Might Sony go with a two-SKU strategy next-gen (like Microsoft did with Series S)?

[Bofferbrauer2]

Pemalite said:

Bofferbrauer2 said:

I don't think that the bandwidth is as much of an advantage. I mean, the 1050Ti has about twice the Bandwidth yet is only similar to the performance of the 680m in the 6800H. And while it's a bit limiting on the APU, it's also the reason why there's only 12CU, and why the Phoenix APUs coming in March are also limited to 12CU (and why AMD stuck so long with Vega 8). The 7040 series APUs will also come with 32MiB of L3 cache vs 16 on the 6000 series, which will certainly also alleviate the bottleneck to some degree.

Newer GPU's can do more with less memory bandwidth... Things like compression, culling, buffering are all a thing, so it's only natural that over time you can do more with less... Which is why IGP's can compete with low-end GPU's over time.

Also... The number of CU's is only part of the story... Remember, AMD actually started out with Vega 11 APU's in it's mobile chips, but eventually downgraded to Vega 8... Which actually offered more performance. - How? Clockspeeds, bandwidth.

Bofferbrauer2 said:

Phoenix will also have more Bandwidth than the 6000 series, which was limited to DDR5-4800 or LPDDR5 of the same speed (linking the 7735 as it's basically a 6900HS refresh, resulting in about 40GB/s). The 7040 will go up to DDR5-5600 or LPDD5-7500 (45GB/s and 60GB/s respectively), reducing the bottleneck, especially with the latter option. Additionally, the GPU clock can go up to 3000 Mhz, though probably not sustainable at that speed.

With all these improvements, I expect the performance improvement of the 780m over the 680m to be ~15-20% with DDR5 and up to 35% (limited by the TDP) for the LPDDR5 variant, which should make it compete with the 1650 or even 1060 - despite them having 2-4 times the bandwidth.

I prefer not to make assumption on hardwares capabilities when they haven't released yet.

However, keep in mind the age of the 1060 which released in 2016... It turns 7 years old this year, it's only natural that integrated graphics would eventually catch up.

I agree. But the way you worded it before sounded like "It can't get close to the performance of the Series S due to it's larger bandwidth", and I just wanted to show that even with less bandwidth, the APU is in the ballpark of GPUs with much more bandwidth to counter that statement.

7 years old and yet still one of the most widespread GPUs on Steam... That's as if the Riva TNT2, a DirectX 6 card, was still one of the most widespread GPUs when the 7900 GTX released, the last DX 9 generation.

Last edited by Bofferbrauer2 - on 09 January 2023