Next Generation Hardware Speculation with a Technical Spin [post E3 2019, pre GDC 2020] [XBSX, PS5]

Status
Not open for further replies.
Tests without knowing how the hardware was setup aren't very informative either. Trying to infer information from an unknown setup = almost pure guesswork. Only slightly more informative than just making something up. It's why virtually any reputable site that does performance benchmarks not only provides details on how the hardware was setup and what hardware was used, but also explains what tests will be run and why those tests were used.

That said, pulling random benchmarks that fit what someone believes conforms to the performance they believe is representative of X thing and then going through those results and looking for some hardware configuration (of the little that is provided) that fits their concept of what the hardware should look like does make for interesting commentary though. :) Just as long as people understand that the information that can be gleaned from it is like throwing a dart at a dartboard with your eyes closed. You may derive the right information, but if you do, it's mostly just luck.

Not saying it's wrong to do that considering how little is actually known about the next generation of consoles, but people should understand that whatever can be gleaned may or may not be relevant in any way. But it does for the most part devolve down to confirmation bias (look for something that fits what you believe and discard anything else) and then hoping that it matches the actual product.

Regards,
SB
 
I wonder if HBM is still a possibility...

The high cost is more of an offer/demand imbalance, and low volume, than a production cost issue now. It has to rebalance eventually. It does cost more to make, but 30% isn't much compared to the 2.5x we heard about a year ago. And a console contract might be a convincing argument to significantly increase HBM production and maybe more competition to get that contract?

https://semiengineering.com/hbm2e-the-e-stands-for-evolutionary/
Jim Handy, principal analyst with Objective Research, estimates the cost of a DRAM wafer at $1,600. HBM2 adds another $500 to that cost, a 30% premium. “DRAM makers would charge a lot more than 30% for it,” he said. “I expect [HBM] will remain in the GPU space mostly, because it’s expensive. If the market gets big enough, then production will get cheaper and open the doors to new apps.

The price has remained high because there has not been a widespread adoption by companies to work on a new cost structure or increase supplies.
 
I guess 8GB HBM + 16GB DDR4 could make some sense, for cost reasons. But I would think it's still better to have one pool.

The only reason to go with hbm would be if the higher price compared to gddr6 is offset by the slightly lower cooling/psu costs. The power consumption or footprint advantages are not really providing anything important for a home console, and there isn't any performance gains with hbm unless they need more than 768GB/s, which they don't.
 
The Playstation 4 family's memory structure is a bit of an odd one, considering games only have access to 4.5 GB GDDR5 on Ps4 and 5GB GDDR5 on the 4Pro. I guess they could expand the PS5 alternate DDR4 memory pool enough to fit the entire OS there instead of only using it as a cache and swap scratch-pad.
 
I think the whole theory of having a slower DDR4 or LPDDR4 memory pool for OS use is a total non-starter. It's like asking for the OS to be laggy and unresponsive. I think of it a bit like buying an SSD to install games on and installing Windows on a mechanical hard drive so you don't have to sacrifice SSD space for the OS. In a system with an SSD it probably doesn't even make sense for tombstoning apps anymore.
 
I think the whole theory of having a slower DDR4 or LPDDR4 memory pool for OS use is a total non-starter. It's like asking for the OS to be laggy and unresponsive. I think of it a bit like buying an SSD to install games on and installing Windows on a mechanical hard drive so you don't have to sacrifice SSD space for the OS. In a system with an SSD it probably doesn't even make sense for tombstoning apps anymore.

lpddr4 memory makes sense if one assumes next gen consoles keep super low power arm cores around to do background tasks like download/installs/... That is unless amd has such a good x86 chip that arm doesn't give good ROI anymore.

Do we have any rumors on next gen consoles ditching arm cores for low power/background tasks?
 
Do we have any evidence that current-gen consoles actually use the arm cores for low power or background tasks?
 
Is there any combo of GDDR and HBM that makes clear sense?
Not really IMO. HBM would make sense if paired with cheap RAM like DDR4.


Considering that for now even AMD reduced the demand...
AFAIK there's no reduced demand now for HBM than there was during the Vega56/64 days. HBM is being used in a number of non-GPU chips, and there's also the MacBook Pro GPU.

I guess 8GB HBM + 16GB DDR4 could make some sense, for cost reasons. But I would think it's still better to have one pool.
UMA isn't the best approach if you're trying to squeeze the most bandwidth you possibly can from the system.
We know from Sony's presentations that in the PS4 the effective memory bandwidth for the GPU is significantly lower than the theoretical 156GB/s (176GB/s total - 20GB/s that the CPU can occupy). Flooding the memory controller with access requests from different subsystems reduces its effectiveness.


The Playstation 4 family's memory structure is a bit of an odd one, considering games only have access to 4.5 GB GDDR5 on Ps4 and 5GB GDDR5 on the 4Pro. I guess they could expand the PS5 alternate DDR4 memory pool enough to fit the entire OS there instead of only using it as a cache and swap scratch-pad.
IIRC the DDR4 pool in the PS4/Pro is only there to alleviate the main RAM to cache the non-gaming apps (netflix, spotify, etc. which you can instantly switch to/from the games). The CPU accesses it through the southbridge and it has very low bandwidth and high latency.
If the PS5 were to use an exclusive RAM pool for the OS, they'd need to make something very different from the PS4's southbridge RAM.
 
I think the whole theory of having a slower DDR4 or LPDDR4 memory pool for OS use is a total non-starter. It's like asking for the OS to be laggy and unresponsive. I think of it a bit like buying an SSD to install games on and installing Windows on a mechanical hard drive so you don't have to sacrifice SSD space for the OS. In a system with an SSD it probably doesn't even make sense for tombstoning apps anymore.

Does your Windows (or MacOS) install feel laggy and unresponsive? I think the whole point is that the OS doesn't need the incredibly high bandwidth a game would demand. If anything, DDR4 is more suitable for OS like tasks.
 
I think the whole point is that the OS doesn't need the incredibly high bandwidth a game would demand. If anything, DDR4 is more suitable for OS like tasks.

But that would be to the detriment of GDDR5 memory bandwidth, considering an entirely secondary and separate memory bus would take up pin-outs and die-space on the main SOC that would be better spent on a single memory pool. Net result is having less available or costing more. It doesn't seem economically feasible in terms of di space or cost. Right?
 
I'm going to make a giant reach here. It would be interesting to have fairly powerful set of arm cores to help with background tasks. Arm cores could act as ssd controller, network downloads, installations and maybe even run simpler apps like store, netflix, browser. Maybe if arm cores are low power enough and x86 side can be shutdown/idle super efficiently console could be completely quiet on media use cases and use something like 5-10W. In this case the memory attached to arm core can work as fairly fast cache/memory both for ssd and apps.

SSD controller for the assumed specs of next gen need to be fairly powerful. Another option would be to attach ssd directly to cpu via pci-e and use one of the x86 cores for io. Or have the traditional approach where there is separate controller just for ssd but in closed box this would feel not optimal.

It's very hard to find any concrete information on what the arm cores in ps4/xbox do. Will be interesting to see if arm carries over or gets removed.
 
Last edited:
It wouldn't be safe for Sony to go HBM route because of the latest Japan South Korea spat. The only manufacturing of HBM right now I think is Hynix and Samsung both from South Korea. If Japan government limits the the export of raw materials to South Korea needed to produce memory, cpus, and oled screens, there just wouldn't be enough supply. With GDDR6, Sony can go to Micron.
 
It wouldn't be safe for Sony to go HBM route because of the latest Japan South Korea spat. The only manufacturing of HBM right now I think is Hynix and Samsung both from South Korea. If Japan government limits the the export of raw materials to South Korea needed to produce memory, cpus, and oled screens, there just wouldn't be enough supply. With GDDR6, Sony can go to Micron.

While the concern can be a valid one, I doubt Sony architected the PS5's memory subsystem based on a spat that started a month ago.
 
Yeah at this point the ram type should be taped out. If it's HBM2 then a possible shortage could occur if the spat continues.
 
Status
Not open for further replies.
Back
Top