dual 8180m miniITX build

Venturi

Limp Gawd
Joined
Nov 23, 2004
Messages
264
Asus c621e sage
Dual Xeon 8180M
384GB of ram
4x Titan V including Titan V CEO ed 32GB
960pro nvme (OS)
10x SSD RAID (Apps and Backup 8x 2x )
1600W PSU (digital) noiseless
MS 2019 Data Center, Ubuntu
TT miniITX case

IMG_9263.JPG

IMG_9013.jpg

IMG_9019.JPG

IMG_9068.JPG

IMG_9061.JPG
 
Last edited:
VERY nice. What's the use case for this beast?


edit: did you get a custom mount setup for the SSDs or does the P1 support that out of the box? Either way, looks great.
 
VERY nice. What's the use case for this beast?


edit: did you get a custom mount setup for the SSDs or does the P1 support that out of the box? Either way, looks great.

I made the mount for the SSDs - it is a CNC backplate for all of them. (Thermaltake only made the original P1 case, and there is very little of the original left)


It is my personal rig but on occasion I do work on it: isotopes, medical imaging, deep learning algorithms (ANN CNN) and

gaming, email, papers, school, billing, etc
 
Ray Tracing is alive and well in metro exodus at 4k with my Titan Vs

IMG_9230.JPG
 
What's your monitor? Love the rest of that rig. Doubt the wife would let me build one like that. (Not to mention the crazy electric bill cooling that thing in an Arizona summer...)
 
The monitor is an asus pq329

I’ll trade Virginia for Arizona in a heartbeat!
 
Holy fuck, that's a $25,000+ build. What the fuck? School papers? Did daddy win the lottery?
 
Now that you have the 8180M's, the obvious next step is to upgrade to a quad-socket. Sadly it doesn't seem like a "normal" form factor quad Platinum board exists yet - all you get is this, which will be a nightmare to fit in any case. I wonder how your rig compares to a quad E7 V4 system - the Skylakes are certainly faster and 2P's scale better than 4P's, but it is 96 cores versus 56...
Building the "fastest possible computer" has gotten trickier and trickier. The EVGA SR2 was the last board that you could get everything on - it was never slower than a 980X, and sometimes twice as fast. After that we lost overclocking, and to further salt the wounds, the internal layout of the Xeons became more and more complicated. These days even with something like dual 8180M's or 2699V4's, you need to keep a spare 9700K rig around for things that don't scale well.
 
Now that you have the 8180M's, the obvious next step is to upgrade to a quad-socket. Sadly it doesn't seem like a "normal" form factor quad Platinum board exists yet - all you get is this, which will be a nightmare to fit in any case. I wonder how your rig compares to a quad E7 V4 system - the Skylakes are certainly faster and 2P's scale better than 4P's, but it is 96 cores versus 56...
Building the "fastest possible computer" has gotten trickier and trickier. The EVGA SR2 was the last board that you could get everything on - it was never slower than a 980X, and sometimes twice as fast. After that we lost overclocking, and to further salt the wounds, the internal layout of the Xeons became more and more complicated. These days even with something like dual 8180M's or 2699V4's, you need to keep a spare 9700K rig around for things that don't scale well.


I will agree, I loved the sr-2 and the sr-x (minus the cheesy trickle like bios support from evga). It did have a quirky memory map with uneven ram slots per cpu....

As for a quad set up:
it is not possible to get a quad intel or dual Epyc board with: full 4x pcie dual spaced gpu sli capable board. It is also not possible to get a motherboard manufacturer that is willing to put that much effort into a workstation board.

My supermicro x10drg-q was painful to put into a workstation. I tried the z9 and z10 asus builds as well.

Before that I tried a bulozer 4 socket build, also painful...

Before that I built a tyan dual socket board with a dual socket daughter board (4 socket total) and that was a near impossible feat just to put it into a workstation enclosure.




Sadly, the days of uber workstation builds is coming to an end. The environment is rack-space not desktop. Yes, I could put a whole rack system on top of my desk but I like silent (and also not an eye sore).



whats worse... I just sourced a pair of 8280M.... and the ecc 2933 ram. But before I pull the trigger... the performance increase is so negligible that I can't justify it. I got my 8180Ms for less than 1/4 the price. Otherwise I would never part with that kind of money. Also in order to buy those I sold 2x e5-2699 v4, 2x e5-2699 v3, a pair of v2 2690s, and the all the ram I had accumulated from the other two builds.

I work 3 jobs, I'm getting my doctorate, wife, kids, ... the hobby has to pay for itself.





in this thread, if you scroll you can see a brief history of many of those builds (with pictures)

https://hardforum.com/threads/prototype-4x-titan-v.1950322/page-3#post-1043401212
 
Last edited:
I'm impressed you were able to find 8180M's for that much below list price - the 8173M's are a pretty good deal (under $2K a piece) but 8180's are pretty elusive.
I really like the compact build - I almost put the quadsocket in my sig into a Thermaltake P3, but decided that putting it in a huge case (Thermaltake W100) would be kind of entertaining. Maybe I will revisit the compact concept for my 8890V3's...
 
What a fantastic build, but man you are disgracing that awesome PC, and your mac pro with some awful cable management. Get some velcro asap!
 
What a fantastic build, but man you are disgracing that awesome PC, and your mac pro with some awful cable management. Get some velcro asap!

my cable management on the PC is Zen like BUT he cable management around the desk...shameful. Its the glass desk! I have tried and tried, but have not come up with the perfect solution yet. I need those Star Trek EPS conduits....
 
Small update

Moved to LRDIMM

So the unit now has 768GB of ram


I’m waiting for when I try to turn it off and it says “What are doing Dave?’
 
I am curious about "medical imaging". What are you doing that you need that much power in a workstation?
 
I am curious about "medical imaging". What are you doing that you need that much power in a workstation?

I work in medical amd my dissertation straddles many parts.

*the dissertation - machine learning to generate training dat / verification data for DICOM images and radiologist report correlation

*the daytime job - 3r diagnostic reconstruction, PACS, radiology,

*2nd daytime job - nuclear isotopes,

*3rd daytime job - hospital database support



so, my personal rig gets a few things done ....
 
So if I understand correctly. Your desertation is about using AI to double check radiologist diagnosis using the image they read from?

Regarding the recon, what software are you using and is it related to processing data for your desertation? What modality?
 
So if I understand correctly. Your desertation is about using AI to double check radiologist diagnosis using the image they read from?

Regarding the recon, what software are you using and is it related to processing data for your desertation? What modality?

not exactly - I'm focusing on the training data on pulmonary embolisms and a scoring criteria ( in this case CT, not MRI) my MRI component is using Neuroquant derived data and a separate research scope on hippocampal atrophy scoring and AD (for a separate NHS study).
 
Yes, the training data is for the machine learning

So with this data the computer will be able to detect pulminary embolism in a CT scan and then give you some kind of a confidence rating or flag it for the doctor to review?
 
hence the scoring criteria

This is cool stuff! The closest I have come to a similar technology was a Roche slide scanner that could detect cancerous cell in pathology specimen slides.

And for the AD you are using MRI to look for the Alzheimer's markers and train a computer to flag that?

How are you obtaining your imaging? Are you using phantoms or real patient data (masked obv)?
 
This is cool stuff! The closest I have come to a similar technology was a Roche slide scanner that could detect cancerous cell in pathology specimen slides.

And for the AD you are using MRI to look for the Alzheimer's markers and train a computer to flag that?

How are you obtaining your imaging? Are you using phantoms or real patient data (masked obv)?


I'm using real patient data. All the identifiable patient information was anonymized. But I did keep age and sex

Prior imaging (Petabyte + PACS archives) is part of training data. New imaging is from 3T MRI units, CT is from 64-slice units. Images are lossless compression with no smoothing or filtering.

Its not as romantic as it sounds, I've had to re-learn old math skills to pull off some the algorithms - my wax boards look like chalk boards from some Einstein biography. If I had to do it again I would have picked some social sciences dissertation and laid back .

;)
 
Last edited:
What was the reason for getting the M? I thought the only advantage of the M is to address 1TB+ per CPU.
 
What was the reason for getting the M? I thought the only advantage of the M is to address 1TB+ per CPU.

There are some other enabled/available features on the M, but the real reason is that it was available at 1/6 the price of retail.... so it was that purchase that inspired me to go from my x10DRG-q dual xeon E5-2699 v4 build to the sage e621 dual 8180M build
 
Back
Top