By using this site, you agree to our Privacy Policy and our Terms of Use. Close
Pemalite on 26 September 2025
Edit Thread
Darc Requiem said:
Bofferbrauer2 said:

There are other and better alternatives than this. For doubling the bandwidth you'd need to thicker boards made of more layers, which strongly increases their price. It's one of the reasons why entry-level GPUs come with such small bandwidths

  • HBM: Would eliminate the bottleneck but at a hefty price, so probably not worth it
  • GDDR7: Would be very fast, but capacity could be a concern.
  • LPDDR5X: Samsung is now producing LPDDR5X-9600 memory, which would strongly increase the bandwidth over the LPDDR5 memory in the Steam Deck with minimal changes
  • DDR6: Expected in 2027, it increases bandwidths considerably as it comes with speeds ranging from DDR6-8800 to DDR6-17600. So if there's no successor until then this could be a very viable option.

So yeah, there are a lot of alternatives, some more sensible than expanding the bus to 256 bit.

The only one of those memory types that would see use in a PC handheld is LPDDR5X. The other memory types you listed are a complete non starter. Pemalite's suggestion of using larger cache sizes to compensate for the low memory bandwidth seems to be the best option. Still, I worry about the growing gulf in bandwith between PC handhelds and even low end dGPUs. 

What I really was going to post about is the $1000 Xbox Rog Ally X price. Remember when Phil Spencer said this...

https://wccftech.com/microsoft-gaming-ceo-gives-subtle-dig-at-ps5-pro-price-were-not-growing-the-market-with-1000-consoles/

The guy is a walking contradication. Mattrick may started Xbox's decline but Phil Spencer was the one that killed the brand.

Handhelds use integrated graphics, there has always been that gulf in memory bandwidth between integrated graphics and low-end discreet... But we also need to remember they aren't chasing 1440P or 4k resolutions so the bandwidth and fillrate is far less important relative to other parts... They are best for 720P-900P.

Strix Halo mixed things up a bit though with bandwidth up to 256GB/s and 32MB of cache, but it still drops behind the 9060XT... And it's a stupidly expensive and large chip that consumes way to much power and produces far to much heat for a handheld.

AMD could optimize Strix Halo though by cutting the CPU cores, GPU CU's, memory bus and boosting the cache. Doubt they will though.

Bofferbrauer2 said:

DDR6 too, at least as a bridgegap between LPDDR5X and LPDDR6. After all, DDR6 should consume as little as LPDDR5, so somewhat less than LPDDR5X.

And of course the extra cache as Pemalite suggested could go a very long way in terms of staving off Bandwidth limitations. AMD already proved this with RDNA2/3 and their Gamecache, allowing for similar bandwidth as NVidia on older and slower memory technology. Now if AMD would go ad put 3D Vcache over the GPU part instead of the CPU part (which, let's face it, is totally useless in a handheld where the GPU is limiting either way - unless AMD wants to use it as a way to lower the clock speeds to lower power draw without losing performance), this could be very useful to avoid going to the memory altogether in many cases.

I mean... AMD has proven to be competitive with RDNA4 with slower and cheaper GDDR6 verses nVidia with GDDR7.
The 9060XT handles itself very well against the 5060Ti despite having 128GB/s (Almost a 3rd!) less bandwidth than the 5060Ti.

And it's mostly down to the cache, culling and compression.

Caches are the answer if DRAM can't keep up.




www.youtube.com/@Pemalite