Now you're talking

Message boards : Number crunching : Now you're talking

To post messages, you must log in.

AuthorMessage
bill

Send message
Joined: 16 Jun 99
Posts: 861
Credit: 28,721,888
RAC: 37
United States
Message 1247177 - Posted: 16 Jun 2012, 23:33:15 UTC

I wonder what the RAC would be on this monster

Link
ID: 1247177 · Report as offensive
Profile Fred J. Verster
Volunteer tester
Avatar

Send message
Joined: 21 Apr 04
Posts: 3252
Credit: 31,903,643
RAC: 0
Netherlands
Message 1247218 - Posted: 17 Jun 2012, 1:15:54 UTC - in response to Message 1247177.  
Last modified: 17 Jun 2012, 1:21:01 UTC

I wonder what the RAC would be on this monster

Link


Sure would like to play with such a 'monster'..............
4,096 cores, if they'll compaire to FERMI/KEPLER cores, this is
4096/16= 256. A GTX580(/590?) has 16 cores, so 256 GTX580s

RAC of 5000 (?) 5000 x 256 = 1,280,000 would be possible, theoretical, all
depends on RAC of a single GTX580(590?) Since thats hard to achieve, a CPU
has to load and unload the GPU and write an output file, etc.

How much electricity this machine will use, when fully occupied?
But (far) more efficient compaired to an average CPU-cluster.
ID: 1247218 · Report as offensive
Profile betregerProject Donor
Avatar

Send message
Joined: 29 Jun 99
Posts: 4850
Credit: 11,277,428
RAC: 10,028
United States
Message 1247240 - Posted: 17 Jun 2012, 2:10:55 UTC - in response to Message 1247218.  

I doubt Seti could feed such a machine to keep it full.
ID: 1247240 · Report as offensive
bill

Send message
Joined: 16 Jun 99
Posts: 861
Credit: 28,721,888
RAC: 37
United States
Message 1247279 - Posted: 17 Jun 2012, 5:55:52 UTC - in response to Message 1247218.  

One of Toshibas baby nuclear reactors could handle it.

Then run a couple of wide pipes to the lab.

California dreaming.
ID: 1247279 · Report as offensive
Horacio

Send message
Joined: 14 Jan 00
Posts: 536
Credit: 75,967,266
RAC: 0
Argentina
Message 1247313 - Posted: 17 Jun 2012, 7:29:38 UTC

They are talking of standard CPU Cores (Intel Xeon E5 Sandy Bridge)... See: Big Brain

Anyway, an I7-2600 with HT on (8 cores) gives around 8K RAC... around 1K per core so if the E5 performance were equivalent to the i7 then you can get a RAC of around 4,096,000 (or even more... if Im not wrong the E5 is like the i5 and dosnt use HT which gives more RAC per core)...

It took me 12 years to get 4 millions credits!!! (Im going faster now... but anyway... LOL)

ID: 1247313 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6108
Credit: 155,331,426
RAC: 50,142
United States
Message 1247492 - Posted: 17 Jun 2012, 17:39:57 UTC - in response to Message 1247313.  

They are talking of standard CPU Cores (Intel Xeon E5 Sandy Bridge)... See: Big Brain

Anyway, an I7-2600 with HT on (8 cores) gives around 8K RAC... around 1K per core so if the E5 performance were equivalent to the i7 then you can get a RAC of around 4,096,000 (or even more... if Im not wrong the E5 is like the i5 and dosnt use HT which gives more RAC per core)...

It took me 12 years to get 4 millions credits!!! (Im going faster now... but anyway... LOL)

The Xeon E5 processors come in 2,4,6, and 8 core configurations. Most have HT, but some of the 4 & 6 core versions do not. Depending on the thermal spec they are shooting for I suppose. As the E5 series is designed for 2 socket configurations I would guess they are settings up 2 socket nodes and interconnecting them in some fashion to get the shared memory access.

As they state the base configurations is 16 core configuration I would guess it is two 8 core processors. Unless they are counting cores with HT. Then it is probably just a single processor. It is still impressive that in the end they will have 256 or 512 processors in total.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the BP6/VP6 User Group today!
ID: 1247492 · Report as offensive
Profile Ex
Volunteer tester
Avatar

Send message
Joined: 12 Mar 12
Posts: 2895
Credit: 2,080,451
RAC: 420
United States
Message 1247621 - Posted: 17 Jun 2012, 21:20:47 UTC
Last modified: 17 Jun 2012, 21:23:05 UTC

My Xeon E3-1230 (3.2GHZ) (4 core 8 thread) would do about 11-12k RAC at 100% with op-apps.

So I would think the E5 would do more credit than the 1K per core/thread that was estimated above.
-Dave #2

3.2.0-33
ID: 1247621 · Report as offensive
Horacio

Send message
Joined: 14 Jan 00
Posts: 536
Credit: 75,967,266
RAC: 0
Argentina
Message 1247811 - Posted: 18 Jun 2012, 4:14:45 UTC - in response to Message 1247621.  

My Xeon E3-1230 (3.2GHZ) (4 core 8 thread) would do about 11-12k RAC at 100% with op-apps.

So I would think the E5 would do more credit than the 1K per core/thread that was estimated above.


With that values then a full loaded Big Brain will give around 6 million RAC or more!!

(Anyway, I think that for the some money we can get a higher RAC using several hosts with the fastest/latest GPUs... and, for sure, several hosts will have better luck getting enough work from SETI than a single host that is crunching 4096 concurrent WUs... IIRC, the scheduller has a pool of 200 WUs to assign, so in the best -and impossible- case a host will be getting 200 every 5 mins... That means 1.7 hs to get 4096 tasks... that's way longer than the average crunching time for MB tasks on those CPUs...)



ID: 1247811 · Report as offensive
Profile Slavac
Volunteer tester
Avatar

Send message
Joined: 27 Apr 11
Posts: 1932
Credit: 17,952,639
RAC: 0
United States
Message 1247819 - Posted: 18 Jun 2012, 5:33:36 UTC - in response to Message 1247811.  

Hrm, I want one.


Executive Director GPU Users Group Inc. -
brad@gpuug.org
ID: 1247819 · Report as offensive
Profile Ex
Volunteer tester
Avatar

Send message
Joined: 12 Mar 12
Posts: 2895
Credit: 2,080,451
RAC: 420
United States
Message 1247860 - Posted: 18 Jun 2012, 9:52:36 UTC - in response to Message 1247819.  

Hrm, I want one.


Don't we all. <evil grin>

If someone here starts crunching with a machine like that, a lot of us will end up without work to crunch. ;-)

-Dave #2

3.2.0-33
ID: 1247860 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6108
Credit: 155,331,426
RAC: 50,142
United States
Message 1247910 - Posted: 18 Jun 2012, 13:30:13 UTC - in response to Message 1247860.  

Hrm, I want one.


Don't we all. <evil grin>

If someone here starts crunching with a machine like that, a lot of us will end up without work to crunch. ;-)

I configured a T620 on Dell's configurator with these main specs:
2 - E5-2690 2.90GHz, 20M Cache, 8.0GT/s QPI, Turbo, 8C, 135W
8 - 8GB RDIMM, 1600 MHz
8 - 300GB 15K RPM SAS 6Gbps 2.5in Hot-plug Hard Drive (In RAID 60 configuration)
1 - Dual, Hot-plug, Redundant Power Supply (1+1), 1100W
Which came to just under $20,000. If you reduced the memory and drives, which you wouldn't really need for a computing farm, it comes closer to $12,000.
So for the neighborhood of $3,000,000-$5,000,000 you could have the same amount of computing grunt as your disposal.

The T620 also supports some GPU's which could be used as well. It says up to 4 can be used, but I was only allowed to select 3 with my configuration.
2 - E5-2690 2.90GHz, 20M Cache, 8.0GT/s QPI, Turbo, 8C, 135W
8 - 8GB RDIMM, 1600 MHz
2 - 1TB 7.2K RPM SATA 3.5in Hot-plug Hard Drive (In RAID 1 configuration)
1 - Dual, Hot-plug, Redundant Power Supply (1+1), 1100W
3 - NVIDIA Tesla C2075 GPU computing processor
Comes to just under $23,000.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the BP6/VP6 User Group today!
ID: 1247910 · Report as offensive
EPG

Send message
Joined: 3 Apr 99
Posts: 110
Credit: 10,410,528
RAC: 0
Hungary
Message 1248328 - Posted: 19 Jun 2012, 12:09:31 UTC

And what about Xeon Phi?

Intel is sticking to the goal of offering 1TFLOP of real world double-precision (FP64) performance; for comparison Tesla M2090 and Radeon HD 7970 have a theoretical FP64 throughput of 665GFLOPs and 947GFLOPs respectively.

http://www.anandtech.com/show/6017/intel-announces-xeon-phi-family-of-coprocessors-mic-goes-retail
ID: 1248328 · Report as offensive
Profile Ex
Volunteer tester
Avatar

Send message
Joined: 12 Mar 12
Posts: 2895
Credit: 2,080,451
RAC: 420
United States
Message 1248394 - Posted: 19 Jun 2012, 16:08:08 UTC - in response to Message 1248328.  

And what about Xeon Phi?

Intel is sticking to the goal of offering 1TFLOP of real world double-precision (FP64) performance; for comparison Tesla M2090 and Radeon HD 7970 have a theoretical FP64 throughput of 665GFLOPs and 947GFLOPs respectively.

http://www.anandtech.com/show/6017/intel-announces-xeon-phi-family-of-coprocessors-mic-goes-retail



Oh yes!!!!! HAL!, They packaged it into pci-e cards!!!!!!! :-D That makes my day!
-Dave #2

3.2.0-33
ID: 1248394 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6108
Credit: 155,331,426
RAC: 50,142
United States
Message 1248430 - Posted: 19 Jun 2012, 21:23:34 UTC - in response to Message 1248394.  

And what about Xeon Phi?

Intel is sticking to the goal of offering 1TFLOP of real world double-precision (FP64) performance; for comparison Tesla M2090 and Radeon HD 7970 have a theoretical FP64 throughput of 665GFLOPs and 947GFLOPs respectively.

http://www.anandtech.com/show/6017/intel-announces-xeon-phi-family-of-coprocessors-mic-goes-retail



Oh yes!!!!! HAL!, They packaged it into pci-e cards!!!!!!! :-D That makes my day!

You want to place a bet on if they will be x8 or x16 cards? :)

I would imagine there should be some version of these in an x8 configuration. As most servers are setup for x8 cards instead of x16.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the BP6/VP6 User Group today!
ID: 1248430 · Report as offensive
Profile Ex
Volunteer tester
Avatar

Send message
Joined: 12 Mar 12
Posts: 2895
Credit: 2,080,451
RAC: 420
United States
Message 1248606 - Posted: 20 Jun 2012, 5:20:49 UTC - in response to Message 1248430.  
Last modified: 20 Jun 2012, 5:28:51 UTC

I would imagine there should be some version of these in an x8 configuration. As most servers are setup for x8 cards instead of x16.


That's the very hope I'm living on...

I see way more x8 slots on server boards. It's marketed with the Xeon name, I'm hoping they plan accordingly.



I'd say that's an 8x pictured here... :-D


(I may need to wait for some time to see some prices and probably wait for them to drop, but THIS looks like THE solution to adding parallel processing to my server. It would fit oh-so-nicely, and would work great with my existing hardware. Plus it can function as a system co-processor OR dedicated cruncher :-D, my dreams are coming true here... My only gripe, I could only fit one. (I only have 2 x8's and they are next to each other, and this is a double-width card))
-Dave #2

3.2.0-33
ID: 1248606 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6108
Credit: 155,331,426
RAC: 50,142
United States
Message 1248799 - Posted: 20 Jun 2012, 16:39:39 UTC - in response to Message 1248606.  

I would imagine there should be some version of these in an x8 configuration. As most servers are setup for x8 cards instead of x16.


That's the very hope I'm living on...

I see way more x8 slots on server boards. It's marketed with the Xeon name, I'm hoping they plan accordingly.


[img ]http://images.anandtech.com/doci/6017/XPhiCard_575px.jpg[/img]
I'd say that's an 8x pictured here... :-D


(I may need to wait for some time to see some prices and probably wait for them to drop, but THIS looks like THE solution to adding parallel processing to my server. It would fit oh-so-nicely, and would work great with my existing hardware. Plus it can function as a system co-processor OR dedicated cruncher :-D, my dreams are coming true here... My only gripe, I could only fit one. (I only have 2 x8's and they are next to each other, and this is a double-width card))

That is a x16 card. If you notice the 3 short pins along the connector after the notch. Those are more or less the indicators of where it would end if it were an x1, x4, or x8 connector.
Also I found a large image of one of their sample cards where all 82 pins can be counted.
http://www.version2.dk/sites/v2/files/styles/original/public/wp_000237.jpg

I still bet they will make x8 cards though. Perhaps even x4 versions.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the BP6/VP6 User Group today!
ID: 1248799 · Report as offensive

Message boards : Number crunching : Now you're talking


 
©2016 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.