Advice Needed: CPU Upgrade to Pair with Upcoming Nvidia 5090 for Gaming, Editing, and ML

[jonneymendoza]

Hey everyone,

I’m looking for advice on what CPU to upgrade to in preparation for the upcoming Nvidia 5090 GPU. Here’s a breakdown of my use case:

1. Gaming: I play the latest, graphically intensive AAA titles on a 4K monitor.


2. Development: I code and compile Kotlin and Python projects regularly.


3. Photo Editing: I use Lightroom Classic to process and edit 50MP RAW files (it feels pretty slow for me right now).


4. Video Editing: I create and edit 10–20 minute 4K videos recorded on my camera.


5. Future Workload: I plan to start running machine learning models, specifically LLMs.



My current CPU is a Threadripper 3960x, and I’ve been considering either upgrading to the Threadripper 7960x or waiting for the 9950x3D

Questions:

Is the Threadripper 7960x worth the upgrade for my workload, or should I hold out for the 9950x3D?

Are there other CPUs I should consider instead for this kind of workload?

Are there any interactive benchmark tools or graphs where I can compare my current CPU (3960x) with the latest ones?


Any advice or resources would be much appreciated!

Thanks in advance!

 

[Rroff]

Honestly a horrible place at the moment for your kind of use - the 6 core CCDs are not forward thinking for gaming with some newer games starting to push their capabilities when paired up with a high end GPU, the 285K is strong for some of your usage but would be weak for other stuff and silly overpriced for the gaming performance it provides which at times struggles to beat the 14700K. Probably the best option would be the 9950X3D.

EDIT: GN review here includes the older TR as well: https://gamersnexus.net/cpus/amds-cheap-threadripper-hedt-cpu-7960x-24-core-cpu-review-benchmarks for gaming the 7960x doesn't look like a good pairing with a 5090.

 

[Meddling-Monk]

Personal thoughts:

1. Gaming. Performance is mostly based on your GPU. However, having used an RX580 8GB on an i7 4930K and then moving to a 7950X and getting much more out of the RX580 (in Baldur's Gate 3), as well as an RTX 3060 12GB running at less than 30fps on 4K (native and with dlss) on an i5 3570 to a 9800x3D (and going all the way to around 45FPS in 4K with everything on native and with 60+FPS using ultimate dlss 4K - all in Act 3 of BG3 so you know it's potentially lag city), I would say there's some impact from CPU performance of your GPU for gaming. I wouldn't go less than either the single 7800x3D or 9800x3D if gaming is most important. In a pinch, depending on other factors, I would consider also the 7950X, 9950X and 9950x3D (not out yet). 7950X as that's a proven brute force CPU (especially for productivity and if you clock it well, it can really go far even for gaming), the 9950X as that is the current gen version of the previous chip, and the 9950x3D as that's the supposed upgrade "middleground" between gaming and productivity in this current gen. The 7950x3D is not suggested because the 7000 series of CPU's are 100% silicon lottery based on getting any large amounts of memory going at any decent speed. Whilst the 9000 series at least with a decent board has a better chance of doing so by limited accounts.

2. Development. Unsure on how multi-threaded they are these days, but I'm assuming mostly still remains single threaded during compile and stuff or at best mildly multi-threaded. So you're looking at a fast clocker or something that just pumps out performance. You probably don't need the heavy multi-core chips like the 7950X, 7950x3D or their current gen variants. The 7800x3D and 9800x3D will likely serve you best in this regard.

3. Photo Editing. Haven't used any current modern version of stuff so can't really say much on that front. But I've never really seen anything really be heavily multi-threaded when editing unless if it's a batch run. So in most likelihood, you're again after a fast CPU and not necessarily a high multi-core one. So I would suggest of the AMD line to once again go with the 7800x3D or 9800x3D given your gaming considerations thus far also.

4. Video Editing. From my limited experience, I would say if you have memory for it already, you're looking at more the cores available on the CPU than just the speed. Although it does vary a lot also with codecs, engines, apps used, etc. But if I were to pick any CPU for this work load, it would be the 7950X, 9950X, or 9950x3D (not released yet). 7950X and 9950X only because if you can clock them higher, the performance difference shouldn't be that far off and it can brute force when your case is limited by cores, so the 7800x3D and 9800x3D will not be able to keep up when it does count (which Video content creation can do, but is limited by what is used). The 9950x3D because of the middleground again.

5. Machine Learning. It's mostly limited by your GPU, however, there are some factors that can impact on the speed of the models also; your CPU definitely has a say, depending on model being used, you may require frequent swaps from your memory or storage to the VRAM in the GPU to keep it fed; failure to do so basically causes to slow to a crawl (slowest component in chain). The RTX 3060 12GB I am using was previously testing Machine Learning / AI stuff on an i5 3570, and when loaded into the VRAM (and can all fit in there) it was OK, not great, but OK. But it will cause the CPU to go full tilt. Swapping it to the 9800x3D now, and it still fills the VRAM but when swapping the data through to it, the 9800x3D still ends up at least 50% all core use. So I would imagine that with a 5090, if you're not at least using a 9800x3D or 7950X/9950X or 9950x3D, it'll sit there waiting on your CPU a bit too much (which can be a problem if it needs to pull from long term storage like SSDs as well as the CPU is loaded). This can be especially vexing if it's your main rig as well; I do mine on the secondary rig so it can run on its own and not impact on me gaming or doing anything else on my main rig.

So given that, my (limited experience) recommendations would be:
1. 9950x3D (when released and you want middle ground and hope no core stuff handling issues like on 7950x3D)
2. 9800x3D (as most uses cases aren't multi-threaded enough to push it beyond this and especially so if you know your uses cases aren't multi-threaded enough to take advantage of more cores)
3. 9950X (If you want to clock the chip higher and also know that the stuff you do is guaranteed to use more cores over just speed)
4. 7950X (If you need an alternative to the 9950X that's cheaper and more likely available)

 

[jonneymendoza]

Personal thoughts:

1. Gaming. Performance is mostly based on your GPU. However, having used an RX580 8GB on an i7 4930K and then moving to a 7950X and getting much more out of the RX580 (in Baldur's Gate 3), as well as an RTX 3060 12GB running at less than 30fps on 4K (native and with dlss) on an i5 3570 to a 9800x3D (and going all the way to around 45FPS in 4K with everything on native and with 60+FPS using ultimate dlss 4K - all in Act 3 of BG3 so you know it's potentially lag city), I would say there's some impact from CPU performance of your GPU for gaming. I wouldn't go less than either the single 7800x3D or 9800x3D if gaming is most important. In a pinch, depending on other factors, I would consider also the 7950X, 9950X and 9950x3D (not out yet). 7950X as that's a proven brute force CPU (especially for productivity and if you clock it well, it can really go far even for gaming), the 9950X as that is the current gen version of the previous chip, and the 9950x3D as that's the supposed upgrade "middleground" between gaming and productivity in this current gen. The 7950x3D is not suggested because the 7000 series of CPU's are 100% silicon lottery based on getting any large amounts of memory going at any decent speed. Whilst the 9000 series at least with a decent board has a better chance of doing so by limited accounts.

2. Development. Unsure on how multi-threaded they are these days, but I'm assuming mostly still remains single threaded during compile and stuff or at best mildly multi-threaded. So you're looking at a fast clocker or something that just pumps out performance. You probably don't need the heavy multi-core chips like the 7950X, 7950x3D or their current gen variants. The 7800x3D and 9800x3D will likely serve you best in this regard.

3. Photo Editing. Haven't used any current modern version of stuff so can't really say much on that front. But I've never really seen anything really be heavily multi-threaded when editing unless if it's a batch run. So in most likelihood, you're again after a fast CPU and not necessarily a high multi-core one. So I would suggest of the AMD line to once again go with the 7800x3D or 9800x3D given your gaming considerations thus far also.

4. Video Editing. From my limited experience, I would say if you have memory for it already, you're looking at more the cores available on the CPU than just the speed. Although it does vary a lot also with codecs, engines, apps used, etc. But if I were to pick any CPU for this work load, it would be the 7950X, 9950X, or 9950x3D (not released yet). 7950X and 9950X only because if you can clock them higher, the performance difference shouldn't be that far off and it can brute force when your case is limited by cores, so the 7800x3D and 9800x3D will not be able to keep up when it does count (which Video content creation can do, but is limited by what is used). The 9950x3D because of the middleground again.

5. Machine Learning. It's mostly limited by your GPU, however, there are some factors that can impact on the speed of the models also; your CPU definitely has a say, depending on model being used, you may require frequent swaps from your memory or storage to the VRAM in the GPU to keep it fed; failure to do so basically causes to slow to a crawl (slowest component in chain). The RTX 3060 12GB I am using was previously testing Machine Learning / AI stuff on an i5 3570, and when loaded into the VRAM (and can all fit in there) it was OK, not great, but OK. But it will cause the CPU to go full tilt. Swapping it to the 9800x3D now, and it still fills the VRAM but when swapping the data through to it, the 9800x3D still ends up at least 50% all core use. So I would imagine that with a 5090, if you're not at least using a 9800x3D or 7950X/9950X or 9950x3D, it'll sit there waiting on your CPU a bit too much (which can be a problem if it needs to pull from long term storage like SSDs as well as the CPU is loaded). This can be especially vexing if it's your main rig as well; I do mine on the secondary rig so it can run on its own and not impact on me gaming or doing anything else on my main rig.

So given that, my (limited experience) recommendations would be:
1. 9950x3D (when released and you want middle ground and hope no core stuff handling issues like on 7950x3D)
2. 9800x3D (as most uses cases aren't multi-threaded enough to push it beyond this and especially so if you know your uses cases aren't multi-threaded enough to take advantage of more cores)
3. 9950X (If you want to clock the chip higher and also know that the stuff you do is guaranteed to use more cores over just speed)
4. 7950X (If you need an alternative to the 9950X that's cheaper and more likely available)

Thank you very much for this.

Lots to unpack and think about!

But from my initial overview, it seems as though I should wait for the 9850x3d.

One thing I forgot to add is that I also use quite a few hdd in my current Threadripper. I have 3 3.5 and 3 2.5 ssd plus 3 m.2 attached and one of the benefits of the current trx50 mobo is the huge amount of pci lanes available.

 

[Meddling-Monk]

But from my initial overview, it seems as though I should wait for the 9850x3d.

I would imagine this is your best bet, especially if you are thinking of doing the Machine Learning / AI stuff on the same rig (primary rig). As if all CPU cores are loaded, you're not doing anything else at that point. And the 9800x3D is already 50% all core loaded on a 3060 with 12GB VRAM. 32GB VRAM and 5090 is likely to saturate the 9800x3D fully, so you're not going to get much else done at the same time. So the 9950x3D is likely your better bet (need to wait on release though) if it's the same system and you can at least still do stuff at the same time (mostly at least, you can't play the latest AAA games at the same time as doing full fledged Machine Learning / AI).

One thing I forgot to add is that I also use quite a few hdd in my current Threadripper. I have 3 3.5 and 3 2.5 ssd plus 3 m.2 attached and one of the benefits of the current trx50 mobo is the huge amount of pci lanes available.

There are some motherboards out there with onboard 6x SATA ports that can handle your 3.5 and 2.5 drives along with the 3 m.2 drives without needing to share anything, you'll need to look around to see which is right for you though (some don't have decent onboard NICs for example but do everything else well; but then you can just use the WiFi for the network, so not a major deal etc, unless if your residence has issues with WiFi signals. Or other variants of weaker components that you can get around for all the drives and lanes on hand to use).