Im planning my new config and one of the primary issue is that Im unsure whether 7950x3d is worth the hassle. If I went with 7950x, I would surely want to fiddle with overclocking.
As far as Im aware, 7950x3d can be overclocked, but only the one without the 3d cache. How reliable is this?
Also, I suspect that the TDP difference is a non-issue, as it is mostly just a limitation of the 7950x3d thermals rather than actually better power consumption.
Also the L3 cache is 64mb for the 7950x vs 128mb for 7950x3d. How much does that matter?
Things I want to do: gaming, software development, virtual machines, VR, AI related work. Im asking here since I mainly want to know about the status on linux.
The cache matters a lot. That’s basically why the x3d chips are so great at gaming. With over clocking you’ll always get a lower clock than the stock boost clock with these newer chips. That’s why PBO (precision boost overdrive) in combination with CO (curve optimiser) will give the most performance. Both chips can do this.
Tl;Dr: get the x3d.
Shit. Forgot the price. What if I can get 7950x for like ~500 and the 7950x3d for like ~700?
I’d go for the 7800x3d if gaming is your priority. If you need the cores the 7950x would be the economical option.
Since you mention ‘gaming’ first, I’d say go for the X3D. Overclocking Zen3+ CPUs is hardly worth the effort anyway, you’ll hit the point of diminishing returns real fast and hit thermal limits. Undervolting might give you better results though.
The only downside is some games are bad when they run on the wrong cores so you might need to manually sign them
CS:GO needs the fast frequencies, other ones like more cache
When it comes to gaming performances the 7800X3D actually outperforms the 7950X3D.
How I see it, its more that if you use multi CCD cpus, you have to manage the CPU manually as some games prefer cache, some games prefer clockspeed, and the OS picks the wrong one at times (usually the clock speed) as the OS assumes the higher clock speed = faster. As the thread director works correctly with Intel chips for the most part (where the e cores have a lower clock) so the e cores are often not selected for performance.
The solution AMD will have to transition to is to exchange the other CCD for the core count focused design (e.g Zen 4 vs Zen 4c) which would be clocked lower to receive similar benefits from the current itteration of thread director.
Just fyi there’s a bug currently preventing newer AMD CPUs from falling into low power states. May require manual intervention (i.e. turning off certain c-states) on some distros. Not sure how concerned you are with power efficiency, but something to consider.
Lol I would like to heat my room with it.