Welcome to our website

To take full advantage of all features you need to login or register. Registration is completely free and takes only a few seconds.

8-way, 16-core Iwill system Benchmarked

Posted by: duke on: 05/27/2005 01:44 PM [ Print | 18 comment(s) ]

Extended64 has posted some preliminary benchmarks of the 8-way, 16-core Opteron system from Iwill.

Since The Inquirer wrote an article yesturday about IWill demoing this server, I decided to let the world see the benchmarks that I took last month on the testing system. Below are screenshots with some quick benchmarks, and the numbers speak for themselves. As expected, this box basically beats every P4/Xeon/Opteron/Athlon64 in ANY configuration.

You can look at all of the benchmarks over here.

Comment

duraid SMP Qualified Posts: 387 Joined: 2002-03-31

#36609 Posted on: 05/27/2005 06:17 PM or you could visit http://www.spec.org , it won't bite you :\

Comment

MultimediaMan S3n10R D3s!gN 3ng!n33r Posts: 1490 Joined: 2000-05-12

#36610 Posted on: 05/28/2005 03:00 AM The ultimate Freecell machine! Thus far, you have been adrift within the sheltered harbor of my patience. Heat All of my heatware comes from Transactions on this board. 2CPU.com is the only board I trade on. I can provide references from other board members on request.

Comment

XWRed1 Registered User Posts: 185 Joined: 2001-08-27

#36611 Posted on: 05/28/2005 12:22 PM Is this supposed to be more interesting than the SGI 64-way IA64 box w/ 256gb of ram I saw at Linuxworld 2 years ago? I guess with the combinations of Windows and 128gb of memory... I could run an insane amount of spyware at once!

Comment

duraid SMP Qualified Posts: 387 Joined: 2002-03-31

#36612 Posted on: 05/28/2005 05:33 PM Originally posted by XWRed1 Is this supposed to be more interesting than the SGI 64-way IA64 box w/ 256gb of ram I saw at Linuxworld 2 years ago? yeah but how are you gonna run sandra to know if its good or not???? :confused: and i heard you cant OC those SGI systems??? whats up with that!!! :mad: ;)

Comment

scythe Quad nutty Posts: 555 Joined: 2002-09-25

#36613 Posted on: 05/28/2005 06:46 PM Originally posted by XWRed1 I guess with the combinations of Windows and 128gb of memory... I could run an insane amount of spyware at once! The realy beauty is that if a person purchases a machine configured as such they could run all the spyware that they apparently want and not notice any slow downs. - Q6600 @ 3.0ghz (333x9) - 4x1gb DDR2 800 - HD4890 - Asus Maximus Formula X38 - 150GB RaptorX - Antec P180v2 w/ 550w Real Power Pro -

Comment

terminalrecluse is home now Posts: 3802 Joined: 2004-08-07

#36614 Posted on: 05/29/2005 12:47 AM Sandra benchmarks? Are you serious? That's not benchmarking. Where are the number crunching ones? How much of a folding monster is it? How fast will it compile a linux / freebsd kernel? Oh, and the obligatory will it run Doom 3? 121 total Ghz, 304GB in total memory... Arch Linux - stable : 3930k @ 4.4Ghz, 64GB DDR3, 120GB Samsung 840, HX850 DAS - Norco 4020 - 20x Seagate 3TB, 1KW PSU FBSD 10 ZFS server - SM Chassis, SM X8DTE, 2x L5520 Xeons, 48GB Reg DDR3 ULP, IBM 5015 w/ 512MB DDR2 cache, RAID-50 - 42TB storage Dell C6100 - 4 nodes, 2 1.1KW psu's, 2x L5520s, 6x e5530s, 192GB reg DDR3 (48GB each), F@H, etc

Comment

proffesso Winner of the Internet Posts: 1532 Joined: 2001-06-25

#36615 Posted on: 05/29/2005 04:42 AM 8 cpu's / 16 way probably very expensive, but much, much cheaper than the sgi alternative. its kind of that 'if I could justify it (ie, claim it on tax) then yeah...its possible. SGI on the otherhand is hugely priced gear. chasing tokyo girls

Comment

rmn oh my, it's huge! Posts: 6013 Joined: 2002-01-26

#36616 Posted on: 05/29/2005 05:22 AM This kind of system only makes sense for applications that need lots of processing power and lots of memory bandwidth. For things like rendering it's much more cost-effective to set up a network with several dual-CPU / dual-core nodes. RMN ~~~

Comment

proffesso Winner of the Internet Posts: 1532 Joined: 2001-06-25

#36617 Posted on: 05/29/2005 06:02 AM depends. for setting up shaders / lighting etc, then a huge 16 way system could be very effective for that. becuase even though mentalray can be used for bucket rendering accross the network for single frames (ie, the renderregion when setting up shaders / lights / render settings) the network just becomes a bottleneck. its known that 4 machines, even on GbE is the usually the max for plugging into rayhosts. purely rendering sequential frames however, your absolutly correct. chasing tokyo girls

Comment

XWRed1 Registered User Posts: 185 Joined: 2001-08-27

#36618 Posted on: 05/29/2005 07:19 AM probably very expensive, but much, much cheaper than the sgi alternative. So the impressive thing is supposed to be the price, and not "OMGZ MICROSOFT WINDOWS SERVER 64-BIT EDITION FOR 64-BIT EXTENDED SYSTEMS ENTERPRISE EDITION VERSION 2003 AND 128GB RAM" ?

Comment

duraid SMP Qualified Posts: 387 Joined: 2002-03-31

#36619 Posted on: 05/29/2005 10:52 AM Originally posted by proffesso 8 cpu's / 16 way probably very expensive, but much, much cheaper than the sgi alternative. Actually, if you include the cost of the RAM (or do you claim that on tax? ) I think you'll find that the total prices will be within 20% or 30% of each other, and the SGI may actually end up being cheaper. (as i'm about to explain to rmn...)

Comment

duraid SMP Qualified Posts: 387 Joined: 2002-03-31

#36620 Posted on: 05/29/2005 10:59 AM Originally posted by rmn This kind of system only makes sense for applications that need lots of processing power and lots of memory bandwidth. For things like rendering it's much more cost-effective to set up a network with several dual-CPU / dual-core nodes. You're assuming an Altix equipped with several numalink 3 or 4 routers, which to be fair is how most large altix systems are sold. What you might not know (it's not _that_ obvious from the website) is that you can get away with fewer routers, or even none at all (and so you end up with a normal ethernet cluster of dual socket altix nodes). As you can imagine, the price goes down accordingly. Depending on your application, you might find the altix to be better price-performance than many other systems out there, incuding dual-core opteron systems. Not all of these applications are floating-point, either. I recently came across an artificial intelligence program (automatic theorem prover anyway) that simply flies on itanium, a good 4x faster per clock than anything else, including xeon, pentium M, K8 and PPC970. - http://spass.mpi-sb.mpg.de/ (don't believe me? try it yourself!)

Comment

elec999 Registered User Posts: 1268 Joined: 2003-01-31

#36621 Posted on: 05/29/2005 02:00 PM Does each cpu have it own memory, or is it shared. Thanks [BOTH BOINC][SETI@HOME][2CPU TEAM][http://www.boincstats.com/signature/user_126915_project0.gif

Comment

duraid SMP Qualified Posts: 387 Joined: 2002-03-31

#36622 Posted on: 05/29/2005 04:17 PM Originally posted by elec999 Does each cpu have it own memory, or is it shared. Yes.

Comment

i_wolf labhair dom as gaelige Posts: 2097 Joined: 2002-11-19

#36623 Posted on: 05/29/2005 04:27 PM First Order Predicate Calculus prover ? I'm assuming then it uses some type of modus ponus ( If A => B and A is true we can infer B. ) and the usual term rewriting rules (DeMorgans theorem etc...) to arrive at its conclusions. But then there will be things he wants to prove are true by proving the the opposite could not be true (refutation) so he would have to use resolution. Problem with this is converting what the user enters to clause form.... I would imagine a lot of the Itanium horse power would go into this.... especially the part where we calculate the conjunts of disjunts. To actually code something like this would require a heck of a lot of conditional branches... I'd imagine that accounts for a big reason why the Itanium fares so well on it. That thing is a monster for handling these types of scenarios. Hung like a donkey. Go like a horse!

Comment

terminalrecluse is home now Posts: 3802 Joined: 2004-08-07

#36624 Posted on: 05/29/2005 09:16 PM AI, eh? I don't care if she is powered by a couple of Itaniums, just that she is a hotty! All joking aside that prover program is pretty cool. 121 total Ghz, 304GB in total memory... Arch Linux - stable : 3930k @ 4.4Ghz, 64GB DDR3, 120GB Samsung 840, HX850 DAS - Norco 4020 - 20x Seagate 3TB, 1KW PSU FBSD 10 ZFS server - SM Chassis, SM X8DTE, 2x L5520 Xeons, 48GB Reg DDR3 ULP, IBM 5015 w/ 512MB DDR2 cache, RAID-50 - 42TB storage Dell C6100 - 4 nodes, 2 1.1KW psu's, 2x L5520s, 6x e5530s, 192GB reg DDR3 (48GB each), F@H, etc

Comment

elec999 Registered User Posts: 1268 Joined: 2003-01-31

#36625 Posted on: 06/04/2005 08:58 AM Imagine how long it will take to get 16 f@h or s@h running. Its easy to turn them when using a script. But to make 16 folders for each client. It will take sometime . Thanks [BOTH BOINC][SETI@HOME][2CPU TEAM][http://www.boincstats.com/signature/user_126915_project0.gif

Comment

proffesso Winner of the Internet Posts: 1532 Joined: 2001-06-25

#36626 Posted on: 06/04/2005 05:06 PM Originally posted by i_wolf First Order Predicate Calculus prover ? I'm assuming then it uses some type of modus ponus ( If A => B and A is true we can infer B. ) and the usual term rewriting rules (DeMorgans theorem etc...) to arrive at its conclusions. But then there will be things he wants to prove are true by proving the the opposite could not be true (refutation) so he would have to use resolution. Problem with this is converting what the user enters to clause form.... I would imagine a lot of the Itanium horse power would go into this.... especially the part where we calculate the conjunts of disjunts. To actually code something like this would require a heck of a lot of conditional branches... I'd imagine that accounts for a big reason why the Itanium fares so well on it. That thing is a monster for handling these types of scenarios. you make my brain hurt. chasing tokyo girls