This technology offers about 10-100 times preformance improvements.
1 x Opteron -> 3.2 - 4.5 GFlops
+ 1 x DRC -> 32 - 450 GFlops
Neat. What's the price-tag? The mid-sized Virtex-4 ain't not that expensive; around ~500-900 USD each.
What i find more interesting are the Cell CPUs; they're quite interesting as well and Mecury is offering Cell-based PCIe card with 179 single-precision GLFOPS. And according to their press-release their FFT-routines are supposed to be quite fast.
Rant: http://www.bloglines.com/blog/ITnomad?id=130
Cell accel.-board: http://www.mc.com/cell/products/view/index.cfm?id=106&type=boards
I guess that the card will be around ~3000€, so we'd end up with 16€/GFLOP. That's not too bad compared to an Opteron, although i doubt that it'll be a wise investment for a hobbyist considering that CPUs will get faster all the time - or, more modern said: There will be more cores on a die in the future.
But since Cell will be in the Playstation 3... go figure :) Break some DMCA and mod it up.
And regarding the Clear Speed... Dunno - a friend of mine who's using them at work banged in a small benchmark [3] which i used to toy around on an old Cray Y-MP EL and basically: The CX sucked big time when doing vector-operations.
The interesting thing about the CX is that you don't have to screw around with VHDL or other strange stuff, but you just pump your binaries into that board and voila! - there we go, cowboy. And it has it's own RAM.
...And the Cray still kicked my Pentium's[1] arse, which was quite surprising to me :)
Still good at vectors.
"I am tired of all this sort of thing called science here... We have spent
millions in that sort of thing for the last few years, and it is time it
should be stopped."
-- Simon Cameron, U.S. Senator, on the Smithsonian Institute, 1901.
I guess that the card will be around ~3000 EUR, so we'd end up with 16 EUR/GFLOP.
Heise just released some more information about the Cell-board; the price is 8000 USD and will be released by spring next year.
Cheers, Alex.
"I am tired of all this sort of thing called science here... We have spent
millions in that sort of thing for the last few years, and it is time it
should be stopped."
-- Simon Cameron, U.S. Senator, on the Smithsonian Institute, 1901.
Alright, more details: The 179 GFLOPS are just single-precision. Downgraded to double-precission the board would've about the power of a quad-Opteron 244, but more energy-efficient.
We discussed the whole thing in the Rosetta-Fori, see: http://boinc.bakerlab.org/rosetta/forum_thread.php?id=2058#21741
Good night,
Alex.
"I am tired of all this sort of thing called science here... We have spent
millions in that sort of thing for the last few years, and it is time it
should be stopped."
-- Simon Cameron, U.S. Senator, on the Smithsonian Institute, 1901.
Oh my !
8,000 $ for 1 coprocessor
+ 1 dual-processor mainboard
+ 1 power supply able to sustain the whole system
+ memory
-------------------------
about 10,000 $
for that price, you can buy 10 cheap systems or more and that farm will crunch more floats than that high-priced beast.
I am crazy, I may invest 1,200 $ per year to upgrade my computers, nevertheless I may not spend 10,000 $ a year.
That'd be MAD :-)
Quote:
Alright, more details: The 179 GFLOPS are just single-precision. Downgraded to double-precission the board would've about the power of a quad-Opteron 244, but more energy-efficient.
We discussed the whole thing in the Rosetta-Fori, see: http://boinc.bakerlab.org/rosetta/forum_thread.php?id=2058#21741
Good night,
Alex.
"Entia non sunt multiplicandam praeter necessitatem" (OKHAM)
I found this developers board http://www.hitechglobal.com/Boards/V4LXPCIE.htm
for less than $5K. One of you sharp coders could make this work, still more then I would like to spend. I like the little DRC unit, perhaps one Opteron 290 and a DRC unit in the other 940 socket, wonder if that would foul up my access to the second pcie x16 slot?
Very interesting, is there a way to figure out what the gflops performance of this card would be? That pcie 1x path is good for over 2 gb/s. It also has an expansion board, hmmmm. And the price is very nice.
I just happen to have an 8x pcie slot (at 4x) available.
RE: Coprocessor for
)
Neat. What's the price-tag? The mid-sized Virtex-4 ain't not that expensive; around ~500-900 USD each.
What i find more interesting are the Cell CPUs; they're quite interesting as well and Mecury is offering Cell-based PCIe card with 179 single-precision GLFOPS. And according to their press-release their FFT-routines are supposed to be quite fast.
Rant: http://www.bloglines.com/blog/ITnomad?id=130
Cell accel.-board: http://www.mc.com/cell/products/view/index.cfm?id=106&type=boards
I guess that the card will be around ~3000€, so we'd end up with 16€/GFLOP. That's not too bad compared to an Opteron, although i doubt that it'll be a wise investment for a hobbyist considering that CPUs will get faster all the time - or, more modern said: There will be more cores on a die in the future.
But since Cell will be in the Playstation 3... go figure :) Break some DMCA and mod it up.
And regarding the Clear Speed... Dunno - a friend of mine who's using them at work banged in a small benchmark [3] which i used to toy around on an old Cray Y-MP EL and basically: The CX sucked big time when doing vector-operations.
The interesting thing about the CX is that you don't have to screw around with VHDL or other strange stuff, but you just pump your binaries into that board and voila! - there we go, cowboy. And it has it's own RAM.
...And the Cray still kicked my Pentium's[1] arse, which was quite surprising to me :)
Still good at vectors.
Alex.
[1] P4 Northwood 2.4GHz, 800MHz FSB, 1GB DDR not the slowest machine, but SSE is nothing compared to a good vector-unit of that old Cray
[2] http://www.cray-cyber.org/systems/yel.php
[3] https://opz.ynfonatic.de/61
Edit: Corrected URL [3]
"I am tired of all this sort of thing called science here... We have spent
millions in that sort of thing for the last few years, and it is time it
should be stopped."
-- Simon Cameron, U.S. Senator, on the Smithsonian Institute, 1901.
RE: I guess that the card
)
Heise just released some more information about the Cell-board; the price is 8000 USD and will be released by spring next year.
Cheers, Alex.
"I am tired of all this sort of thing called science here... We have spent
millions in that sort of thing for the last few years, and it is time it
should be stopped."
-- Simon Cameron, U.S. Senator, on the Smithsonian Institute, 1901.
Alright, more details: The
)
Alright, more details: The 179 GFLOPS are just single-precision. Downgraded to double-precission the board would've about the power of a quad-Opteron 244, but more energy-efficient.
We discussed the whole thing in the Rosetta-Fori, see: http://boinc.bakerlab.org/rosetta/forum_thread.php?id=2058#21741
Good night,
Alex.
"I am tired of all this sort of thing called science here... We have spent
millions in that sort of thing for the last few years, and it is time it
should be stopped."
-- Simon Cameron, U.S. Senator, on the Smithsonian Institute, 1901.
Oh my ! 8,000 $ for 1
)
Oh my !
8,000 $ for 1 coprocessor
+ 1 dual-processor mainboard
+ 1 power supply able to sustain the whole system
+ memory
-------------------------
about 10,000 $
for that price, you can buy 10 cheap systems or more and that farm will crunch more floats than that high-priced beast.
I am crazy, I may invest 1,200 $ per year to upgrade my computers, nevertheless I may not spend 10,000 $ a year.
That'd be MAD :-)
"Entia non sunt multiplicandam praeter necessitatem"
(OKHAM)
RE: I am crazy, I may
)
Crap, that makes it official... I'm MAD
I found this developers board
)
I found this developers board http://www.hitechglobal.com/Boards/V4LXPCIE.htm
for less than $5K. One of you sharp coders could make this work, still more then I would like to spend. I like the little DRC unit, perhaps one Opteron 290 and a DRC unit in the other 940 socket, wonder if that would foul up my access to the second pcie x16 slot?
RE: I found this developers
)
Cheaper variations with less performance:
Spartan-3 PCI Express Starter Kit
Spartan-II PCI Development Kit
RE: RE: I found this
)
Very interesting, is there a way to figure out what the gflops performance of this card would be? That pcie 1x path is good for over 2 gb/s. It also has an expansion board, hmmmm. And the price is very nice.
I just happen to have an 8x pcie slot (at 4x) available.
RE: RE: I found this
)
What about einstein's source code compilation ? I read there are some C/C++ GNU compiler available. But the sources aren't available.
"Entia non sunt multiplicandam praeter necessitatem"
(OKHAM)
The extreem card
)
The extreem card http://www.mc.com/products/view/index.cfm?id=106&type=boards , 180 GFLOPS for only $7999.