Best Practices For Building A Multiple GPU System

Discuss anything you like on this forum.
Post Reply
User avatar
Notiusweb
Licensed Customer
Posts: 1285
Joined: Mon Nov 10, 2014 4:51 am

smicha wrote:Guys,

Do you know anything about pricing of the Klimax 210?
I don't know, but the name alone is priceless! :D
Win 10 Pro 64, Xeon E5-2687W v2 (8x 3.40GHz), G.Skill 64 GB DDR3-2400, ASRock X79 Extreme 11
Mobo: 1 Titan RTX, 1 Titan Xp
External: 6 Titan X Pascal, 2 GTX Titan X
Plugs: Enterprise
User avatar
smicha
Licensed Customer
Posts: 3151
Joined: Wed Sep 21, 2011 4:13 pm
Location: Warsaw, Poland

Notiusweb wrote:
smicha wrote:Guys,

Do you know anything about pricing of the Klimax 210?
I don't know, but the name alone is priceless! :D

I am trying to get in touch with them. If I get any news I'll post here.
3090, Titan, Quadro, Xeon Scalable Supermicro, 768GB RAM; Sketchup Pro, Classical Architecture.
Custom alloy powder coated laser cut cases, Autodesk metal-sheet 3D modelling.
build-log http://render.otoy.com/forum/viewtopic.php?f=9&t=42540
User avatar
rappet
Licensed Customer
Posts: 1962
Joined: Fri Apr 06, 2012 3:57 pm
Location: The Netherlands
Contact:

smicha wrote:Guys,
Do you know anything about pricing of the Klimax 210?
my few cents here:
Looks to me it a a very narrow spaced box with narrow spaced aircooled GPU, which probably will not get the best performance per gpu.
Probably best solution for big big datacentre wher you have so many cases like this stacked.. then it might be good performance per square area used in that space.,
and not having to look at price-performance.
I think two regular PC cases with good airflow with 8 gpu might have better performance, for even cheaper price, then this one box with 11 gpu badly aircooled.
maybe I am mistaken, but that is my first gut feeling :roll:

cheers,

4090+3089ti & Quad 1080ti
ArchiCAD25, ofcourse Octane & OR-ArchiCAD plugin (love it)
http://www.tapperworks.com
http://www.facebook.com/pages/TAPPERWOR ... 9851341126
http://www.youtube.com/user/Tapperworks/videos
User avatar
smicha
Licensed Customer
Posts: 3151
Joined: Wed Sep 21, 2011 4:13 pm
Location: Warsaw, Poland

rappet wrote:
smicha wrote:Guys,
Do you know anything about pricing of the Klimax 210?
my few cents here:
Looks to me it a a very narrow spaced box with narrow spaced aircooled GPU, which probably will not get the best performance per gpu.
Probably best solution for big big datacentre wher you have so many cases like this stacked.. then it might be good performance per square area used in that space.,
and not having to look at price-performance.
I think two regular PC cases with good airflow with 8 gpu might have better performance, for even cheaper price, then this one box with 11 gpu badly aircooled.
maybe I am mistaken, but that is my first gut feeling :roll:

cheers,
rappet,

it has 20 pcie slots 16x so I am thinking about cutting 20 gpus into single slotted and watercool'em

PS. do you think I owe you a good beer? :) Monday you say is the day?
3090, Titan, Quadro, Xeon Scalable Supermicro, 768GB RAM; Sketchup Pro, Classical Architecture.
Custom alloy powder coated laser cut cases, Autodesk metal-sheet 3D modelling.
build-log http://render.otoy.com/forum/viewtopic.php?f=9&t=42540
User avatar
rappet
Licensed Customer
Posts: 1962
Joined: Fri Apr 06, 2012 3:57 pm
Location: The Netherlands
Contact:

smicha wrote::
rappet,

it has 20 pcie slots 16x so I am thinking about cutting 20 gpus into single slotted and watercool'em

PS. do you think I owe you a good beer? :) Monday you say is the day?
Yep, Ofcorsere you are putting them on water ;)
Then it is another story ofcourse... man 20 singleslotted GPU... that would be something

AS for inspiration: Linus is watercooling a server here (cpu though):


About the beer, yes, I think you do.. ;).... it is 28th feb or 1st March
I would be surpsrised if the green team pull another rabbit out of the head after this countdown 'announcement' with a giveaway in the source code of the webpage and the Ti in bold written hint :roll:

we'll see.. then cheers with digital beers :P

4090+3089ti & Quad 1080ti
ArchiCAD25, ofcourse Octane & OR-ArchiCAD plugin (love it)
http://www.tapperworks.com
http://www.facebook.com/pages/TAPPERWOR ... 9851341126
http://www.youtube.com/user/Tapperworks/videos
User avatar
Tutor
Licensed Customer
Posts: 531
Joined: Tue Nov 20, 2012 2:57 pm
Location: Suburb of Birmingham, AL - Home of the Birmingham Civil Rights Institute

My 22 GPU 3 Level Condo
A tribute to the bitcoiners who helped to motivate me to think outside of the box.
philliplakis wrote:Hey guys,

Sorry to chime in late. Does the X10DRX have open ended PCIe slots? So we could effectively insert 11 - 16x slot cards?
The Supermicro X9DRX and the X10DRX each have 11 PCI-e slots. Those slots are all x8 sized and are closed on the front so that populating them with GPUs with an x16 sized PCI-e male connectors requires cutting the front of the slots or using the low profile x8 to x16 riser cards that I've reference earlier in this thread [ viewtopic.php?f=40&t=43597&start=190#p238446 ]. On the X10DRX, the first eight of the eleven slots (on the left-hand side) aren't blocked by the memory slots; on the X9DRX it's seven of the eleven on the left-hand side that aren't blocked by the memory slots.

Condo level 1 - I generally place the display GPU in the PCI-e slot farthest to the left, using/placing it upon a low profile x8 to x16 riser card. On my latest build ( mainly GTX 1070s air only), I placed the display GTX 1070 GPU into slot no. 1 and have placed the second and third 1070 GPUs into slots nos. 4 and 7 respectively. All three are all well-spaced apart and they sit atop low profile x8 to x16 riser cards. I use #4-40 x1 machine screws to bind these the video cards' back brackets to the card connectors on the Lian LI D8000's frame where the regular (but now too short) screws would have otherwise been used to secure the video cards.

Condo level 2 - Since the second and third Level 1 1070 GPUs, like the first Level 1 GPU, are atop those riser cards, I still have access to the PCI-e slots that those GPUs would otherwise be blocking/covering (that's why I prefer the risers over a Dremel/Exacto Knife cutting session - I like turning seeming obstacles into my further advantage). In those PCIe slots (as well as a third PCIe slot) I have placed x1 to x16 riser cables ( but I may soon order, for replacement - x8 to x16 riser cables for them) and connected them to 3 more 1070s. If you've been counting, I know that you see that at this point in the description that I've used six of my eleven PCIe slot allotment, for six level 1 and level 2 1070s. In four of the remaining five PCIe slots (I'm leaving the eleven slot free for future expansion), I've placed four x4 splitters that I've distributed as follows: two of the splitters to the remaining PCI-e slots fed by CPU1 (the front one) and the other two splitters are fed by CPU2 (the back CPU). All three of these GPUs are all well-spaced apart and are staggered so as not to sit directly above the GPUs on Level 1.

Now some math is in order:

20 - 6 = 14.

Condo level 3 - Fourteen 1070s are feed by splitters nos. 1, 2, 3 and 1/2 of 4. The other half of splitter no. 4 feeds 2 x GTX 980 Tis. All sixteen of these level 3 GPUs are all well-spaced apart and the ones inside the complex are staggered so as not to sit directly above the GPUs on Level 2.

P.S. It's still a WIP. Moreover, the Condo has to have the aka of the "Linux Tower" because currently available Windows OSes will support, at best, no more than 14 GPU processors.
You do not have the required permissions to view the files attached to this post.
Because I have 180+ GPU processers in 16 tweaked/multiOS systems - Character limit prevents detailed stats.
User avatar
smicha
Licensed Customer
Posts: 3151
Joined: Wed Sep 21, 2011 4:13 pm
Location: Warsaw, Poland

Tutor,

Could you please measure ms/s speed the way I did here
https://youtu.be/ABfxuxKcEWc?list=PLj4j ... 6BTrFetgz0

and post ms/s results but only on 1070s (14x of them) and core and memory mhz speed?

PS. The setup is very impressive.
3090, Titan, Quadro, Xeon Scalable Supermicro, 768GB RAM; Sketchup Pro, Classical Architecture.
Custom alloy powder coated laser cut cases, Autodesk metal-sheet 3D modelling.
build-log http://render.otoy.com/forum/viewtopic.php?f=9&t=42540
User avatar
Tutor
Licensed Customer
Posts: 531
Joined: Tue Nov 20, 2012 2:57 pm
Location: Suburb of Birmingham, AL - Home of the Birmingham Civil Rights Institute

smicha wrote:Tutor,

Could you please measure ms/s speed the way I did here
https://youtu.be/ABfxuxKcEWc?list=PLj4j ... 6BTrFetgz0

and post ms/s results but only on 1070s (14x of them) and core and memory mhz speed?

PS. The setup is very impressive.
Sure. In the next few days in addition to posting core and memory mhz speeds, I'll post ms/s for four, eight, twelve, fourteen, sixteen and twenty 1070s if possible.
Because I have 180+ GPU processers in 16 tweaked/multiOS systems - Character limit prevents detailed stats.
User avatar
rappet
Licensed Customer
Posts: 1962
Joined: Fri Apr 06, 2012 3:57 pm
Location: The Netherlands
Contact:

Tutor wrote:....I'll post ms/s for four, eight, twelve, fourteen, sixteen and twenty 1070s if possible.
This post by itself is like an encounter of the third kind :ugeek:

4090+3089ti & Quad 1080ti
ArchiCAD25, ofcourse Octane & OR-ArchiCAD plugin (love it)
http://www.tapperworks.com
http://www.facebook.com/pages/TAPPERWOR ... 9851341126
http://www.youtube.com/user/Tapperworks/videos
User avatar
smicha
Licensed Customer
Posts: 3151
Joined: Wed Sep 21, 2011 4:13 pm
Location: Warsaw, Poland

Tutor,
I am looking at photos of X10DRX and cannot find any info - is the 2011v3 socket a narrow ILM or a regular mount?
3090, Titan, Quadro, Xeon Scalable Supermicro, 768GB RAM; Sketchup Pro, Classical Architecture.
Custom alloy powder coated laser cut cases, Autodesk metal-sheet 3D modelling.
build-log http://render.otoy.com/forum/viewtopic.php?f=9&t=42540
Post Reply

Return to “Off Topic Forum”