Threadripper vs Ryzen 9950X

GazzaMataz

GazzaMataz

GazzaMataz

Associate
Joined
2 Nov 2006
Posts
141
Location
Kent, UK
I have a Threadripper 1920 system which I built back in 2018/9. I bought it since it allowed a lot more lanes on the PCI bus since I wanted to use the machine with multiple GPU for 3D rendering. It has served me well but is getting on a bit especially the single core speed in Cinebench 2024 which is about 60... I am thinking of upgrading to the Ryzen 9950x cpu which I think scores 130 in Cinebench on the single core front?

Anyway, I was wondering most recent motherboards only seem to have two GPU slots these days but then again the GPUs are a lot faster so you wouldn't need four of them in a build - maybe two? Also I wonder how much more performance I would be getting over my Threadripper 1920 against the Ryzen 9950X - twice as fast? Not interested in the Ryzen 9950X3D since it doesn't make any difference for what I want to use it for.

Looking forwards to anything anyone has to shed on the scenario.
 
Consumer motherboards have only sufficient PCIe lanes for one GPU and may have to resort to multiplexing if you want more than 3 or 4 NVME drives. Don't rule out Threadripper 3000 CPUs which will enable you to reuse your RAM:


Unfortunately you might have to go second-hand with the motherboard.

The Level1Techs YouTube channel is probably what you want to mine for info, but the Adamant IT channel found that the Arc B580 rendered video about as fast as a RTX 4080, and Maxsun do a dual B60 (2x 24 GB professional B580) card so as long as your PC supports PCIe bifurcation. Cost, according to Reddit, is £1800 from an OCUK competitor - I expect @Gibbo can do much better. The single B60 is a third of the price.
 
The PCI-e lane thing is a wet blanket on pretty much all consumer platforms in recent generations, sure most people don't hit the limits of it but still. You are pretty much relegated to 1x x16 slot and then either splitting to 2x x8 or 1x x16 and using an x4 slot, etc. for an extra GPU and a struggle with more than 2.

CPU wise a lot of stuff will obliterate that Threadripper today.
 
Sounds like unless I want to shell out on a new threadripper system, which seem to be MUCH more expensive these days, I would be better off with a Ryzen 9950x cpu and a newer GPU. I use two RTX 2800ti cards which have been great but I guess an RTX 4080 or RTX 5080 would be more than twice the speed and only one card.
 
Problem with a 9950x system
Is the insane price of ddr5 right now
Assuming you'll want a fair amount of ram
If a workstation is the intended use

The 5080 is terrible value too

Really is probably the worst time I remember
When It comes to building a decent pc
Thanks to ai data centers grabbing everything
 
I'd buy a used system of some sorts or sweat your current hardware another 18-24 months. This is, as Mcnumpty says, the worst possible time to do a new build.
 
Should probably say
Despite the current awful situation
If its for business use
Rather than personal use
That may alter things somewhat
As the painful state of ram prices
And the current gpu situation could be passed
On to customers and recouped

Must admit I don't really follow
Workstation prices new or used
But as suggested a second hand Workstation
May be a good option
 
First of all, you don't have to worry about PCI-E lanes for GPU rendering (realistically). The bigger problem with consumer boards is the physical spacing of the slots unless you're willing to buy PRO/Quadro cards, or can get a couple of 5090FE's. As long as you buy a board that will give you 2 x 8x slots at 4.0 you're absolutely fine. I run 2 x 4090FE's on an Asus Dark Hero with 3 internal m2 drives and it still flies. There is practically no performance difference for offline GPU rendering between x16 4.0 and x8 4.0. In fact you can go even lower.

You need to work out what you want in the machine and then figure out what you'll need to make it all fit, including the case. I couldn't run a case with a PSU shroud at the bottom for example, because the second card overhangs the motherboard, so I got a Lian-Li O11D XL that has tons of space below the motherboard tray.
 
Last edited:
We have replaced some 1st/2nd gen TR workstations at work and have went with 9950X based systems, I would say Zen 3 performance per core is the minimum these days. When it first launched TR was very cheap for what you got core wise, now its very expensive and you have to be sure its what you need.
Multicore 9950X is about 50% faster than 2990WX and about 300% the 1950X, single thread wise they are more than 200%.
 
I don't think I would go down the threadripper route these days since those machines are very expensive and I don't do as much work as I used to. I have been told that the logic board I had my eye on, the MSI X870E Carbon, apparently cannot run 2 GPU's at x16, it can run two GPU's but it would be 8/4 configuration as they share the bandwidth. The ASUS ProArt X870E-Creator will run 8/8 in certain configurations... But as Adrian said this will probably have no effect on GPU rendering. I mean what's the point of two GPU slots on these boards if you cannot use them?
 
GazzaMataz said:
I mean what's the point of two GPU slots on these boards if you cannot use them?
Click to expand...

Yeah something I find annoying albeit not many people run 2x GPUs these days, though that will increase a bit with the rise of AI.

On my Intel Z790 setup I can only do x16 + x4 for the GPUs and hilariously there is a 3rd x16 size socket but it only runs at x1.
 
Well they are PCIe slots not strictly GPU slots, so capture cards, high performance NICs, Extra NVME via PCie, specalist cards etc many uses but yeah indeed most users are just using the x16 slot and the rest of the stuff provided by the onboard components.

HEDT like x79/99 and then X399 used to provide an entry level middle ground between consumer platform and full on server/workstation and you would only pay a bit more but now you pay way more from the get go, boards are like 3-4x the price of an average consumer board and cpu is 2x price of 9950x for the lowest.
 
Toasty said:
Well they are PCIe slots not strictly GPU slots, so capture cards, high performance NICs, Extra NVME via PCie, specalist cards etc many uses but yeah indeed most users are just using the x16 slot and the rest of the stuff provided by the onboard components.

HEDT like x79/99 and then X399 used to provide an entry level middle ground between consumer platform and full on server/workstation and you would only pay a bit more but now you pay way more from the get go, boards are like 3-4x the price of an average consumer board and cpu is 2x price of 9950x for the lowest.
Click to expand...

It was all over for me once we got 16 cores in consumer parts. I had a TR workstation with a 3970x and I built a 5950x system for home pretty much the week they came out. Naturally work ends up bleeding a bit between the two and besides being a bit more limited on ram at the time so much stuff was absolutely fine on the 5950x. Now we have things like the 9950x that are not only faster for a bunch of tasks, but we can easily have 192gb+ ram in them (apparently 256gb but I've never seen this in the wild, not that the price is sensible on either these days :mad:), and realistically still enough PCI-E lanes for a ton of different use cases. Unfortunately with covid gouging + current AI driven prices it does feel like that period was a bit of a golden age that is now coming to an end/already over.
 
Adrianr said:
Unfortunately with covid gouging + current AI driven prices it does feel like that period was a bit of a golden age that is now coming to an end/already over.
Click to expand...
I am beginning to think the same Adrian. It was bad enough with the cost of GPUs a few years ago with the miniers which is why I bought both of my RTX 2800ti GPUs second hand. And they have been fine. The only thing I wouldn't buy second hand would be hard drives - data is not replaceable ;)
 
You must log in or register to reply here.
Back
Top Bottom