For software development purpose is threadripper pro 3955wx enough for me ?, do you think 3970x or 3955wx or 3975wx is better for compiling large codebase ?, in my area 3970x is very hard to find
Welcome to the forums!
Do you need the ram and\or pcie lanes the platform allows for ?
if no, then a 5950x could be a better and cheaper choice.
still a 16core part, but faster.
Else I would go for the 32core part.
Absolutely, I needs a lot of PCIe lanes, for pure computing power can 3955wx beat 3970x ?, 3970x is very hard to find also 3975wx is overpriced in my local area
Can 3955wx IPC be 200% of 3970x IPC ?
Unlikely. That would be a heck of a jump. Check this CPU compare site. There are many others to check if you want an overall impression.
The only 3970x available are through scalpers and horribly overpriced. Wait for the TR PRO 5000’s to come out if you want a Threadripper. It depends on your workload and budget so [VIIgraphics] (Level1Techs Forums) advice for the 5950X may work for you.
1 Like
3955 is a 16core part,
3970/3975 are 32 core parts, all of those are based in zen2 architecture, ipc per core should be very close and overall performance should be scaling quite good, but definitely not linearly.
The best you can do (if you can) is wait for the 5000 series, I’m doing the same.
Judjing from what is written, I think that you are referring to the 64core one, which is 3995wx (and not 3955)
My advice saddly is the same, since the consumer platform will leave you unsatisfied if you need both ram and pcie lanes.
Not 64 core version, its 16 core version 3955x, but someone in youtube say 16 pro version can beat 32 non pro version, under some circumstance and workload, do you think thats true ?
3955wx is slower than 5950x which both are 16core parts, the latter one is based on a newer node.
Pro or non pro version doesn’t matter much, as they are the same architecture, but in some cases non pro is faster as well.
For example non pro threadrippers have higher clocks and are unlocked, compared to the 3000 series wx models (this will change in 5000)
In single / few threaded tasks (photoshop, 3d editor etc)
the faster chip may has the upper hand, but in all core load, corecount is much more important.
I’m not familiar with code compiling, and the intricacies of your workload, why you absolutely need the pcie lanes etc. If you really do, you have no choice, get the 3955wx and you can upgrade later to a 5000 series part. that would be the more sensible thing right now, even if it will cost you a bit extra for the stop gap solution.
if you do find a 32core part, it is the sweetspot for most people.
Could you explain this a bit more? PCIe lanes would only help if you mean that you’re adding a bunch of GPUs or something. PCIe lanes don’t have anything to do with CPU performance.
That’s something you have to judge as I don’t know your work load or expectations.
My opinion: If you are just doing your codebase then a 5950x is fine. If you want to run that load plus a couple of VM’s and render video at the same time then get a TR. YMMV.
The market is about to flood with 5000 series threadripper. but then it is also going to cost more
The 5950x/16 cores… if 128gb ram is enough for you? the 5950x is a better choice.
It would be nice to know what @meowsoft is going to use the pcie lanes for, but the 32 core non pro will be faster than the 16 core pro except, perhaps, in computational fluid dynamics situations.
Devops and code compile? The 5950x will beat the 3955 except unless your devops job is as large as microsoft or you really need 256-512gb memory.
Could I interest you in an epyc rome or milan “P” series cpu however? It is till possible to find cpus like the 24-core epyc rome 7502p for around $1500 US and this is a killer deal if you dont mind running a server platform as a workstaion a la our swerkstation videos.
Details man, detail!s!! 
2 Likes
Well said! Do we know when this ‘‘about’’ is going to happen?
Yes, and individuals will need to look at their own needs and assess price vs. performance. At the end of the day, it all comes down to what you want to pay based on your workloads.
HEDT TR would have been the sweet spot for me in that scenario. However, that option is not present today, so I have to choose to go with the cheaper 5950x and live with that performance or over-buy and go with a 32 core PRO TR. For future proofing, I suspect I need to go the PRO route.
Unless, of course, some tall handsome YouTuber offers to sell me one of his used TR 3970x when the 5000 series rolls in. 
I needs a lot of virtualization, also experiments with large codebase, and of course I use six GPU setup for deep learning, I thin 3970x is better but is out of stock
I think R9 5950x plus TR 3955WX or TR 3970x plus TR 3955WX will be perfect choice, two workstation
@wendell do you think 3955wx is good for deep learning with 6x RTX 3090 ?, I have a lots of 3090 for deep learning, maybe in such case I can build 12 gpu setup with PCIe splitter (each x16 splitted to two x8)
And I R9 5950x still good for large codebase and much virtualization ?, I think it will be good using 3970x but it is out of stock
I think two workstation is better
It was tested with R9 5900x but still not enough performance (for software dev purpose only, no deep learning)
the motherboard will be critically importan – youd need a tr pro motherboard with extra power inputs to run that many gpus at once. even then 6 is pushing it. Even 3 would be pushing it on a regular board if it didnt have supplimental pcie power inputs.