In the November 2009 top500 list the first three systems are Opteron based, the fourth,seventh and eighth are PowerPc based, the fifth, sixth and tenth are Xeon based and the ninth is another Opteron. But wait for the new list in about a week.
Tullio
In the November 2009 top500 list the first three systems are Opteron based, the fourth,seventh and eighth are PowerPc based, the fifth, sixth and tenth are Xeon based and the ninth is another Opteron. But wait for the new list in about a week.
Tullio
Oh, the previous graph is based on Einstein@Home TOP 10000. :-)
In the November 2009 top500 list the first three systems are Opteron based, the fourth,seventh and eighth are PowerPc based, the fifth, sixth and tenth are Xeon based and the ninth is another Opteron. But wait for the new list in about a week.
Tullio
Oh, the previous graph is based on Einstein@Home TOP 10000. :-)
I know that the PCs running E@home cannot be compared with "big iron". But still, as I am running a low cost Opteron 1210 with 6 BOINC projects plus a Solaris virtual machine running a BOINC client and SETI app by a developer called Dotsch i still think of the AMD architecture as a good one, even if its marketing cannot compete with Intel's. Having worked for Honeywell-Bull Italy I have seen the power of marketing. I was glad to see a BULL machine working for the ITER fusion experiments. Cheers.
Tullio
Noone said it's a bad architecture. But all the advantages you name are actually just software - you could do that with an Intel just as well.
The problem we're having here with AMD is not its marketing but rather its performance at Einstein@home. A Phenom needs quite a bit more CPU clock to match a Core 2, let alone an i3/5/7. Sandy Bridge is going to increase performance per clock further, whereas Bulldozer is not going to be a FP monster (as needed for E@H).
However, the AMD 6-cores are an interesting offering. In general performance they are a good match to the equivalently priced Intel 4-cores, trading blows in highly threaded apps and loosing in lightly threaded ones and power consumption. However, BOINC can make very good use of the additional 2 cores, so the throughput should be quite competitive.
That looks to show that throughout, AMD has about a 18% market share until the top 1000 where it drops to about 10% market share.
That likely fits very closely with their production capacity vs Intel's production capacity! The rest is pricing and Marketing... (As shown by the kink at the top 4000 hosts point?)
That likely fits very closely with their production capacity vs Intel's production capacity! The rest is pricing and Marketing... (As shown by the kink at the top 4000 hosts point?)
Yes, 18% should be about the current market share, biased a bit to the high end market due to us selecting only reasonably high performing hosts. But the 10% due to pricing and marketing? Well.. that's a strange way of putting it.
A single Intel 6-core typically gets an RAC of 13 - 14k, whereas the throughput of the best AMD host featuring 2 x 12-cores of RAC ~11k can almost be reached with an Intel 4-core OCed to an estiamted 3.7 GHz (RAC ~10k). So if you want to put it like that: buying an AMD 2 CPU board (340€) and 2 12-cores (2 x 1100€) is more expensive than buying an i7 930 for 250€. So price is the reason we don't see that many top performing AMDs. Oh, and marketing.
Us plebes aren't going to touch the top 10,000 hosts - they're owned by
Universities, laboratories, and technology firms. What CPU's they run
isn't terribly relevant - they're current and fast. That means Intel,
which owns about 85% of the market. For us, it comes down to bang-for-buck,
not Ghz or work-per-cycle. Most lf us don't have $3-$5K to spend on a current
XEON server (minimum). I don't even have money to afford a Core I7 920.
What I can do, is buy a Phenom II 965 BE for $179 and overclock it modestly
(modestly is all most AMD chips can manage) to 3.7-3.8Ghz and it's just a few
percent shy of an Intel Core I7 920 @$300-$320. Better yet, would be a an AMD X6
@$200). I also have an aging 2xOpteron 275 server (4-cores total). circa Q2 2005.
Amazingly, the Opteron server is just about half the performance of a core I7
920, which is circa Q4 2008!
If you have many threads, and under Boinc, you can spin off as many tasks as you
want, one per core (or 2, in the case if the Core I7 and later series), AMD
iron is really cost effective. If you're egalitarian, and must have the fastest
core-for-core machine you can build, undoubtedly, that's Intel.
My Phenom II X4 965 @3.7Ghz earns about 23 credits/cpu-hour, running einstein.
My Opteron machine earns about 11 credits per cpu-hour. For comparison,
benchmarking an 8800GT GPU, it has continuous floating point performance about
11 times greater than a Phenom II X4 965 CPU at 3.7Ghz. Unfortunately,
einstein utilizes only a small fraction of a GPU for FFT, a few percent to about
15% in most cases (and a significant part of that is coprocessor communication).
The future of computation is the GPU (or GPCPU - General Purpose computing Unit).
Unfortunately, neither Einstein nor most of the other BOINC projects really
utilize the GPU effectively.
GPUs are overloading SETI@home with too many results/hour so the admins must shutdown the servers three full days a week. Do we want the same thing to happen at Einstein?
Tullio
Yes but in the top500 list
)
Yes but in the top500 list all first ten but one or two use Opterons, maybe also with CELL, like Roadrunner by IBM.
Tullio
Here is the graph of
)
Here is the graph of percental distribution.
Each column has 500 computers.
blue -> Intel
green -> AMD
red -> other
In the November 2009 top500
)
In the November 2009 top500 list the first three systems are Opteron based, the fourth,seventh and eighth are PowerPc based, the fifth, sixth and tenth are Xeon based and the ninth is another Opteron. But wait for the new list in about a week.
Tullio
RE: In the November 2009
)
Oh, the previous graph is based on Einstein@Home TOP 10000. :-)
RE: RE: In the November
)
I know that the PCs running E@home cannot be compared with "big iron". But still, as I am running a low cost Opteron 1210 with 6 BOINC projects plus a Solaris virtual machine running a BOINC client and SETI app by a developer called Dotsch i still think of the AMD architecture as a good one, even if its marketing cannot compete with Intel's. Having worked for Honeywell-Bull Italy I have seen the power of marketing. I was glad to see a BULL machine working for the ITER fusion experiments. Cheers.
Tullio
Noone said it's a bad
)
Noone said it's a bad architecture. But all the advantages you name are actually just software - you could do that with an Intel just as well.
The problem we're having here with AMD is not its marketing but rather its performance at Einstein@home. A Phenom needs quite a bit more CPU clock to match a Core 2, let alone an i3/5/7. Sandy Bridge is going to increase performance per clock further, whereas Bulldozer is not going to be a FP monster (as needed for E@H).
However, the AMD 6-cores are an interesting offering. In general performance they are a good match to the equivalently priced Intel 4-cores, trading blows in highly threaded apps and loosing in lightly threaded ones and power consumption. However, BOINC can make very good use of the additional 2 cores, so the throughput should be quite competitive.
MrS
Scanning for our furry friends since Jan 2002
RE: Here is the graph of
)
That looks to show that throughout, AMD has about a 18% market share until the top 1000 where it drops to about 10% market share.
That likely fits very closely with their production capacity vs Intel's production capacity! The rest is pricing and Marketing... (As shown by the kink at the top 4000 hosts point?)
Happy crunchin',
Martin
See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)
RE: That likely fits very
)
Yes, 18% should be about the current market share, biased a bit to the high end market due to us selecting only reasonably high performing hosts. But the 10% due to pricing and marketing? Well.. that's a strange way of putting it.
A single Intel 6-core typically gets an RAC of 13 - 14k, whereas the throughput of the best AMD host featuring 2 x 12-cores of RAC ~11k can almost be reached with an Intel 4-core OCed to an estiamted 3.7 GHz (RAC ~10k). So if you want to put it like that: buying an AMD 2 CPU board (340€) and 2 12-cores (2 x 1100€) is more expensive than buying an i7 930 for 250€. So price is the reason we don't see that many top performing AMDs. Oh, and marketing.
MrS
Scanning for our furry friends since Jan 2002
Folks, Us plebes aren't
)
Folks,
Us plebes aren't going to touch the top 10,000 hosts - they're owned by
Universities, laboratories, and technology firms. What CPU's they run
isn't terribly relevant - they're current and fast. That means Intel,
which owns about 85% of the market. For us, it comes down to bang-for-buck,
not Ghz or work-per-cycle. Most lf us don't have $3-$5K to spend on a current
XEON server (minimum). I don't even have money to afford a Core I7 920.
What I can do, is buy a Phenom II 965 BE for $179 and overclock it modestly
(modestly is all most AMD chips can manage) to 3.7-3.8Ghz and it's just a few
percent shy of an Intel Core I7 920 @$300-$320. Better yet, would be a an AMD X6
@$200). I also have an aging 2xOpteron 275 server (4-cores total). circa Q2 2005.
Amazingly, the Opteron server is just about half the performance of a core I7
920, which is circa Q4 2008!
If you have many threads, and under Boinc, you can spin off as many tasks as you
want, one per core (or 2, in the case if the Core I7 and later series), AMD
iron is really cost effective. If you're egalitarian, and must have the fastest
core-for-core machine you can build, undoubtedly, that's Intel.
My Phenom II X4 965 @3.7Ghz earns about 23 credits/cpu-hour, running einstein.
My Opteron machine earns about 11 credits per cpu-hour. For comparison,
benchmarking an 8800GT GPU, it has continuous floating point performance about
11 times greater than a Phenom II X4 965 CPU at 3.7Ghz. Unfortunately,
einstein utilizes only a small fraction of a GPU for FFT, a few percent to about
15% in most cases (and a significant part of that is coprocessor communication).
The future of computation is the GPU (or GPCPU - General Purpose computing Unit).
Unfortunately, neither Einstein nor most of the other BOINC projects really
utilize the GPU effectively.
Stuart
GPUs are overloading
)
GPUs are overloading SETI@home with too many results/hour so the admins must shutdown the servers three full days a week. Do we want the same thing to happen at Einstein?
Tullio