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

[Strange]

And it's a simple cost change, just like the 4GB=>8GB thing they did with PS4 (and much easier too imo). Other issues should be relatively simple to solve using software.

 

[Vhatt]

Guys, your recent discussion on memory contention lead me to do some googling to try and better understand what it was. While doing so I came across the following http://pages.cs.wisc.edu/~basu/isca_iommu_tutorial/IOMMU_TUTORIAL_ASPLOS_2016.pdf

I think the bits that pertain to the next-gent consoles begin on page 122 and end at 172. It's mainly grafts and large text so it shouldn't make for a heavy read. Could some of the more technical members have a quick look through and see if this could have been designed as part of the two systems set to launch later this year? Thanks in advance.

 

[vjPiedPiper]

IF the XSX wanted to have all it's ram accessible @ Max speed would that mean, increasing the total amount of ram or decreasing?
ie. could they do a sneaky last minute 6Gb RAM upgrade and let it all run @ max speed?

 

[see colon]

IF the XSX wanted to have all it's ram accessible @ Max speed would that mean, increasing the total amount of ram or decreasing?
ie. could they do a sneaky last minute 6Gb RAM upgrade and let it all run @ max speed?

If max theoretical bandwidth was the only concern, increasing or decreasing the memory so that all chips are the same size would achieve that goal. But that leads to other questions. Is 10GB too little? Is 20GB too expensive. Or, is 16GB the right total, but we need a portion of that to be as fast as possible because the main bandwidth sapping part of realtime 3d graphics is writing and reading to relatively small buffers.

 

[jlippo]

Maybe a depth-only pass where depth is easier to compress, and so having double the units would still be effective with less raw bandwidth.

Most likely.

ROPs in ps4pro shouldn't get near maximum outside really simple cases even without bandwidth limitation.

Has there been tests or napkin math on how much texture fetches or shader math can be done before they become limiting factor?
It shouldn't be much.

 

[Frenetic Pony]

If max theoretical bandwidth was the only concern, increasing or decreasing the memory so that all chips are the same size would achieve that goal. But that leads to other questions. Is 10GB too little? Is 20GB too expensive. Or, is 16GB the right total, but we need a portion of that to be as fast as possible because the main bandwidth sapping part of realtime 3d graphics is writing and reading to relatively small buffers.

Vendors only sell ram in limited sizes, with 8gb currently being the sweet spot for a lot of ram producers as far as DDR is concerned, I'd assume the same is true for GDDR. I'd also assume it was the best deal MS could get per gig with things like assembly and total size taken into account as well.

 

[goonergaz]

I disagree with the idea that you can't have cross-gen games that "fully take advantage" of the new hardware. Cross-gen does not have to mean parity. We've even seen the opposite in the past where next-get versions of cross-gen games were cut down compared to the prior-gen version. Fortunately, I don't expect to see that during this transition.

I understand that you could certainly create a game that fundamentally relies on an upgraded hardware spec that can't deliver even a down-graded experience on older hardware, but that's not going to create an inherently superior game, just a different one. I am eager to see the first example that delivers on both, but that won't preclude me from appreciating or enjoying scalable experiences simply because they are scalable.

I don’t disagree, but then you could have Star Citizen on a PS1 with that kind of mindset!

Also, we don’t yet know what the SSD might help deliver that’s not been spoken about by Sony. This could be the biggest game changer since moving from cart to CD or the requirement to have HDD or the move to 3D.

We do know the HDD is a massive bottleneck for devs and that PS5 exclusives don’t need to work to that limitation, I think it would be mad for Sony not to show what it brings to the table having put so much effort into its implementation in PS5.

 

[RDGoodla]

Also it was about what they felt was the best way to get to 10TF, not that a higher raw TF value isn't better.

Agreed, He wasn't lying, and he did explicitly also state that by increasing clocks there were limitations because memory wasn't going any faster. Most people choose to ignore his caveat. So while his statement is that a rising tide lifts all boats; the pipeline goes faster everywhere, this places additional pressure on memory being able to delivery on both it's bandwidth and latency.

If we see his example of 36CUs vs. 48CUs with the same Tera Flops, he mentioned the overall performance is raised due to higher frequency, and it is easier to fully use 36 CUs of the narrower GPU.
The point is faster GPU pipelines can increase performance significantly, a different way compared with more TereFlops.

And it’s very easy to deduce that PS5 GPU can beat a 48 CUs GPU with 10.3 TFs operating at 1.673 GHz. I assume PS5 probably matches the performance of 48CUs at 1.825 GHz. In other words real world in-game performance between xsx and PS5 is roughly 52/48 which is 8.3% and adding 2~3% of PS5 down clocking. Overall difference is 10~11%.

We can wait for multi platform games to reveal the performance results.

 

[RDGoodla]

That would be a first for sony, as for every gen it mostly where tech demos (order 1886) or just games that won't make any impression both graphics and gameplay wise (shadowfall, FC3 looks better). I rather have corporations put their resources and time into proper AAA titles, most don't have access to these new machines anyway at launch, which make it a niche market.

You're also forgetting scaling, which is something that didn't exist in the scope we know today.

Bloodborne is also SONY IP and has metecritic score 92.

( It is developed by other company just like 1886)

And now Sony studios become much more stronger than 2015 (e.g. Insomniac Games).

 

[BRiT]

The 448 GBps ibandwidth is alright if the GPU was clocked lower around 9.2TF but their upclocking of only the GPU makes it a bottleneck much earlier.

 

[PSman1700]

If we see his example of 36CUs vs. 48CUs with the same Tera Flops, he mentioned the overall performance is raised due to higher frequency, and it is easier to fully use 36 CUs of the narrower GPU.
The point is faster GPU pipelines can increase performance significantly, a different way compared with more TereFlops.

And it’s very easy to deduce that PS5 GPU can beat a 48 CUs GPU with 10.3 TFs operating at 1.673 GHz. I assume PS5 probably matches the performance of 48CUs at 1.825 GHz. In other words real world in-game performance between xsx and PS5 is roughly 52/48 which is 8.3% and adding 2~3% of PS5 down clocking. Overall difference is 10~11%.

We can wait for multi platform games to reveal the performance results.

As noted by DF, its the other way around. Also thats a 20 to 30% difference, depending on the boost.

 

[Karamazov]

That would be a first for sony, as for every gen it mostly where tech demos (order 1886) or just games that won't make any impression both graphics and gameplay wise (shadowfall, FC3 looks better). I rather have corporations put their resources and time into proper AAA titles, most don't have access to these new machines anyway at launch, which make it a niche market.

You're also forgetting scaling, which is something that didn't exist in the scope we know today.

What ? Killzone SF looked amazing at launch, FC3 ? Hope That's not far cry 3 ? Tried ps4 version last week, looked Really rough.

 

[function]

And it’s very easy to deduce that PS5 GPU can beat a 48 CUs GPU with 10.3 TFs operating at 1.673 GHz. I assume PS5 probably matches the performance of 48CUs at 1.825 GHz. In other words real world in-game performance between xsx and PS5 is roughly 52/48 which is 8.3% and adding 2~3% of PS5 down clocking. Overall difference is 10~11%.

That's a bit of a reach IMO.

There are a number of differences between the systems, including a large bandwidth advantage for XSX, almost certainly more GPU L2, and for workloads reliant on random reads probably somewhat lower latency in terms of cycles. I think the performance difference will probably be both less than and greater than the 18% flop difference depending on what's going on at the time.

Regarding Cerny's comments on 36 vs 48 CUs for 10.3 TF, if he said it, it must be true, at least generally and if all else remains equal. But I'd rather have the hypothetical 1.673 GHz 48CU system if it didn't throttle. And there may be some workloads where the large increase in memory latency puts the faster, narrower chip at a bit of a disadvantage. Ray tracing, for example, is supposed to be quite demanding in terms of random reads.

 

[goonergaz]

The 448 GBps ibandwidth is alright if the GPU was clocked lower around 9.2TF but their upclocking of only the GPU makes it a bottleneck much earlier.

So if we're comparing TF and B/W then;

PS5 = 43.6 GB/s @ 10.28TF to 48.7 GB/s @ 9.2TF
XSX = 46.1 GB/s @ 12.15TF

Or is that bad math?

If correct it hardly seems a big difference (~7% in very worst case)

 

[ultragpu]

Let's say it's 18% difference on average that would make it almost 2000p vs 2160p if the power is to be focused on resolution. If both are rendering at the same res then the 18% power might give the X better shadow quality, higher res volumetric lighting, slightly better raytracing? While PS5 might get better LOD, more varied texture, more high res assets per frame due to faster SSD?

So if we're comparing TF and B/W then;

PS5 = 43.6 GB/s @ 10.28TF to 48.7 GB/s @ 9.2TF
XSX = 46.1 GB/s @ 12.15TF

Or is that bad math?

If correct it hardly seems a big difference (~7% in very worst case)

It seems the extra bandwidth is not much of a surplus at all, more like what's required in proportion to the CU count.

 

[BRiT]

might get better LOD, more varied texture, more high res assets per frame due to faster SSD?

Only if the devs are incapable of using RAM.

 

[ultragpu]

Only if the devs are incapable of using RAM.

So why do you need an ultra fast SSD for? Just for faster load time and the lulz? Does it not help you stream more detail faster in tandem with RAM?

 

[goonergaz]

Only if the devs are incapable of using RAM.

Isn’t one of the benefits of the SSD that it potentially increases those things? We know it’s not just about load times, in fact Cerny seemed to want to make that fact clear.

 

[Inuhanyou]

Ram is fast but the hdd bottleneck makes a lot of things harder and more taxing on the entire system if not outright impossible. There is tons of usage systemwide for getting rid of something devs had to always plan around

 

[London Geezer]

Only if the devs are incapable of using RAM.

SquareEnix says hello, enters the room.