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

[mrcorbo]

You could be correct on the GPU, potentially, being bandwidth limit (in certain cases). Seeing a lot of top PS3 GDC tech breakdowns, it didn't seem to be a limiting factor for those studios. It could've been for other studios. I wish we had more data on that.

PS3 developers making exclusive games didn't do things in their games that the PS3 wasn't good at.

 

[Shifty Geezer]

Biggest disappointment of PS3 was the drop in particle quality from PS2. The special effects of the PS2 games I enjoyed really made them zing. Shadow quality took a nose-dive too with shadow volumes replaced with shadow mapping.

 

[Deleted member 11852]

PS3 developers making exclusive games didn't do things in their games that the PS3 wasn't good at.

Yup, like Cell's read speed from local memory was 16MB/s - leading Sony to have to write "(no, this is not a typo)" on dev documentation.

 

[Arwin]

Yup, like Cell's read speed from local memory was 16MB/s - leading Sony to have to write "(no, this is not a typo)" on dev documentation.

That is a completely irrelevant remark though as you would DMA data in there and then use the blazing fast local storage RAM.

 

[Shifty Geezer]

Yes. You'd work with the HW's strengths, not its weaknesses, which is the point being made.

 

[Arwin]

Yeah no, that wasn’t really the reason why framebuffer effects sometimes ran at quarter-res though, that wouldn’t have allowed for even a fraction of that [emoji51]

 

[PSman1700]

Biggest disappointment of PS3 was the drop in particle quality from PS2. The special effects of the PS2 games I enjoyed really made them zing. Shadow quality took a nose-dive too with shadow volumes replaced with shadow mapping.

ZoE2

 

[Lucid_Dreamer]

WRT to PS3 GPU bandwidth limitations being a common concern on PS3. Many PS3 games used quarter resolution alpha effects while Xbox 360 versions did not.

True. Alpha effects were not generally stressed as much with 1st party PS3 titles, IIRC. 3rd party and 1st party X360 titles cut alpha effects down to the size of the X360's eDRAM.

Besides alpha effects, what else can you think of that was greatly limited GPU bandwidth. Also, what effects do you think would/could cause the next-gen consoles "bandwidth fever"?

 

[Inuhanyou]

360s edram was great. Cerny once talked about 1tb/s edram on ps4...in wonder if those kind of solutions will be something to look at in future. 256mb of 5tb/s anyone?

 

[Deleted member 11852]

360s edram was great. Cerny once talked about 1tb/s edram on ps4...in wonder if those kind of solutions will be something to look at in future. 256mb of 5tb/s anyone?

This was covered in Mark Cerny's 2013 GameLabs talk 'The road to PS4', he said:

We didn’t want the hardware to be a puzzle that the developers would need to solve to make quality titles.

And just to give a specific example, this gets technical really quickly, please forgive me. The architecture we ended up with for PlayStation 4 used a 256-bit bus and a type of memory found in top of the line graphics cards called GDDR5. And the combination of this wide bus and this fast memory gives us a 176Gb per second which, and many of you will have to take my word for it, is quite a lot.

So with that much bandwidth straight forward programming techniques usually result in some pretty impressive graphics. Now we knew that was an alternative architecture that would be a bit easier to manufacture. In this architecture we would use a narrow 128-bit which would drop the bandwidth to 88Gb per second, which is not particularly good in next generation terms and would therefore really hurt the graphic performance.

So we then use some very fast on-chip memory to bring the performance back up. If we used EDRAM for this on-chip memory, we know that bandwidths as much as one terabyte per second, that’s 1,000 gigabytes a second, would be achievable. The catch though, is that the on-chip memory would need to be very small. And each game team would need to develop special techniques in order to manage it.

So to compare these two architectures, the one on the left [actual PS4 design shown] has 176Gb per second for any access, and the one on the right [alternate PS4 design shown] 88Gb per second if data is in system memory or 1,000 gigabytes per second if the data is in that tiny EDRAM. And at first glance the architecture on the right looks far superior to the one on the left. Sure it takes a while to figure out how to use it but once you figure out how to use that little cache of EDRAM you can unlock the full potential of the hardware. But to our new way of thinking, the straight forward approach on the left is definitely advantageous.

Since Cerny began leading PlayStation hardware direction (Vita, PS4, PS4 Pro, PS5), he has delivered hardware based on feedback from devs on what they need to solve problems. So I'd say, as long as Mark Cerny continues to lead PlayStation hardware, and devs continue to shun esoteric solutions, you'll never seen something like EDRAM in a PlayStation.

 

[Tkumpathenurple]

Since Cerny began leading PlayStation hardware direction (Vita, PS4, PS4 Pro, PS5), he has delivered hardware based on feedback from devs on what they need to solve problems. So I'd say, as long as Mark Cerny continues to lead PlayStation hardware, and devs continue to shun esoteric solutions, you'll never seen something like EDRAM in a PlayStation.

I wonder what, if anything, might make them reconsider that to any extent? The example that faced Cerny was quite a dramatic example; the positive developer feedback and swift utilisation of its resources indicates that they made the right choice.

If the on-chip memory could be larger, or the sacrifice to the main memory's bandwidth lesser, I wonder if they would necessarily make the same decision?

If, for example, HBM3 is released (~512GB/s per stack) and becomes relatively affordable by the time of an equally hypothetical PS5Pro. Relatively affordable in the sense that it will be cheap enough to supplement the main memory, but not replace it.

Hypothetical PS5Pro memory configurations:

1. 20GB's 16Gbps GDDR6 = 640GB/s
2. 20GB's 18Gbps GDDR6 = 720GB/s
3. 16GB's 16Gbps GDDR6 + 1 stack 4GB's HBM3 = 1024GB/s

The first two options are the most straightforward, but the last one has relatively high bandwidth for the memory pool with the largest capacity, and relatively high capacity for the memory pool with the largest bandwidth.

Perhaps ray tracing might warrant this shift? Don't BVH's benefit from very high bandwidth?

 

[Inuhanyou]

Iirc the main issue of 360s edram was that they didnt put in enough, so devs had to sacrifice and scrape by to get everything into that 10mb framebuffer. But even then the main bw of 360 was around the ps3s even without usage of the edram...correct me of im.wrong.

So like the post above me says...if they could increase the amount of edram and minimize the bw penalty to main ram it would be pretty interesting

 

[Dictator]

Perhaps ray tracing might warrant this shift? Don't BVH's benefit from very high bandwidth?

In my talks with devs, yeah memory bandwidth is very important. According to one: "RT is bw hungry as hell".

 

[Shifty Geezer]

Hypothetical PS5Pro memory configurations:

1. 20GB's 16Gbps GDDR6 = 640GB/s
2. 20GB's 18Gbps GDDR6 = 720GB/s
3. 16GB's 16Gbps GDDR6 + 1 stack 4GB's HBM3 = 1024GB/s

The first two options are the most straightforward, but the last one has relatively high bandwidth for the memory pool with the largest capacity, and relatively high capacity for the memory pool with the largest bandwidth.

In a pro, no way. It'll require specific, considerable efforts to use the machine. A pro should just run the base game with more room for improvements without needing an engine reworking to manage a small memory pool.

Given faster SSDs, I could definitely see something like [fast SSD > relatively small fastish DDR > very fast HBM] in future hardware. The DDR would sit more as a cache between storage and working RAM. Everything will be streamed and tiled and virtualised!

 

[Deleted member 11852]

I wonder what, if anything, might make them reconsider that to any extent? The example that faced Cerny was quite a dramatic example; the positive developer feedback and swift utilisation of its resources indicates that they made the right choice.

Things change, but as long as devs are not embracing weird tech I don't see Cerny-led Sony hardware returning to the mad hardware of PS2 and PS3. Cerny's time-to-triangle metric is vague and a bit trite, but in terms of expressing the ease/effort with which you can start putting something onscreen, it does it's job and doing things easily and quickly will tend to find favour with most people - particularly those paying the bills.

 

[fehu]

A pro should just run the base game with more room for improvements without needing an engine reworking to manage a small memory pool.

coff coff one x coff transparent coff

 

[Shifty Geezer]

coff coff one x coff transparent coff

It's very easy to fit data designed for small pools into a larger pool, but not so easy the other way around.

 

[BRiT]

Xbox One games can run unpatched on OneX perfectly fine.

So what was the intended point, besides spreading virus through coughing?

 

[Kaotik]

Xbox One games can run unpatched on OneX perfectly fine.

So what was the intended point, besides spreading virus through coughing?

I was wondering why Windows Defender started screaming every time I visit this thread, now I know why!

 

[Jay]

In a pro, no way. It'll require specific, considerable efforts to use the machine. A pro should just run the base game with more room for improvements without needing an engine reworking to manage a small memory pool.

Given faster SSDs, I could definitely see something like [fast SSD > relatively small fastish DDR > very fast HBM] in future hardware. The DDR would sit more as a cache between storage and working RAM. Everything will be streamed and tiled and virtualised!

I wouldn't have gone with the PS5 Pro as an example, I'd have gone with PS6.
But his example would still mean it runs the base game better with no code changes.

But to make full use of pro features would need code changes, just like current mid gen with id buffer, or single pool on 1X.

Unless you think it will just be faster with no IP changes at all so different than the way current mid gen was approached? I think that would be very limiting.