Help me choose 7900xtx or 5070ti
Posted by Ready-Pay2087@reddit | buildapc | View on Reddit | 18 comments
Hello everyone I’m in Australia and a first year engineering student, I’ve saved up enough to buy myself a workstation which will mainly be for personal use.
Here are my options:
Ultra 7 250k
32gb ddr5 ram at 6000 MT/s CL38
RX 7900xtx
1tb nvme
$2700 AUD
or
Ultra 5 245KF
32gb ddr5 ram at 6000 MT/s CL38
RTX 5070TI
1tb nvme
$2800 AUD
I do play games that require some juice but I’ve been on laptop till now so I’ve only ever really played games at 1080p (I play intensive games such as Mount and Blade Bannerlord, Beyond all reason, general grand strategy/triple A games). That being said gaming is not the primary concern, I need a rig to 3d model in, rendering is not a major concern, I’m aware that the 5070ti outperforms the 7900xtx by about 85% in cycles but I don’t render too much plus the 7900xtx does slam or match he power of a 3090 in terms of rendering performance so I’m happy with that. My main question is whether it’s worth going for the 7900xtx and a faster cpu, som other things I will be suing the rig for, Machine learning, hosting local LLMs and a lot of CAD work/simulation/electronic design. Thanks for any help.
Ughh__@reddit
If ur workflow has local LLM and machine learning then nvidia is a no brainer If gaming takes precedence over work (as in your mentioned workflows are a hobby or smth) then the amd config is better bang for ur buck.
Lil_Hater112@reddit
I had a 7900xtx and now have a 5070ti and feel like a massive upgrade to me lol
Ready-Pay2087@reddit (OP)
What’s your use case been? I’d be curious to know what type of load your machines are being put theough
Lil_Hater112@reddit
Pretty much all type of games.
In multiplayers where you dont use Framegen or maxed settings ye, 7900xtx was fine ( if it didnt crash, had random crashes with that card since i bought it)
In singleplayer, compared to 5070ti, it s shit. 5070ti can get OC+uv to 5080 stock levels from what I have seen(or close) , has DLSS which will look way better than FSR supported for 7900xtx, which will let u gain more frames for same image quality cuz u can turn it lower(even for multiplayer this can be the case).
You also get PT and RT, which is shit on 7900xtx and you also get MultiFramegen , which is like comparing iphone 17 to iphone 7 vs AMD FG.
xiscf@reddit
And now we have "Dynamic Frame Generation" which could be very interesting compared to MFG.
Lil_Hater112@reddit
ATM, I dont find it usefull, I prefer having x2/x3
kmkm2op@reddit
Replace the 245kf in the second build with a 250k plus which in multicore is almost a 265k and ur golden for a workstation.
xiscf@reddit
I understand you don’t really care about rendering, however, you should not forget about the viewports, and the animations inside the viewport.
In the viewports of blender, during an animation, you will be happy to have a rtx 5070ti.
Ready-Pay2087@reddit (OP)
How much of a difference can I expect in terms of performance for viewports and animations?
I don’t really have much use for animations as my game dev projects thus far have been 2D and I plan on keeping them low poly if I ever step into the 3d space. As for the 3D modelling work, I do a lot of table top and war gaming so I’ll mainly be making custom minis like space marines for WH40K for 3d printing, given that would the difference in performance still be that considerable?
xiscf@reddit
According to opendata.blender.org, the RTX 5070 Ti has a median score of 7,571.42, while the Radeon XTX 7900 has a median score of 4,049.85.
However, these values primarily relate to rendering performance.
If you only care about modeling performance then take the one you prefer.
It’s more about which technology will you use. CUDA is often more implemented. And if you have to use a local LLM during your studies you may want to have CUDA; it’s not mandatory though.
Ready-Pay2087@reddit (OP)
I suppose that’s the main thing I’m deciding on before making the purchase, how vast of a difference is there on the ROCm infrastructure compared to CUDA for LLMs? I’d ideally like to fit a larger model hence why the larger vram amount is useful.
That being said though I’m not really using local LLMs for anything too crazy, I’m working on a text based RPG system so whatever language model I end up picking will need to have a focus on lucidity and large context windows rather than needing to do complex problems - for those I usually just use the free versions of Claude or something - machine learning is a bit of a different story though, I’m going to be running sim pretty intensive simulations in UE5 or maybe Unity for things like robotics, I’ll also be using the rig for things like computer vision and reinforcement learning. If the performance by Nvidia is far better for these things given I will be going for an Nvidia based system.
MrFartyBottom@reddit
Why are those your two options? Go to https://au.pcpartpicker.com/ and build out a system that is better than both of those for the same money.
Ready-Pay2087@reddit (OP)
There’s currently a massive sale going on which is selling these systems are pre rampocalypse prices so I actually end up saving around $500-$600 on these system if I were to buy them together from the retailer
MrFartyBottom@reddit
Not really dude. I mean the 7900xtx might have 24GB of VRAM but it is an aging card that draw a shitload of power and neither of those CPUs are great. Go check out prebuilts at Umart, Scroptech or Mwave and see what they have for that kind of money.
Ready-Pay2087@reddit (OP)
I’ve checked out the other major retailers and Scorptec, mwave, umart don’t really have anything better.
Also why aren’t the CPUs a good option? I was under the impression that Intel CPUs were better for simulation and workstation tasks?
MrFartyBottom@reddit
They are two very different CPUs. The Ultra 7 265kf is way better if they are your needs. The Ultra 5 245KF is a very mid range chip that has just been shit on by the the Core Ultra 5 250K Plus.
Ready-Pay2087@reddit (OP)
Sweet stuff, thank you for the heads up. Just in terms of the GPUs, would the rx7900xtx still be able to perform adequately in AI/ML tasks? I know ROCm has gotten a lot better and I will be using Linux but I’m not sure how much of a difference there will be between the 7900xtx and the 5070ti, I know the vram difference for one is pretty big and LLMs have been becoming more viable on the amd infrastructure
MrFartyBottom@reddit
I am not an AI dude but my friends who are still say Nvidia all the way. If the model you are running can fit into the VRAM of your card then Nvidia will treat you better.