How Much CAN Nintendo Charge for Switch Pro

[Bofferbrauer2]

Pemalite said:

Bofferbrauer2 said:

I know Flops are irrelevant, hence why I said Navi with similar performance, not similar flops. But Since both the PS4 and Vega are GCN, comparing their performance by TFlops was a legit way to get them to the same level more or less.

Nope. Vega/Graphics Core Next 5 is a more efficient GPU than the Graphics Core Next 1.0 derived Graphics processors in the base Playstation 4 and Xbox One.
Thus making Tflops irrelevant.

Just because they are "Graphics Core Next" doesn't mean they have the same performance per clock, AMD actually did allot of engineering work updating Graphics Core Next over the years.

Bofferbrauer2 said:

Dropping the clock speeds doesn't cut it. I'll have to check if I can find a test in English, but the 3700 U can't hold the GPU speed at all and drops to about 800 Mhz under full load (CPU at it's base clock of 2.3Ghz). At that point, it can reach at best 2/3rd of the PS4's performance. The 35W 3750H also can't hold the clock speed, but at least only drops to 1150Mhz under fullload (with same CPU clock).o this yet.

I have a 2700u Ryzen mobile chip in my notebook and done extensive clock/TDP/voltage scaling testing... It's basically the 14nm version of the 12nm 3700u, so not much difference.

You can hold the Graphics Processors Clock if you reduce your Central Processors clock... In games that aren't CPU intensive you can actually see some rather dramatic increases in performance. - For most titles I actually limit my CPU clock to 70% of it's base clock, not allowing it to turbo... Good thing Ryzen has decent IPC.

It's tied to TDP, once you hit your TDP limit on both the CPU and GPU, then the clocks between the two get "leveled out" so to speak.

But the clockspeed of the Graphics Processor isn't what is holding back AMD's APU's. - It's actually memory bandwidth, DDR4 2400mhz on a 128bit memory bus isn't doing anyone any favors in regards to performance, especially once allot of bandwidth hungry alpha effects start getting thrown around.

You don't need the same flops or clockrates or CU's to match the base Playstation 4 in terms of performance... Or even memory bandwidth for that matter, GPU's have come a long way since 2013 on the efficiency front making such comparisons directly useless, even if they are all Graphics Core Next. (Which I would argue Navi falls under Graphics Core Next anyway.)

May I ask what Laptop you got? I've been looking for a good one with an APU for quite some time now...

Don't forget that Laptops dissipate lots of heat through the keyboard, which a handheld doesn't have, and are bigger in general. You really need to bring the TDP down to about 5W for a handheld device - and not just for heat, also for battery life reasons. Also, don't forget that you possibly just got a very good sample from the center of the die and your experience doesn't necessarily translate to it being the case for all the chips in the series.

I know Vega is more efficient, but not very much more performant. For instance, the performance difference between GCN3 and GCN4 at same CU and clock speed boils down to just 4%, and that comes entirely from increasing the L2 cache from 512kb to 2MB. The increases from GCN2 to 3 boil down to a change in the Instruction set. All in all, from GCN1 to GCN5, the uptick in performance per clock and CU is below 20% unless you use half precision, and most of it was just getting it's power that it already had to the ground. It just gained lots in efficiency along the way - which AMD then threw away again by clocking their chips so high to at least keep up somewhat with NVidia in the mid-range.

I do agree with you that GCN isn't terribly well suited for gaming. But for work, it seems to be better suited than RDNA. At least the upcoming AMD Arcturus professional cards seem to be based on Vega, not Navi.