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

Status
Not open for further replies.
I do not know any current SSD capable of 4.8 GB/s. Samsung 970 EVO and Intel Optane 905p cannot reach that value (3.2 GB/s is the best max on 970 Evo).
What other do you know?
Pretty much every PCIe 4.0 SSD out there, the ones you mentioned are PCIe 3.0.

What about the RX 480/580? Any idea the amount of L2 they have? I want to know if the amount of L2 in Scorpio was something common to Polaris. I know Scorpio more closely resembles the X1 in its makeup but they did include some features found in Polaris. Curious if the L2 amount more closely conforms to Polaris or Hawaii.
2MB L2 on Polaris 10/20/30

4X16 = 64
RX 5700 XT has 2 SEs, 4 SAs, 64 ROPs
If PS5 and XSX have 4 SEs like you said, they should have 8 SAs and thus 128 ROPs.
 
What about the RX 480/580? Any idea the amount of L2 they have? I want to know if the amount of L2 in Scorpio was something common to Polaris. I know Scorpio more closely resembles the X1 in its makeup but they did include some features found in Polaris. Curious if the L2 amount more closely conforms to Polaris or Hawaii.
Polaris 10 featured 2MB as well (256-bit bus). TBH, it probably had more to do with the node density, as Hawaii was on 28nm, so naturally 2x density on 14/16nm (Polaris/Scorpio) allowed them to shove more on there.

It's a bit of a weird setup for Scorpio since it features a 384-bit bus whereas Hawaii was 512-bit, so there's a non-pow2 oddity between L2 slices and the number of memory channels. Should have been 3MB. :devilish:

The delta compression for the ROPs was the only other major feature added I think.
 
If PS5 and XSX have 4 SEs like you said, they should have 8 SAs and thus 128 ROPs.

4SAs nothing changed since 5700XT it seems.

I.e. 64 ROPS, 4 rasterizers, 4 primitive units, 4 L1 pieces (probably the same size), 1 Geometry unit.
For both PS5 and XBSX.
 
Clearly, they want people to know about their TF, sram and VRS. They picked a few points to focus on, and there has to be a reason! Same with Sony and their audio and ludicrous speed SSD.
Cerny also bragged about having lots of esram.. in the I/O block.
 
I hated lvl 19 of Manic Miner. That damn level haunts my 1980s childhood to this day. :yep2:
true I think it was the hardest, level 20 was pretty easy IIRC
I wonder if that game had anything to do with making platformers my least favorite game genre?
 
We do? How many can present 13.1 Tflops and still spare power for that

I'm curious where the 13.1 Tflops figure has come from. Is this something Cerny has said? I know he did say that their decompression tech is the equivalent of 13 of their (3.5Ghz) Zen2 cores which equates to only 1.5 TF of raw floating point throughput. Still a significant amount of course.

That's not to say I think decompression on the GPU is possible or desirable in software. But if that compressed data (or perhaps just a significant portion of it) isn't needed in an uncompressed form until it's in graphics memory then it should be a pretty straight forward matter to put dedicated decompression hardware on the GPU. It has significant precedent with video decompression which is arguably far more niche given a GPU's intended market. Obviously if the data needs to be decompressed at the CPU stage though then that whole idea can be binned.

I do not know any current SSD capable of 4.8 GB/s. Samsung 970 EVO and Intel Optane 905p cannot reach that value (3.2 GB/s is the best max on 970 Evo).
What other do you know?

There are several here all based on the current industry leading Phison PS5016-E16 controller:

https://www.scan.co.uk/shop/computer-hardware/solid-state-drives/m2-pcie-40-nvme-ssds

They're expensive now but within a few months these will be supplanted by the next gen drives based on the upcoming PS5018-E18 controller capable of 7GB/s so should drop considerably in price. Lexar is the fist vendor to start promoting it's upcoming 7GB/s drives:

IMG_20200108_112747-1024x614.jpg
 
Le'ts hope the new storage benefits translate on PC. Right now there's no difference in loading speeds between PCI 3.0 and PCI 4.0 drives on the PC. Effectively useless for gaming.
 

Nice... 250 euros is a lot though.
I'm curious where the 13.1 Tflops figure has come from. Is this something Cerny has said? I know he did say that their decompression tech is the equivalent of 13 of their (3.5Ghz) Zen2 cores which equates to only 1.5 TF of raw floating point throughput. Still a significant amount of course.

Typo... I meant 12.1, the Xbox GPU Tflops.
 
Last edited:
IIRC DirectStorage is coming to Windows as part of DirectX.

It is, and from my limited understanding it's designed directly to address the legacy I/O protocols used in PC's which result in faster drives being bottlenecked today.

I'm particularly interested in whether Microsoft intend to bring the other major aspect of their velocity architecture to PC via DX as well, I.e. establishing direct comms between the GPU and SSD to allow it to act as additional graphics memory.

AMD already demonstrated the technology years ago in their Pro cards, but will it now be brought into the consumer space via RDNA2 within the framework of DirectStorage? And is NVIDIA in a position to support it if they do?
 
Was it not required 7 GB/s to compensate for only 2 priority channels?
I don't think so, that was just speculation that maybe it needs more than 5.5 to compensate variations in controller implementations, we have no idea how much.

And the storage expansion doesn't need that many priority channels if those related to the OS are only present on the primary drive. The second drive is used exclusively for the game.
 
Was it not required 7 GB/s to compensate for only 2 priority channels?

Edit: Yes it was 7 GB/s in the "Road to PS5 video" at 21:45

I think he said it needed to be 'a little bit faster' than the 5.5 GB/s offered in the PS5's native drive. 7GB was on the screen behind him but I think that was in the context of the new 7GB/s drives coming to the PC market this year. So yes you will need one of those because the current 5.0GB/s drives aren't quite fast enough, but that doesn't mean you'll need the full 7GB/s to equal the PS5's base performance.
 
I took the 8-9GB/s figure to be a simple calculation of the raw SSD speed * the ZLIB/Kraken compression ratio which given he stated Kraken to be about 10% better than zlib gives you a ratio between 150% - 165%. Or exactly 8.25GB/s - 9.1GB/s.
This makes sense. But does that mean the compression will never go above 165%? As in never ever?
 
Status
Not open for further replies.
Back
Top