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

Status
Not open for further replies.
I see. That's more reasonable and indeed possible, and I think the touch bar MacBooks kinda already do that with the touch bar. But that maybe more restrictive and not worth the cost.

Currently at least on Xbox One, the games' and apps' CPU quota changes based on whether the game is in foreground, and background services like friends already take very few cycles. GameDVR on the other hand, I guess there's a dedicated hardware encoder somewhere.

Right now your wasting resources for a fancy UI and your ear marking resources for video recording , voip and other features that we have come to expect from a console since the dreamcast / xbox 360. A second apu running arm would allow you to off load all that and free up the resources for the actual games. You'd also sip power when watching netflix or doing non game related tasks in the os. MS has windows running really well on arm and collaborated on the sq1 and 2 apus so they can modify a gpu as they want. They added a lot of ML stuff to it already. I would wager the only question is how much ram you'd need to run the os and stuff you want to unload. 2gigs ? 4 gigs. So what would be the price of the apu and ram and is it worth the down side. Although again if they are going to do a streaming stick you could have had the same arm apu in both products bringing the cost down for both and you'd already have had to have done all the work
 
Right now your wasting resources for a fancy UI and your ear marking resources for video recording , voip and other features that we have come to expect from a console since the dreamcast / xbox 360. A second apu running arm would allow you to off load all that and free up the resources for the actual games. You'd also sip power when watching netflix or doing non game related tasks in the os. MS has windows running really well on arm and collaborated on the sq1 and 2 apus so they can modify a gpu as they want. They added a lot of ML stuff to it already. I would wager the only question is how much ram you'd need to run the os and stuff you want to unload. 2gigs ? 4 gigs. So what would be the price of the apu and ram and is it worth the down side. Although again if they are going to do a streaming stick you could have had the same arm apu in both products bringing the cost down for both and you'd already have had to have done all the work
You might as well go one step further and remove the fancy dashboard completely. Relinquish every secondary function like voice chat to your phone, I’m sure everyone has got one these days, and there’s usually more than one high performance ARM cores inside one. Then you can have all the CPU POWER for the game! Now that’s peak performance, not some half-ass hybrid mess.
 
Quick question for you folks. We've seen speculation regarding AMD's new cache setup and while looking at the Series X, I noticed that it was 320 bit.... The 6800xt is 256 bit so why did Microsoft choose to go with a 320bit bus? This has got me thinking about whether or not the Series X soc uses the new cache architecture by AMD. The 6800xt with more CU's requires less bandwidth than the series X and it's 560 gb/s of bandwidth(10GB GDDR6)? Am i off base or does something seem amiss in their design?
 
According to die sizes, neither nextgen consoles have Infinity Cache architectures.
 
Interesting....Then can you fill me in why Cerny was emphasizing Cache Coherency and stale data in his talk? I don't want to start a vs war here. I'm just interested as to why (potentially)neither manufacturer decided to implement that into their Soc design? From first glimpse, it appears to be an architectural design with significant gains.
 
Interesting....Then can you fill me in why Cerny was emphasizing Cache Coherency and stale data in his talk?

It's related to the speed at which the SSD can stream data into the VRAM vs the old HDD's. There's much more opportunity for cache coherency issues when you're overwriting large chunks of VRAM on a constant basis.

I'm just interested as to why (potentially)neither manufacturer decided to implement that into their Soc design? From first glimpse, it appears to be an architectural design with significant gains.

My guess is that it would have made the dies too large and expensive. And possibly that there was no guarantee the tech would be ready in time for the consoles launch.
 
Quick question for you folks. We've seen speculation regarding AMD's new cache setup and while looking at the Series X, I noticed that it was 320 bit.... The 6800xt is 256 bit so why did Microsoft choose to go with a 320bit bus? This has got me thinking about whether or not the Series X soc uses the new cache architecture by AMD. The 6800xt with more CU's requires less bandwidth than the series X and it's 560 gb/s of bandwidth(10GB GDDR6)? Am i off base or does something seem amiss in their design?
larger bus width means more bandwidth. And usually when you have more compute, you're going to need to have more bandwidth. how you accomplish that, could showcase the differences as to why Series X may have gone with 1 design and 6800XT with another.

But we will need to see the final release of 6800XT to get a better idea of what's going on there.
Ultimately design goals are to keep costs and power down for consoles and neither of this may apply to the 6800XT
 
Interesting....Then can you fill me in why Cerny was emphasizing Cache Coherency and stale data in his talk? I don't want to start a vs war here. I'm just interested as to why (potentially)neither manufacturer decided to implement that into their Soc design? From first glimpse, it appears to be an architectural design with significant gains.
The coherency engines and cache scrubbers seem to be intended to reduce performance loss due to cache flushes resulting from the GPU reading from locations that have been overwritten by an SSD read.
How that fits into the functionality isn't clear, in terms of whether it reduces the cost of flushes, or somehow avoids some of these stalls.
The exact usage model hasn't been outlined. There may be certain allocation strategies or phases in the setup or cleanup of a frame that are sped up, or it can enable more rapid use of SSD data.
However, there are likely different methods that do not see significant differences or questions as to how reliable the SSD can be in timing to lean heavily on the scrubbers.

PC parts likely have a problem where they cannot reliably assume there's an SSD present, or the games cannot assume it. Scrubbers matter less if the disk and overall IO subsystem are too long latency to be as aggressive as Sony seems to be, and RAM capacity tends to scale more readily for the PC.
It's possible other vendors don't see a significant enough win, or they have other priorities considered more important like greater compute capability or additional features in other areas. If there's a certain usage model Cerny has in mind, it may be too narrow a field for the other vendors (example: hardware checkerboarding was pretty quickly skipped over by many games)

Cerny seems to care more about certain latency events in frame processing or more serial sections versus raw throughput, and since the PS5 invests more in its SSD subsystem it may be counting on more aggressive leveraging of low-latency storage. However, how aggressive it can be since SSDs still vary a fair amount in latency (and 3rd party drives can vary a lot) is up in the air.
 
And possibly that there was no guarantee the tech would be ready in time for the consoles launch.

Yes, likely there's a difference of at least 9 months, possibly more, on getting the "Infinity Cache Architecture" into desktop AMD GPUs versus when next-gen SOCs were finalized.
 
I dont think it much matters with reliability. Though it would be cheaper to repair.

From original manufacture perspective, one may be a bit cheaper in parts (no slot and mount kit) but then it may be costlier to manufacture (more items to solder in and larger pcb). I really don't know the costs of these tradeoffs.

Not just cheaper to repair, but if you want to release a model with more storage, you just plop in a larger SSD module versus having to make a batch with a soldered on SSD.

This is even more relevant if the console maker has 2 models on the market with different storage capacities. A 1 TB model and say a 2 TB model. It doesn't matter what market demand is like, the manufacturer is still using the same mainboard with a slotted SSD versus a soldered on on SSD where the manufacturer is stuck with whatever storage amount comes with that mainboard.

It hasn't happened yet, but it wouldn't surprise me if, once the launch period is over, MS decides to release a higher capacity SKU.

Regards,
SB
 
Last edited:
The 6800xt is 256 bit so why did Microsoft choose to go with a 320bit bus? This has got me thinking about whether or not the Series X soc uses the new cache architecture by AMD.
The SeriesX isn't using infinity cache for sure because they've shown the die's x-ray and there's a slide with the chip's total SRAM breakdown with no sign of any GPU LLC.
For the PS5 it's still unknown. Cerny did mention a "generous amount of SRAM" that the I/O complex would have access to, but the diagrams shown suggest it was for exclusive use of that block.
 
The SeriesX isn't using infinity cache for sure because they've shown the die's x-ray and there's a slide with the chip's total SRAM breakdown with no sign of any GPU LLC.
For the PS5 it's still unknown. Cerny did mention a "generous amount of SRAM" that the I/O complex would have access to, but the diagrams shown suggest it was for exclusive use of that block.
weird, isn't today RDNA 2 day?

Why isn't the rest of the threads going nuts yet? It's nearly end of day, you should be sleeping by now lol and I haven't seen jack on it.
 
I remember the days we had a huge thread about the advantages of esram on xbox one. That small 32MB that pushed upwards to 192GB/s. So is this infinity cache basically just another esram only larger? I was under the impression that HBM was the true successsor to esram due to it's size/cost advantages while still being able to pump out large bandwidth numbers.
 
I remember the days we had a huge thread about the advantages of esram on xbox one. That small 32MB that pushed upwards to 192GB/s. So is this infinity cache basically just another esram only larger? I was under the impression that HBM was the true successsor to esram due to it's size/cost advantages while still being able to pump out large bandwidth numbers.
In principle they do sound similar. But X1's eSRAM was more of a scratchpad whereas Infinity Cache is probably... a cache?
 
In principle they do sound similar. But X1's eSRAM was more of a scratchpad whereas Infinity Cache is probably... a cache?
Ye I guess in principle. But when you’re cache is 128MB it’s not entirely clear how it works in conjunction with L2 and L1.
 
Status
Not open for further replies.
Back
Top