Current Generation Hardware Speculation with a Technical Spin [post GDC 2020] [XBSX, PS5]

Status
Not open for further replies.
http://industrial.adata.com/en/technology/74
What's the logic in having tiered storage with faster SLC but then you can only use it for writes? Your claims don't make much sense to me.

That's really talking about setting aside some TLC flash to use as SLC.

"SLC cache is the primary method adopted in speeding up read/write of TLC SSD. The name "SLC cache" implies that it is not true SLC NAND Flash. SLC NAND Flash outperforms MLC and TLC in read/write speed because its cells can save 1 bit of data. SLC cache employed by TLC SSD changes the I/O speed of SSDs by reserving some space in the TLC NAND Flash and saving only 1 bit of data in each cell."

So you'd normally only read from this SLC cache (actually made from TLC) if it hadn't been been copied out and stored in a region of TLC. I don't see that you'd copy from TLC, to SLC cache (actually TLC), then to system memory. That would probably be slower than just reading from the TLC region in most cases, at least in terms of latency rather than throughput.

My hope is that the regions of SSD used as "virtual ram" in the XSX are parts of a QLC drive that's been set to SLC mode. This should minimise cost of the drives, but maximise endurance. I hope Sony do something like this too, actually.
 
With mesh shaders happening in addition to fast SSDs and good CPUs I'm looking forward to what devs will come up with if they are allowed to be creative. Definitely the console generation I'm looking forward the most since the PS2.

[...] 10 to 11 Turing seems to match 12.1 RDNA2.

RDNA1 and Turing seem to perform almost the same with exactly the same TFLOPs. Computerbase did a test where they normalized the gpu and memory clocks of the RX 5700 and RTX 2700 (so they have the same TFLOPs and bandwidth), which resulted in equal performance for both.

https://translate.google.com/transl...t-test/4/#abschnitt_amd_navi_vs_nvidia_turing

In case RDNA2 has some perf improvements without the need to use of new features like mesh shaders or variable rate shading then the XSX should sit comfortably between 2080 and 2080ti for unoptimized PC ports.
 
RTX 2080 TI is 10 billion rays per second.

Wow, the XSX will be how many times faster, say 38 times a 2080Ti.

XSX should sit comfortably between 2080 and 2080ti for unoptimized PC ports.

Maybe. Still, il go by what i know now, 2080 level performance, aside from optimisations, as games arent really optimised for modern architectures, untill consoles arrive, and DF could do some tests with modern games.
 
Maybe. Still, il go by what i know now, 2080 level performance, aside from optimisations, as games arent really optimised for modern architectures, untill consoles arrive, and DF could do some tests with modern games.
But your not doing that, you where given a apples to apples RDNA1 to Turning data point, which directly disproves your position of ~10 perf per flop advantage to turning. So the consoles have more flops then a 2080Ti but less bandwidth, more bandwidth and more flops then a 2080 and also methods to better improve I/O to memory performance ( saving bandwidth duh! :) ) . So really inbetween a 2080 and a Ti is a conservative position given known facts.
 
Can the Xbox Series X decompression block be used directly by the developers, or is the 1.2 GB/s headroom (6 - 4.8) only there if Microsoft ever increase the SSD speed?
 
So the consoles have more flops then a 2080Ti but less bandwidth

Do they? A RTX2080Ti is a 13.5TF GPU as far i know. The PS5 might be overshadowing it with it's rumoured 15TF GPU, but the XSX doesn't.
They took a game not optimized for new hardware, i guess gears 5 is optimized for the One and ported/enhanced for One X and PC's. I think a 2080 seems to be about what's in there in raw performance. Optimisations and low level access could lift it to between 2080 and 2080Ti. Right now its 2080 as per DF, we will see how much that lifts compared to that GPU.
More intresting is where it lands compared to RTX3000 series, we should have gotten more info but corona and all.
 
Maybe. Still, il go by what i know now, 2080 level performance, aside from optimisations, as games arent really optimised for modern architectures, untill consoles arrive, and DF could do some tests with modern games.
Assuming they've scaled everything else with the FLOPS it really should be between 2080S and 2080 Ti even if RDNA2 would offer same performance per FLOP as RDNA1 (naive extrapolation from Navi10/5700XT performance)
 
Assuming they've scaled everything else with the FLOPS it really should be between 2080S and 2080 Ti even if RDNA2 would offer same performance per FLOP as RDNA1 (naive extrapolation from Navi10/5700XT performance)

Agree, where the XSX lands yes, on paper specs, 2080/2080Ti between that. Not bad.
 
As long as AMD has not decoupled INT and Float the XSX is not better than a RTX2080.

Right now it's on par with a 2080. Future tests will show more. I think with a RTX2080, you wont be far off, and are able to play next gen games, everything else being equal offcourse.
 
Right now it's on par with a 2080. Future tests will show more. I think with a RTX2080, you wont be far off, and are able to play next gen games, everything else being equal offcourse.

"Future tests"? Why? Only cases i can see is with Mesh-Shading and better GPU sheduling through low level API. Raytracing? Doesnt look like. ML? Nope.
 
Direct Storage + Decryption + Virtualized Memory + Instant Resume + 3D Audio

Probably going to be a while before PCs get that or it's used as much as it will be on consoles.
Well , all related to the ssd (plus 3d audio, didnt read that part). Shifty will be dissapointed without the BVH traversal block and only the intersection hardware.
 
but we do know... assuming an industry standard 10 deep BVH, the 2080Ti can, at 2ghz, in theory do 13.6 gigarays/sec (RT core count , 68 in the 2080Ti's case) times clockspeed , divided by bvh depth) gets you your theoretical peak RT performance, nvidia quotes 10, but they also quote a lower boost clock. for xbox series X, 1825mhz times 208 RT cores gets you the 380 billion/second number MS quoted, divided by the 10 deep BVH gets you ~38 (37.96 to be exact) gigarays/sec, or ~4x that of the 2080Ti. Devs can feel free to chime in on this, but the math is correct on paper. In practice the difference is even bigger because nvidia's design is literally insane, an RT op is warp-wide, so the entire SM is basically idle while RT is being done on a turing RT core, plus it has no latency hiding, which i'd happily go into detail about if anyone actually cares. But no, RT performance on this is not freakin below a 2060, its ~4x that of a 2080Ti minimum
 
Well , all related to the ssd (plus 3d audio, didnt read that part). Shifty will be dissapointed without the BVH traversal block and only the intersection hardware.

Seems so far that PC won't be left out much.


Can't wait to see real RDNA2, full fat GPU's, not just 38times faster then 2080Ti, but way more.
 
Where is the XSX special sauce? Except for maybe ssd data decompresion block all seems standard AMD RDNA 2 tech.
In terms of other customizations, may have to wait to hear.
Example they never mentioned their VRS implementation.
So maybe there's more, maybe this is it.
Also it might be small sauce sachets, than big bottles.
If that's the case, may not hear until next year gdc or hotchips
 
Status
Not open for further replies.
Back
Top