celebrating ;)
@Matt: not sure what you're talking about. With GPUs the answer is almost always "it totally depends on the code and problem". Using the GPU as a co-processor is problematic, unless the chunks of work the CPU is sending to it are small, but compute-heavy, i.e. take a rather long time. In that case a realtively cheap GTX460 will be an order of magnitude faster than your cheap GPU, and so will be overall performance. And there's protein folding running entirely on the GPU. Here raw horse power counts, and not much else. However, if you're talking about the current Einstein app.. well, it hardly counts as "using the GPU", does it?
MrS
Scanning for our furry friends since Jan 2002
RE: RE: Host # 2 now -
)
Things are looking better today (no trickery involved whatsoever) ;-)
$ ps auxc | awk 'BEGIN { ABP=0; GC=0 } /einstein_S/ { GC++ } /einsteinbina/ { ABP++ } END { print "ABP: " ABP ; print "GC: " GC }' ABP: 0 GC: 24 $And to comment on FUD spread in other posts: no sane consumer measures
performance solely with design characteristics (read: number of cores,
die size, etc.). Car analogy: got a V8? Really? Well pitch it against
my turbocharged flat-4 in lightweight chassis :)
Question's always the same: how good it performs desired task
and for what money (and how much of one user is willing to sacrifice
for the other).
On a side note: acquisition cost/system power draw (TCO), performance
per Watt and _space_ (yes!) _may_ also be important (or even esssential)
to some.
Now, which one is better, again?
RE: Question's always the
)
There is wisdom in that. For those who have a glimmer how markets actually work it should not be surprising to find competitive offerings to be in rather close balance as seen by some appropriate summary function on the customer base. Otherwise the overpriced one will rapidly lose market share.
The trick (as I think you imply) is that one's own circumstance may differ rather greatly in the importance of various attributes in a way that makes one offering rather dramatically more attractive than the other. People deciding to build a system with BOINC computational contribution as a high-ranking priority differ rather considerably from the center of the aggregate consumer market, the personal use for business market, the server market, the high-end game-playing market, and most other major identifiable markets. Clusters such as the Atlas AEI Hannover probably come the closest in some ways, but even they differ greatly from a typical home Einstein participant in the way things should matter to them.
So I'd not be surprised on simple market theory grounds to find at one time or another the products of one manufacturer or the other to be found greatly preferable for SETI or Einstein participation by people with this interest who review more evidence than just the manufacturer's name.
Tear, I'm sorry but I
)
Tear,
I'm sorry but I don't know what exactly you want to say. Am I right in guessing that in your quote you're showing some pearl script (or something as cryptic) showing that your machine got only GC tasks for the last 24 WUs?
Do you know that under "your profile" / "Einstein@Home settings" / "Run only the selected applications" you can just opt out of ABP tasks? Doing so would increase the efficiency of your machine and wouldn't hurt, as we're currently handling these tasks just fine.
Did anyone say so?
Of course the hardware has to be judged by the performance in the given task. That's why we're discussing "is an amd cpu better performing than intel at einstein?" here. Judged by Performance along with TCO and maybe other special factors.
Why would we have to say it again? Did anything change based on what you said? I still see a host using 4 expensive 115W TDP CPUs producing about as much GC work as a dual socket Westmere with HT at ~2.7 GHz (see current machine 2 and 3 in top hosts list).
And to actually answer the question we'd have to know at least the TCO of those systems, going by what you just said ;)
And for personal crunching one would actually buy single socket systems instead. But then that's what we've been comparing previously, haven't we?
MrS
Scanning for our furry friends since Jan 2002
Just for the record my
)
Just for the record my Opteron 1210 running Linux at 1.8 GHz completes a GC task in 36k s CPU time against 29k s of ATLAS AEI Hannover with its Xeon also on Linux.
Tullio
Those are old 65 nm Core 2
)
Those are old 65 nm Core 2 Quad with FSB 1066. Just the same as the venerable Q6600. Nice but many years old. If you compare to Pentium 4 based Xeons your Opti starts to shine even more ;)
MrS
Scanning for our furry friends since Jan 2002
I have used a PII Deschutes
)
I have used a PII Deschutes from 2000 to 2008 and the Opteron is faster. The Pentiums I have avoided, but the Cores are faster than the Opteron. But it never errors, except for some CPDN models, but it's not its fault.
Tullio
Would E@h release ati gpu
)
Would E@h release ati gpu crunching option ever?
See here. Cheers, Mike.
)
See here.
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
Well now i am crunching with
)
Well now i am crunching with Samsung i5 laptop, with Geforce GPU.. and i have to say: Prosessor is quite good, but GForce GPU cant do Collaz at all it seems.
No matter, quite strong regular work done. And i gave my old Acer to my best friend, i could have keeped it and got 10 - 20k credit from collaz, but that is not important. More important is my friends, it seems that i5 crunches happily normal tasks, i got now 1k credit income from einsten compared to 200 - 400.
>> Bargained for ye, by Ferengi Scavenger Ink. <<<
<<< Only thriving and employees still living, small business company in Nightcity >>>
>> If ye got the currency, we get job done! <<