Super Computer Build-- Under 10,000k ( for AI, rendering, server)

nidenel1

Distinguished
Apr 12, 2008
7
0
18,510
Hey All,

I am looking to build out a super computer used for AI training, and also as a server.

Here is what I am thinking, multi threading is going to be one of the biggest selling points for me I definitely want something over 20 cores/threads. I am looking at dual Xeon build but I would consider AMD too-- but since this will be for AI research Intel is better suited IMO.

So is Xeon still worth it with the new new I9 coming on line-- it has 18 cores and 32 threads.

My thinking is this:

For the processors spend around 3,000$ (can bump to 4 to 4.5) if needed.

For GPU im thinking Tesla: ~1500$
https://www.newegg.com/Product/Product.aspx?Item=9SIAH597U15219&cm_re=nvidia_tesla-_-14-132-041-_-Product

And for data drive probably a 3 ssd drive raid 0. 1,000 usd

Storage raid 5 ~ 15 tb -- 500$ usd

5.5 k USD for main components.

3k usd for case + PSU + mobo + liquid cooling + accessories

Should be around 10,000.

Any tips on chips / mobos.

-- Is there any networking that I should consider? I would like to consider scalibitly and fast networking switches to add on another similar system and combine them. A la a rack server that is not rack mounted.




 
Look at the new Z4 type workstations from HP. That will give you an idea on how to configure one of the new Xeon scalable hosts. The Xeon scalable have more memory channels. The Xeon scalable uses the same socket for the entire line of CPUs . You can start with a low clock 5000 series and upgrade later.
 
Qty. Product Description Savings Total Price
1
Phanteks Enthoo Luxe PH-ES614LTG_AG Black Aluminum Faceplates Tempered Glass Panel Steel Chassis ATX Full Tower Computer Case with RGB Illumination

Phanteks Enthoo Luxe PH-ES614LTG_AG Black Aluminum Faceplates Tempered Glass Panel Steel Chassis ATX Full Tower Computer Case with RGB Illumination
Item #:N82E16811854044
Return Policy: Standard Return Policy

$179.99
$145.20

Add Phanteks Enthoo Luxe PH-ES614LTG_AG Black Aluminum Faceplates Tempered Glass Panel Steel Chassis ATX Full Tower Computer Case with RGB Illumination to cart
3
SAMSUNG 860 EVO Series 2.5" 1TB SATA III V-NAND 3-bit MLC Internal Solid State Drive (SSD) MZ-76E1T0B/AM

SAMSUNG 860 EVO Series 2.5" 1TB SATA III V-NAND 3-bit MLC Internal Solid State Drive (SSD) MZ-76E1T0B/AM
Item #:N82E16820147673
Return Policy: Standard Return Policy

$599.97
$488.97

Add SAMSUNG 860 EVO Series 2.5" 1TB SATA III V-NAND 3-bit MLC Internal Solid State Drive (SSD) MZ-76E1T0B/AM to cart
4
TOSHIBA MD04ACA400 4TB 7200 RPM 128MB Cache SATA 6.0Gb/s 3.5" Internal Hard Drive

TOSHIBA MD04ACA400 4TB 7200 RPM 128MB Cache SATA 6.0Gb/s 3.5" Internal Hard Drive
Item #:N82E16822149644
Return Policy: Standard Return Policy

$599.96
$515.96

Add TOSHIBA MD04ACA400 4TB 7200 RPM 128MB Cache SATA 6.0Gb/s 3.5" Internal Hard Drive to cart
1
EK-Annihilator EX/EP

EK-Annihilator EX/EP
Item #:9SIAC8W77A7334
Sold by EK Water Blocks What's this?

$159.99

Add EK-Annihilator EX/EP to cart
1
ASUS WS C621E Sage EEB Server Motherboard Dual LGA 3647 Intel C621

ASUS WS C621E Sage EEB Server Motherboard
Item #:N82E16813119064
Return Policy: Standard Return Policy

$544.69

Add ASUS WS C621E Sage EEB Server Motherboard to cart
2
CORSAIR Vengeance LPX 32GB (2 x 16GB) 288-Pin DDR4 SDRAM DDR4 2666 (PC4 21300) Memory Kit Model CMK32GX4M2A2666C16

CORSAIR Vengeance LPX 32GB (2 x 16GB) 288-Pin DDR4 SDRAM DDR4 2666 (PC4 21300) Memory Kit Model CMK32GX4M2A2666C16
Item #:N82E16820233853
Return Policy: Standard Return Policy

$531.98

Add CORSAIR Vengeance LPX 32GB (2 x 16GB) 288-Pin DDR4 SDRAM DDR4 2666 (PC4 21300) Memory Kit Model CMK32GX4M2A2666C16 to cart
1
SAMSUNG 970 EVO M.2 2280 1TB PCIe Gen3. X4, NVMe 1.3 64L V-NAND 3-bit MLC Internal Solid State Drive (SSD) MZ-V7E1T0BW

SAMSUNG 970 EVO M.2 2280 1TB PCIe Gen3. X4, NVMe 1.3 64L V-NAND 3-bit MLC Internal Solid State Drive (SSD) MZ-V7E1T0BW
Item #:N82E16820147691
Return Policy: Standard Return Policy

$277.99

Add SAMSUNG 970 EVO M.2 2280 1TB PCIe Gen3. X4, NVMe 1.3 64L V-NAND 3-bit MLC Internal Solid State Drive (SSD) MZ-V7E1T0BW to cart
1
Intel Xeon Scalable Gold 6140 SkyLake 18-Core 2.3 GHz (3.7 GHz Turbo) LGA 3647 140W BX806736140 Server Processor

Intel Xeon Scalable Gold 6140 SkyLake 18-Core 2.3 GHz (3.7 GHz Turbo) LGA 3647 140W BX806736140 Server Processor
Item #:N82E16819117897
Return Policy: Replacement Only Return Policy

$2,439.99

Add Intel Xeon Scalable Gold 6140 SkyLake 18-Core 2.3 GHz (3.7 GHz Turbo) LGA 3647 140W BX806736140 Server Processor to cart
1
CORSAIR HXi Series HX1200i 1200W 80 PLUS PLATINUM Haswell Ready Full Modular ATX12V & EPS12V SLI and Crossfire Ready Power Supply with C-Link Monitoring and Control

CORSAIR HXi Series HX1200i 1200W 80 PLUS PLATINUM Haswell Ready Full Modular ATX12V & EPS12V SLI and Crossfire Ready Power Supply with C-Link Monitoring and Control
Item #:N82E16817139129
Return Policy: Standard Return Policy

$219.97

Add CORSAIR HXi Series HX1200i 1200W 80 PLUS PLATINUM Haswell Ready Full Modular ATX12V & EPS12V SLI and Crossfire Ready Power Supply with C-Link Monitoring and Control to cart
1
EKWB EK Fluid Gaming A240 Complete Dual 120mm Water / Liquid Cooling Kit 240mm

EKWB EK Fluid Gaming A240 Complete Dual 120mm Water / Liquid Cooling Kit 240mm
Series: N/A Fan Size: 2 Fan Slots (120mm) Color: Black Model: A240
Item #:9SIAC8W5SB2261
Sold by EK Water Blocks What's this?

$159.99

Add EKWB EK Fluid Gaming A240 Complete Dual 120mm Water / Liquid Cooling Kit 240mm to cart
1
NVIDIA Tesla K80 2 x Kepler GK210 900-22080-0000-000 24GB (12GB per GPU) 384-bit GDDR5 PCI Express 3.0 x16 GPU Accelerators for Servers

NVIDIA Tesla K80 2 x Kepler GK210 900-22080-0000-000 24GB (12GB per GPU) 384-bit GDDR5 PCI Express 3.0 x16 GPU Accelerators for Servers
Item #:9SIAH597U15219
Sold by CNE Direct What's this?

$1,550.00

Add NVIDIA Tesla K80 2 x Kepler GK210 900-22080-0000-000 24GB (12GB per GPU) 384-bit GDDR5 PCI Express 3.0 x16 GPU Accelerators for Servers to cart
Grand Total: $7,034.73





 


No, the Z4s are $5K give or take. But a specific model was NOT my point. My point is looking at configurations that are sold commercially to get some info on what makes sense to put together. The HP quickspecs are a treasure trove of configuration do and don't information.

For example, to maximize performance on your dual Xeon 6xxx CPU host, you should have 12 DIMMs. Those CPUs have 6 channel memory controllers. You will maximize performance by adding 6 DIMMs / CPU, so a dual socket config should have 12 DIMMs -- 192GB typically.