New AMD board with four PEG x16 slots


log in

Advanced search

Message boards : Number crunching : New AMD board with four PEG x16 slots

Author Message
Christoph
Volunteer tester
Send message
Joined: 21 Apr 03
Posts: 76
Credit: 262,351
RAC: 0
Germany
Message 1073532 - Posted: 2 Feb 2011, 20:11:24 UTC
Last modified: 2 Feb 2011, 20:12:16 UTC

This board TYAN S8225 (S8225AGM4NRF) have four PEG x16 slots(ALL with 16 lanes) and room for four double wide cards. Right now probably the best board for GPU crunching.......if I had the money.......*sigh*
____________
Christoph

baron_iv
Volunteer tester
Avatar
Send message
Joined: 4 Nov 02
Posts: 81
Credit: 13,578,416
RAC: 7,699
United States
Message 1073613 - Posted: 2 Feb 2011, 23:47:33 UTC - in response to Message 1073532.

When AMD's Bulldozer line comes out, we'll likely see a few of those in the top computers list. They're really increasing the performance when it comes to floating point operations. They have something similar to AVX too, although it's a bit different, because they didn't want to get sued by intel, so it's not 100% compatible with AVX. That sucks, because in a month or so, we're gonna have to burden the developers with ANOTHER new, optimized version for a new CPU instruction set. It's a shame that AMD couldn't make it compatible with AVX.

But that board would be epic with 2 12-core CPUs and 4 GTX 580s. That would be a rocketship to the top of the top computers list. I'd love to have one too, but I'd have to win the lottery to buy one. I mean you figure $1200 for the CPUs, probably $800 or the ECC/Buffered RAM, then $2000 for 4 gtx 580s, you'd probably need 2 power supplies to feed that, so that's another $750 or so and ofc $400 for the motherboard. Maaaaan that would be sweet. ;)
____________
-baron_iv
Proud member of:

Profile [seti.international] Dirk SadowskiProject donor
Volunteer tester
Avatar
Send message
Joined: 6 Apr 07
Posts: 7054
Credit: 59,923,492
RAC: 21,810
Germany
Message 1073679 - Posted: 3 Feb 2011, 2:58:07 UTC - in response to Message 1073532.
Last modified: 3 Feb 2011, 3:14:30 UTC

This board TYAN S8225 (S8225AGM4NRF) have four PEG x16 slots(ALL with 16 lanes) and room for four double wide cards. Right now probably the best board for GPU crunching.......if I had the money.......*sigh*


My AMD Phenom II X4 940 BE with 4x GTX260 OC are on a MSI K9A2 Platinum mobo.
4x PCIe 2.0 x16 slots.
If two used @ PCIe 2.0 x16 speed, if all four used (like I) all @ PCIe 2.0 x8 speed.

IIRC, TYAN mobos are not cheap.
Also the upper mentioned is a server board.

MSI should have now a newer mobo also with 4x PCIe 2.0 x16 slots - which would be much cheaper..

Profile [seti.international] Dirk SadowskiProject donor
Volunteer tester
Avatar
Send message
Joined: 6 Apr 07
Posts: 7054
Credit: 59,923,492
RAC: 21,810
Germany
Message 1073681 - Posted: 3 Feb 2011, 3:01:46 UTC - in response to Message 1073679.
Last modified: 3 Feb 2011, 3:03:20 UTC

Huuhh.. I made a quick Google search.. ~ 430,- € .. wow.. too expensive..

Except it must be a server board.
____________
BR



>Das Deutsche Cafe. The German Cafe.<

Saaby900T
Send message
Joined: 24 Dec 10
Posts: 76
Credit: 4,971,171
RAC: 0
United States
Message 1073715 - Posted: 3 Feb 2011, 5:28:52 UTC - in response to Message 1073613.
Last modified: 3 Feb 2011, 5:30:09 UTC

When AMD's Bulldozer line comes out, we'll likely see a few of those in the top computers list. They're really increasing the performance when it comes to floating point operations. They have something similar to AVX too, although it's a bit different, because they didn't want to get sued by intel, so it's not 100% compatible with AVX. That sucks, because in a month or so, we're gonna have to burden the developers with ANOTHER new, optimized version for a new CPU instruction set. It's a shame that AMD couldn't make it compatible with AVX.

But that board would be epic with 2 12-core CPUs and 4 GTX 580s. That would be a rocketship to the top of the top computers list. I'd love to have one too, but I'd have to win the lottery to buy one. I mean you figure $1200 for the CPUs, probably $800 or the ECC/Buffered RAM, then $2000 for 4 gtx 580s, you'd probably need 2 power supplies to feed that, so that's another $750 or so and ofc $400 for the motherboard. Maaaaan that would be sweet. ;)


I'am pretty sure AMD bulldozer has AVX.

Profile Todd Hebert
Volunteer tester
Avatar
Send message
Joined: 16 Jun 00
Posts: 647
Credit: 217,127,962
RAC: 0
United States
Message 1073720 - Posted: 3 Feb 2011, 5:47:31 UTC

Looks like AMD omitted the Socket 1366 platform and its hex-core processor. This is still geared toward the performance desktop segment - not servers. Although it is also used on the server platform.

This particular board is focused at a GPU compute server - easy to tell by its dual sockets. It has been a long time since either the Pentium III or the AMD Athlon MX line that dual cpu's were targeted at the general user. Or at least cheap enough that a average user could afford to populate dual sockets.

Intel does also offer a true octo-core cpu in its X75xx version. Granted this is a server cpu and does have a huge price associated with it. It is not aimed at the desktop in anyway - but let me tell you the performance is fantasic!

Todd
PS - Yes I am an Intel fan boy - and this was not an attempt to hijack a thread. The motherboard is great and would be very nice to have in ones stable of machines.
____________

Saaby900T
Send message
Joined: 24 Dec 10
Posts: 76
Credit: 4,971,171
RAC: 0
United States
Message 1073724 - Posted: 3 Feb 2011, 6:03:27 UTC - in response to Message 1073720.

I Think they left out 1366 socket cause of price. I think AMD is shooting for bulldozer vs. 1155 . Anyways the 980x is only 25% faster than the x6 so I think that bulldozer is still alot of competition for the price.

Saaby900T
Send message
Joined: 24 Dec 10
Posts: 76
Credit: 4,971,171
RAC: 0
United States
Message 1073738 - Posted: 3 Feb 2011, 8:24:40 UTC - in response to Message 1073724.

I Think they left out 1366 socket cause of price. I think AMD is shooting for bulldozer vs. 1155 . Anyways the 980x is only 25% faster than the x6 so I think that bulldozer is still alot of competition for the price.


Also if Sandy bridge is 20% faster than current i7 (almost as good as the 980x) and bulldozer is 20% faster than the 980x and ivy bridge is 20% faster than sandy bridge. Than they should be about equal or very close. Much different AMD might actually stand a chance at the performance crown.

Profile Karsten Vinding
Volunteer tester
Send message
Joined: 18 May 99
Posts: 140
Credit: 16,506,814
RAC: 3,076
Denmark
Message 1073863 - Posted: 3 Feb 2011, 18:46:04 UTC - in response to Message 1073613.

AMD droppped their own implementation (SSE5), a long time ago, and went the AVX way.

Because of the settlement made last year (the case about Intel strongarming computerbuilders into not selling AMD), where Intel payed AMD more than one billion dollars, agreed on a new longer cross-lincensing agreement and other things, AMD can use AVX without fear of lawsuits. The CPU will also support all current SSE levels all the way up to SSE4.2.

AMD have added a little extra (Fused Multiply Add, FMA), but the benefits from this will be dependent upon programs being written to use it. It will probably make its way into OS's and such, but the majority of programs wont use it at first, if ever. The potential for performance increases are great though, so we should hope that it does get used.

Besides all of this, performance of the new chips are still largely unknown, except that they will be much better than current Phenom's.
Its a pretty different design from most of whats out there today, some have even called it revolutionary and bold, but it still has to perform.

At this time only AMD knows what it can do, and they are pretty tight lipped. Bulldozer will be presented at Cebit in less than a month, and then we will probably know a great deal more.
____________

Profile ignorance is no excuse
Avatar
Send message
Joined: 4 Oct 00
Posts: 9529
Credit: 44,433,274
RAC: 0
Korea, North
Message 1073901 - Posted: 3 Feb 2011, 20:44:27 UTC - in response to Message 1073863.

I was going to get a 6 core AMD but I think I'll waiot a few months for the bulldozer
____________
In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope

End terrorism by building a school

baron_iv
Volunteer tester
Avatar
Send message
Joined: 4 Nov 02
Posts: 81
Credit: 13,578,416
RAC: 7,699
United States
Message 1073960 - Posted: 3 Feb 2011, 23:23:03 UTC - in response to Message 1073901.

I was going to get a 6 core AMD but I think I'll waiot a few months for the bulldozer


Probably a very wise move. I think bulldozer is going to be a huge step up from the current Phenom II X6 line. I'm very happy with my 1090T though. It's mighty fast and chews through seti tasks like nobody's business. Plus they're incredibly overclockable. Mine sits at 4.02ghz 24/7 without a voltage increase and runs 48-50C temps, which is amazing. Mind you, it's not as fast as my 980x, but the 980x is 4x the price (probably a silly "investment" for me, but I like fast things). But yeah, when we're this close to a new architecture that could be quite revolutionary for performance, you'd probably be a bit silly to buy a Phenom II. ;)

http://en.wikipedia.org/wiki/XOP_instruction_set#Compatibility_issues According to this, there are some incompatibilities between AVX and Bulldozer. They're using a different instruction set to attempt to get as close as possible to the same features, but I think we will need another new app for bulldozer. I don't know if this has all changed due to the settlement or not. Usually wikipedia is pretty good about updating when something new like that happens.
____________
-baron_iv
Proud member of:

Profile ML1
Volunteer tester
Send message
Joined: 25 Nov 01
Posts: 8369
Credit: 4,102,738
RAC: 1,069
United Kingdom
Message 1073973 - Posted: 4 Feb 2011, 0:07:06 UTC - in response to Message 1073960.

... http://en.wikipedia.org/wiki/XOP_instruction_set#Compatibility_issues According to this, there are some incompatibilities between AVX and Bulldozer. They're using a different instruction set to attempt to get as close as possible to the same features, ...

Not quite... My interpretation is that AMD have all the Intel features listed but also additionally have their own features.

I just wonder if the instruction op-codes compatibility issues are still a sign that there is still some anti-competitive sabotage going on?

Whichever, I just hope it doesn't mean yet more wasted and duplicated effort for the compiler writers and software developers, and also lost optimisations for all users...


I'll agree that AMD's Bulldozer is looking rather interesting...

We very definitely need to keep AMD alive to avoid Intel expensively stagnating with a monopoly...

Happy fast crunchin',
Martin


____________
See new freedom: Mageia4
Linux Voice See & try out your OS Freedom!
The Future is what We make IT (GPLv3)

Christoph
Volunteer tester
Send message
Joined: 21 Apr 03
Posts: 76
Credit: 262,351
RAC: 0
Germany
Message 1074452 - Posted: 5 Feb 2011, 19:44:47 UTC - in response to Message 1073679.

My AMD Phenom II X4 940 BE with 4x GTX260 OC are on a MSI K9A2 Platinum mobo.
4x PCIe 2.0 x16 slots.
If two used @ PCIe 2.0 x16 speed, if all four used (like I) all @ PCIe 2.0 x8 speed.


Well the difference is that this new board has a total of 64 lanes and must not share them between the slots. There was a thread about how the transfer speed to the cards effect compute times.

Oh, I just saw that it was one of your threads wich I mean. http://setiathome.berkeley.edu/forum_thread.php?id=62704#1064427

So, this board should have faster processing speed whith four cards compared to boards with only 32 available lanes.
____________
Christoph

Profile Todd Hebert
Volunteer tester
Avatar
Send message
Joined: 16 Jun 00
Posts: 647
Credit: 217,127,962
RAC: 0
United States
Message 1074524 - Posted: 5 Feb 2011, 22:59:32 UTC - in response to Message 1074452.

Only if the GPU or add-in card can truely make use of them. Thus far the only devices that I have seen that can max out a x16 PCI-E 2.0 slot is a RAID card with a ton of Cache on it and also a 10GB ethernet card with multi-ports.
Todd
____________

Christoph
Volunteer tester
Send message
Joined: 21 Apr 03
Posts: 76
Credit: 262,351
RAC: 0
Germany
Message 1074745 - Posted: 6 Feb 2011, 15:53:37 UTC - in response to Message 1074524.

Ok, I thought GPUs are the things which realy use and need all these lanes. In this case of course this board offers no advantage in terms of GPU crunching.
____________
Christoph

Profile Todd Hebert
Volunteer tester
Avatar
Send message
Joined: 16 Jun 00
Posts: 647
Credit: 217,127,962
RAC: 0
United States
Message 1074780 - Posted: 6 Feb 2011, 17:07:08 UTC

This might be true in the near future with newer cards needing more bandwidth. Also, Tesla cards do have much more memory onboard - up to 6GB! And this might be useful in the arena - but not many of us are using Tesla cards on this project based on their significant cost. http://www.amazon.com/gp/offer-listing/B0044O7FXS/ref=asc_df_B0044O7FXS1419586?ie=UTF8&condition=new&tag=dealtmp95-20&creative=395165&creativeASIN=B0044O7FXS&linkCode=asm

These are suited for a much more professional purpose - to say the least!
Todd
____________

Christoph
Volunteer tester
Send message
Joined: 21 Apr 03
Posts: 76
Credit: 262,351
RAC: 0
Germany
Message 1075179 - Posted: 7 Feb 2011, 21:35:29 UTC - in response to Message 1074780.

Well, these are far out of range.......with my old job I could have build up a machine using that MoBo equiped with GPUs within one year, but even than that type of card would have been out of range.......no sence to spend so much money compared with consumer GPGPU products.
____________
Christoph

-BeNt-
Avatar
Send message
Joined: 17 Oct 99
Posts: 1234
Credit: 10,116,112
RAC: 0
United States
Message 1075377 - Posted: 8 Feb 2011, 15:15:48 UTC

Whatever happened to the one guy here that was building the home "super computer" with the tesla cards? He just kind of disappeared. Seems I remember from the conversation then that people were expecting that generation of tesla, don't remember or know much about tesla(think it was a 1 series maybe a 2?), was about the same speed as a 295. Some of the reviews have them at about 500 GFLOPS.....yeah $3k dollars.....think I'll stick with consumer level crunchers lol.

Wonder how well they scale crunching wise though? Something has to make them worth the money right???
____________
Traveling through space at ~67,000mph!

Saaby900T
Send message
Joined: 24 Dec 10
Posts: 76
Credit: 4,971,171
RAC: 0
United States
Message 1075432 - Posted: 9 Feb 2011, 5:19:20 UTC - in response to Message 1075377.

Whatever happened to the one guy here that was building the home "super computer" with the tesla cards? He just kind of disappeared. Seems I remember from the conversation then that people were expecting that generation of tesla, don't remember or know much about tesla(think it was a 1 series maybe a 2?), was about the same speed as a 295. Some of the reviews have them at about 500 GFLOPS.....yeah $3k dollars.....think I'll stick with consumer level crunchers lol.

Wonder how well they scale crunching wise though? Something has to make them worth the money right???


They are more accurate that is about it. (it really matter when using them for CAD) .00000001 inch to you or me is nothing. To a 70 story building however that could cause problems.

-BeNt-
Avatar
Send message
Joined: 17 Oct 99
Posts: 1234
Credit: 10,116,112
RAC: 0
United States
Message 1076038 - Posted: 11 Feb 2011, 5:56:14 UTC
Last modified: 11 Feb 2011, 5:58:12 UTC

Did a little digging on them and found that they advertise them(c2070 Tesla) to deliver the same performance of cpu-based cluster at one-tenth the cost and one-twentieth the power. However they don't specify in accordance to what hardware the cluster is, so I suppose the market for these would be people building a cluster cruncher for research that needed to save on power. But the the question begs, why would you still do it instead of a regular gpu?

A bit further down the article they started to go into depth about how the memory on board was ecc, multi-level cache, support c++ programming(I think that's what we have now right?), and up to 1 terabyte of memory, concurrent kernel execution, 10x faster atomic instructions, and recursive functions. Basically a lot of stuff that Seti@Home doesn't use to be as precise, as Saaby said, to be as accurate over a large data based crunch. Unlike out 'little' work units.

From reading the maximum pc article about the Nvidia 'supercomputer' they also act in parallel with the cpu to crunch data, but not sure if that is a defined line because we do the same thing in all reality, just on separate work units. However the article is from 2008 talking about it having 960 cores, so I assume the 1050 series, which is almost what as big as 2x 580's now. So it seems they are possibly a bit ahead. Still not worth the money for us though. Like to see some of these in action in the environment they are meant for, would be interesting for sure.

One interesting note, the largest tesla supercomputer cluster I could find publicized was the Nebulae supercomputer composed of 4,650 Tesla C2050(448 cores per) cards along with Xeon x5620 cpus. Thats ~500 Gflops per card, not counting the cpus?!(120,640 complete) That's one expensive piece of equipment. All of that is at a TDP of 225W per server. The latest numbers published on it I can find put it at top performance was 2.98 PFLOPS. The closet supercomputer to it is Jaguar which consumes 7 MegaWatts of power(@ 2.33 PFLOPS) which is 2x what the Nebula supercomputer uses. Amazing, in that field there is definitely a need for them, when you look at the power consumption and probably heat, versus the desktop comparatives.

So what do you reckon' the RAC on that thing would be haha?
____________
Traveling through space at ~67,000mph!

Message boards : Number crunching : New AMD board with four PEG x16 slots

Copyright © 2014 University of California