Funnily enough, also on GOG: https://www.gog.com/en/game/metro_2033_redux

Edit: Apparently it's to celebrate the 15 years of the release of Metro: 2033

Yeah, had this discussion in another thread on this site when someone said "Just backport it to FP16, it's simple".
But whenever you double precision you tend to halve your performance if everything else is kept equal.

7 times lower performance is a hell of a difference though, so they may be doing something else behind the curtains which comes with such a corresponding hit.

I see.

The performance loss is huge, yes, but since FSR4 is roughly one month old, I'd say that this project is still in the early stages and that there's room for improvement, even if they'll never reach parity.

Mind that there is no native FSR4 on Linux, so you need to:

a) emulate all of AMD's new proprietary WMMA instructions in the first place, which is a bit slow and looks bad at the moment even on RDNA4;

b) then use a hack to compile for yourself a custom version that supports RDNA3, since that can't be done by default with the software renderer being used (Mesa 3D) because it doesn't support these sorts of instructions yet.

Still, I'd call bull on that number, that would be slower than software ray tracing and that requires emulating much more complex BVH math instead. Unless that is referring to the total time the upscaler is taking (say, 14ms vs. 2ms) which is a more reasonable overhead for this sort of thing.

From what I've seen, it does it IRL, not by much, but that's just in turbo mode. I'm guessing next batch of handhelds will have memory bandwidth issues fixed, to a degree.

Little bit of contention on the bus that hampers that 120GB/s of bandwidth from being useful at 1080P when fillrate starts to run up into some serious bottleneck  walls.

That's where the PS4 has the advantage. Otherwise the CPU and GPU is smarter and faster in general in the Ally.

What AMD needed to do was throw in some infinity cache to make up for the bandwidth deficit.

The problem with their Z line is that they are simply rebranded mobile parts. AMD should make an unique series of chip for handheld devices with less powerful CPUs.

• Zen 5c should suffice, since the CPU isn't the limiting factor here, 6 Zen 5c cores should be a good match for handheld gaming.
• Then, couple this with a relatively big GPU. 8-16 CU would be good, more consume simply too much power with 12 RDNA4 CU probably the sweet spot right now.
• Compared to their normal mobile chips, they could cut a lot of the IO of the chip if it's only for handheld devices, which don't have much IO to begin with.
• Finally, add some infinity Last Level Cache addressable from both the CPU and GPU (making it shared memory) to widen the bandwidth, and they'd have a great handheld chip, powerful but relatively easy on the battery without bells and whistles that they don't have any need for.

NVidia 5060/5060Ti announcement videos (and condemnation of NVidia's shady practices for their 8GB cards):

Gamer's Nexus:

Hardware Unboxed:

Pricing and launch day:

• 5060Ti 16GB: $429, April 16
• 5060Ti 8GB: $379, April 16 in theory, in practice potentially later and isn't sampled to board partners
• 5060: $299 Some time in May (no launch day and review embargo end tomorrow with the other cards, so ready to get shadow-dropped)

5060Ti 16GB is supposed to be 20% faster than the old 4060Ti (not precised by NVidia if they meant the 8GB or 16GB 4060Ti). As in the HU chart the 4060Ti has an average of 70FPS and the 5060Ti is supposedly 20% faster, that would give the 5060Ti an average of 84FPS, short of the 4070 with an average of 90FPS and just enough to beat the 7700XT.

The 5060 meanwhile is supposed to be 20-25% faster than the 4060, which would be enough to beat the 4060Ti by a hair... but it's 8GB framebuffer will hold it back a lot, so only so long it lasts...

Oh, and NVidia trying their darnest to hide the 8GB version of the 5060Ti and the 5060 from benchmarks really shows they know it's a waste of sand, but one they'll sell well to unknowing consumers.

AMD’s 5th Gen EPYC Server CPUs Witnesses Adoption By Google’s New AI Servers; Delivering Up To 80% Higher Generational Performance

https://wccftech.com/amd-5th-gen-epyc-server-cpus-witnesses-adoption-by-google-new-ai-servers/

AMD Confirms Next-Gen EPYC Venice “Zen 6” CPUs Are The First HPC Product Made Using TSMC’s 2nm “N2” Process, 5th Gen EPYC Validated At TSMC Arizona

https://wccftech.com/amd-confirms-next-gen-epyc-venice-zen-6-cpus-first-hpc-product-tsmc-2nm-n2-process-5th-gen-epyc-tsmc-arizona/

NVIDIA To Produce AI Servers Worth Half a Trillion Dollars In The US; Blackwell Production Already Underway At TSMC Arizona

https://wccftech.com/nvidia-to-produce-ai-servers-worth-half-a-trillion-dollars-in-the-us/

The only good thing about the 5060 Ti is the 16GB version is $70 less than last gen. It's a shame the only time 60 series saw 12GB of vram was with the 3060 and since then, they have been stuck with 8GB. Nvidia could have given 5060 3GB chips to add in more vram as well since they have access to high density GDDR7 chips but alas.

Hopefully Radeon doesn't goof their pricing up with the 60 class like last time. Launching 7600 with 8GB of vram while being slower than 4060 for a $30 discount was laughable. Hopefully they pull off another 9070 XT type pricing with the 7600.