i am working on putting together a PC, mainly for Linux Flame, possibly dual boot Windows. I am doing this for a client, but i will be the one mainly using it for the first 10 months or so.
Of course, we are trying to get the best bang for our buck, without being tooo stingy
here are my proposed specs:
CPU: AMD 7960X, 24C/48T, 4.20-5.30GHz
GPU: 24GB Gigabyte GeForce RTX 4090 AORUS Xtreme Waterforce Aktiv PCIe 4.0 x16
MB: ASUS Pro WS TRX50-SAGE WIFI-SI (AMD TRX50 CEB)
RAM: 4x 64GB Micron MTC40F2046S1RC56BR DDR5-5600 DIMM CL46 Single
system ssd: Samsung 990 PRO NVMe SSD 4 TB M.2 PCIe 4.0 3D-NAND TLC
framestore/cache: 2xSamsung 990 PRO NVMe SSD 4 TB M.2 PCIe 4.0 3D-NAND TLC RAID0
slow storage: Seagate IronWolf Pro 16TB 7200rpm
network card: TBD
Case: TBD
What are your thoughts? any obvious bottlenecks or compatibility issues ? anyone using a similar build? should we rather go intel than amd threadripper?
Buy a used workstation that has more ram slots.
Buy a workstation class GPU.
Especially if your life span on this machine is limited to 10 months.
Ensure that your life span is compatible, after that it’s someone else’s problem.
I would go this way, with threadripper pro lines if possible for you…not only because of the ram slots but general expansion capability, in case you want to add an extra gpu for whatever reason…extra cards such as a raid card or network card. usb cards, etc…, you’ll have the lanes to do that. If you are absolutely sure you will not need anything else, well, this is fine. 24gb gpu might be OK, but if you want to work with high res plates and maybe play with some AI stuff out there 24gb will hit the wall and should go with 48gb gpu
There will be as many opinions as there are configurations on the market.
Unless you like tinkering and don’t mind occasional frustration (not a good look if the client has visibility or inherits it) - get one of the qualified systems. While many have successfully deployed other configs, it can trip you up. And ADSK is less likely to help you if you get stuck.
For the same reason, buy it from someone who can provide hardware support. You may be ok building a system. But if your client inherits the system, they may not care for it. So having someone from Dell/HP come over and fix hardware is a valuable consideration.
Lastly, you can save money with a used workstation. But more than likely you will then be one generation of CPU and bus speeds behind. There have been some bigger steps in performance that are worth capturing on a new system that will be a primary system, not just a backup or burn node. There’s a reason these systems are being sold off.
And most definitely get a workstation class GPU and skip the GeForce temptation. See above points about getting help if needed. Gamer cards have not been qualified. And nobody will help you if things don’t pan out.
This machine is about 15 years old, the flash storage is over 10 years old.
Workstation: HP Z820 Workstation Not Specified
CPU: Intel(R) Xeon(R) CPU E5-2687W 0 @ 3.10GHz, 32 active CPU(s)
RAM: 515381 Meg of RAM
Graphics: Quadro RTX 4000 Driver: 550.78
OS: Rocky Linux release 9.3 (Blue Onyx) - 64bit OS
Kernel: 5.14.0-362.8.1.el9_3.x86_64 DKUversion 19.1.0-PR199
Config: DKUversion 19.1.0-PR199
TEST DISKS : Sequential Full Res Read: 2932.24 MB/s.
TEST DISKS : Random Full Res Read: 2910.10 MB/s.
You could probably buy this kind of rig for about $1500 if you try.
The workstation would probably cost $200, the RAM would be about $200, the GPU would be about $600, the early Intel flash is probably given away in cereal boxes.
p.s. I used to format the flash in much less safe way, and it would max out the PCIe slot at ~5GB/sec.
thanks for the tip! I know you can get old HP Workstations crazy cheap sometimes! But i will definitely need something more up-to-date than that. I was using a z820 in 2017 and it was really struggling.
For z820 only RAM and GPU and direct-attached storage count.
Everything else is a secondary consideration.
This particular workstation has done some unique projects and been in some unusual locations - the first home was on a workbench, second home a remote farm in northern California in the middle of a 40 acre field, with a fiber connection, trucks on location, cars on set, hotel suites, blah blah blah.
It’s really just a big laptop.
z840 are also cheap, you can add even more ram.
DELL 7920 rack is low profile, quiet, expandable, has better power supply configuration.
Lots of current, high-profile, high-dollar, high-framerate, high-resolution (8K and above) projects are run on used DELL 7920s with A5000s, highpoint cards with Samsung NVMe, blah blah blah.
Also, what’s to stop you calling GUNPOWDER and working out a deal on a virtual machine that you turn on, and off like an appliance?
as to the virtual machine solution… I’m not sure it’s the best fit right now. The project i’ll be using it for will have a TON of footage, which will be needed over a longer period of time, so i guess the price of online storage might not be right for that. Also the prices for virtual workstations, at least to me, seemed to make most sense for short term projects, as the costs would add up to a point where it would have been cheaper just to buy it (considering a runtime of close to a year, and the fact that you might run some rendering jobs overnight, not just 9 to 5)
But i am definitely looking into a cloud flame setup for other freelance jobs! It could definitely be the future.
Two years ago, I debated this same question before buying my certified P620 threadripper. By the time you add up the costs for gpu, framestore, network cards, blah blah, the cost savings percentage of a diy cpu vs certified didn’t seem worth the possible aggravation.
And today I would probably just buy a Mac Studio and enjoy longer render breaks, lol.
