F5F Stay Refreshed Hardware Desktop Which components should be selected for machine learning in 3D medical imaging?

Which components should be selected for machine learning in 3D medical imaging?

Which components should be selected for machine learning in 3D medical imaging?

N
Nybu
Member
Find
160
07-01-2025, 10:21 PM
#1
Hello, I need some guidance on the best choices.
My budget is approximately 1500 dollars.
I currently have: PSU, hard drives, an Intel Ultra 9 285K CPU, and two RTX 3090 graphics cards.
Now I require a motherboard, cooling system, RAM, and a case.
Background information: My main use is prototyping machine learning models in medical imaging, with complex code. Compiling these models takes a long time—sometimes up to three hours—which makes it necessary for me to keep my existing GPU and power supply but upgrade to a better CPU and faster RAM. In about two years, I plan to replace the GPU with a newer model while still keeping two GPUs.
I use Linux Ubuntu.
I don’t mind the appearance of the PC, nor do I need it to be very quiet; it’s placed in a storage room with cables running through the wall to reach my workspace.
I believe my RAM speed is crucial for handling those heavy compilation tasks, so I’m looking for at least 96 GB of RAM.
I want a motherboard that supports two GPUs, though I’m not sure how critical it is to have two PCIe 5.0 slots—especially since my models are often very complex and require more than a second to load after GPU memory updates. I’m unsure whether sharing parameters between GPUs during communication in multi-GPU training is really important.
I’ve been considering the Gigabyte AI Top Z890, which has two PCIe 5.0 slots, and ASUS ROG Hero PCIe 5.0 and PCI 4.0 models that seem to offer better RAM performance.
Additionally, to run the latest CUDA versions (some sold up to 9200 MT/s), I’d need to enable extreme memory profiling on my Intel CPU—this raises concerns about safety, and I’m also wondering if separate cooling for RAM would be necessary.
I’m a software professional and tend to get overwhelmed by hardware decisions.
Thank you for your assistance!
Reply
Reply
N
Nybu
07-01-2025, 10:21 PM #1

Hello, I need some guidance on the best choices.
My budget is approximately 1500 dollars.
I currently have: PSU, hard drives, an Intel Ultra 9 285K CPU, and two RTX 3090 graphics cards.
Now I require a motherboard, cooling system, RAM, and a case.
Background information: My main use is prototyping machine learning models in medical imaging, with complex code. Compiling these models takes a long time—sometimes up to three hours—which makes it necessary for me to keep my existing GPU and power supply but upgrade to a better CPU and faster RAM. In about two years, I plan to replace the GPU with a newer model while still keeping two GPUs.
I use Linux Ubuntu.
I don’t mind the appearance of the PC, nor do I need it to be very quiet; it’s placed in a storage room with cables running through the wall to reach my workspace.
I believe my RAM speed is crucial for handling those heavy compilation tasks, so I’m looking for at least 96 GB of RAM.
I want a motherboard that supports two GPUs, though I’m not sure how critical it is to have two PCIe 5.0 slots—especially since my models are often very complex and require more than a second to load after GPU memory updates. I’m unsure whether sharing parameters between GPUs during communication in multi-GPU training is really important.
I’ve been considering the Gigabyte AI Top Z890, which has two PCIe 5.0 slots, and ASUS ROG Hero PCIe 5.0 and PCI 4.0 models that seem to offer better RAM performance.
Additionally, to run the latest CUDA versions (some sold up to 9200 MT/s), I’d need to enable extreme memory profiling on my Intel CPU—this raises concerns about safety, and I’m also wondering if separate cooling for RAM would be necessary.
I’m a software professional and tend to get overwhelmed by hardware decisions.
Thank you for your assistance!

Reply
Reply
M
Mr_Floobiful
Posting Freak
Find
890
07-01-2025, 10:21 PM
#2
I think you're aiming to integrate PCIe5 GPUs once they're ready and cost-effective, but for now it's better to focus on PCIe4 and not stress too much about what could be available in 2027.
https://www.pcguide.com/gpu/what-gpu-has-pcie-5/
May the best luck come your way when running at 9200MT/s. RAM speed matters, but system memory is key—do your RTX 3090s have enough video RAM? Some RTX A6000 models include 48GB of VRAM.
The quicker you overclock RAM, the higher the chance of crashes. Whether you need separate cooling for RAM is another question. I currently have one old setup with a dual-fan RAM cooler installed over four hot DIMMs. All other systems work just fine without dedicated cooling, but I rely on air coolers for my CPUs, which provides some airflow around the DIMMs and VRMs.
Reply
Reply
M
Mr_Floobiful
07-01-2025, 10:21 PM #2

I think you're aiming to integrate PCIe5 GPUs once they're ready and cost-effective, but for now it's better to focus on PCIe4 and not stress too much about what could be available in 2027.
https://www.pcguide.com/gpu/what-gpu-has-pcie-5/
May the best luck come your way when running at 9200MT/s. RAM speed matters, but system memory is key—do your RTX 3090s have enough video RAM? Some RTX A6000 models include 48GB of VRAM.
The quicker you overclock RAM, the higher the chance of crashes. Whether you need separate cooling for RAM is another question. I currently have one old setup with a dual-fan RAM cooler installed over four hot DIMMs. All other systems work just fine without dedicated cooling, but I rely on air coolers for my CPUs, which provides some airflow around the DIMMs and VRMs.

Reply
Reply
M
Mirrqred
Member
Find
61
07-01-2025, 10:21 PM
#3
Thank you for your reply.
I agree with your thoughts about the 2027 weather, I would prefer to stick with my current motherboard and need PCI5 then, thank you for sharing your ideas.
Regarding CUDIMM, I thought 9200 might be too slow given my plan to go with an Intel 285k processor, so I meant faster options are available.
If it comes to crashes, I assume I could just lower the settings and it should be fine, but I'm worried about the long-term risk of damaging my hardware permanently.
More VRAM and CUDA cores would always help, but right now what's slowing me down is mainly CPU preprocessing, then GPU speed, and finally GPU memory—though I expect things to improve over time. So I'm confident I'll need a better GPU soon, possibly in two years, or maybe sooner if I save more now.
Also, do you have any hardware suggestions?
Thanks!
Reply
Reply
M
Mirrqred
07-01-2025, 10:21 PM #3

Thank you for your reply.
I agree with your thoughts about the 2027 weather, I would prefer to stick with my current motherboard and need PCI5 then, thank you for sharing your ideas.
Regarding CUDIMM, I thought 9200 might be too slow given my plan to go with an Intel 285k processor, so I meant faster options are available.
If it comes to crashes, I assume I could just lower the settings and it should be fine, but I'm worried about the long-term risk of damaging my hardware permanently.
More VRAM and CUDA cores would always help, but right now what's slowing me down is mainly CPU preprocessing, then GPU speed, and finally GPU memory—though I expect things to improve over time. So I'm confident I'll need a better GPU soon, possibly in two years, or maybe sooner if I save more now.
Also, do you have any hardware suggestions?
Thanks!

Reply
Reply
P
PowergirlA
Member
Find
229
07-01-2025, 10:21 PM
#4
PCPartPicker Part List
CPU Cooler:
*
ARCTIC Liquid Freezer III 360 56.3 CFM Liquid CPU Cooler
($136.81 @ Amazon)
Motherboard:
*
Asus ProArt Z890-CREATOR WIFI ATX LGA1851 Motherboard
($489.99 @ Amazon)
Case:
*
Corsair 7000D AIRFLOW ATX Full Tower Case
($209.99 @ Corsair)
Total:
$836.79
Costs cover shipping, taxes, and applicable discounts
*Selected lowest-priced items based on criteria
Created by
PCPartPicker
2025-04-08 03:55 EDT-0400
Links provided for reference
https://www.asus.com/us/motherboards-com...ator-wifi/
https://www.corsair.com/us/en/p/pc-...ow...9011218-ww
Reply
Reply
P
PowergirlA
07-01-2025, 10:21 PM #4

PCPartPicker Part List
CPU Cooler:
*
ARCTIC Liquid Freezer III 360 56.3 CFM Liquid CPU Cooler
($136.81 @ Amazon)
Motherboard:
*
Asus ProArt Z890-CREATOR WIFI ATX LGA1851 Motherboard
($489.99 @ Amazon)
Case:
*
Corsair 7000D AIRFLOW ATX Full Tower Case
($209.99 @ Corsair)
Total:
$836.79
Costs cover shipping, taxes, and applicable discounts
*Selected lowest-priced items based on criteria
Created by
PCPartPicker
2025-04-08 03:55 EDT-0400
Links provided for reference
https://www.asus.com/us/motherboards-com...ator-wifi/
https://www.corsair.com/us/en/p/pc-...ow...9011218-ww

Reply
Reply