@BlueOcean Going back to the power pool, Sony's plan is pretty clear, it's to make the console cheaper and keep the fan quiet. They expect most early games won't be able to keep the CPU and GPU both close to 100% utilisation for a long time, and that has traditionally been the case with most consoles and meanwhile they've made it easier to get there as soon as possible by keeping the system balanced and easy to work on. By the time developers start hitting the upper limits that cause notable clock drops, we'll already be looking to the mid-gen refresh i.e. PS5 Pro and Xbox equivalent, for anyone who cares about maximum performance.
@BlueOcean So some of this isn't quite the case, it's true that Sony has a common power pool, however it's not actually true that you can't run the GPU and CPU at maximum clock at the same time. Power consumption isn't just linked to clock, it's also driven by how many transistors are working at any one point in time. For example if you're performing heavy AVX calculations on the CPU, it's going to consume a lot more power than working through a normal pipeline. As Sony puts it, it'll be working at maximum clock the vast majority of the time, if needed; it's when you perform certain combinations of actions that use a lot of energy, it'll drop the clock, but only by a few percent, as a small drop in clock gives a big improvement in power consumption. These changes are 100% predictable, repeatable and thus profilable.
The PS5 CPU also does have SMT, Sony just never pointed it out because the cases where you wouldn't use SMT are pretty-much non-existent, it's largely marketing blurb to allow a higher clock speed to be quoted. SMT is just AMD's version of hyper threading, which comes as standard in all of their CPUs; in a nutshell, the PS5 has 8 cores, with 16 hardware threads (SMT).
Quite rightfully as you say you could use some RAM for the GPU, some for the CPU and if used correctly this would yield a performance boost, but there are a lot of challenges and choices a developer might face. What if they want to use more than 10GB for the GPU? How can they write their engine to somehow manage the difference in speed for the extra data, is there going to be an algorithm, will that itself eat up performance? What if they want to transfer a lot of data between CPU and GPU, will that data always have to be limited in speed or do they transfer it over to fast RAM? But that takes bandwidth. What if suddenly the CPU becomes the bottleneck, do they have to rethink how all this will work? And remember, what the PS5 does have going for it here is a much faster cache for prefetching from RAM. Also, most developers are 3rd party and coming from PS5 or PC to Xbox or going there after Xbox, they don't want to have to deal with this customisation.
The amazing thing about the SSD on both the Xbox and PS5 is that other than asking for a file to be loaded, there's no CPU intervention at all, they really do run at full speed, all the time, loading data directly into RAM. Subtle differences in performance on the CPU/GPU aren't going to impact that - you load the model into RAM, the GPU renders it and as I mentioned Sony actually reduces a bottleneck here specifically with cache scrubbing. I haven't really spoken about it but the new Geometry shaders on both consoles are also a key component to this, they can handle vertex manipulation traditionally done by the CPU and they can cull any non-visible vertexes, allowing you to have large levels of model detail at a very low cost to bandwidth and performance (this was key to the UE5 PS5 demo).
@BlueOcean I think the theory actually conveys a fairly level playing field, it's the reality I'm not sure about, here's why:
On the face of it: PS5 GPU Teraflops: 10.28, CPU (SMT): 3.5GHz Xbox GPU Teraflops: 12.115, CPU (SMT): 3.6GHz Difference: 17.85% / 2.85%
However, once you dig deeper:
Teraflops is only a measure of the Vector ALU performance.
PS5 GPU has a clock speed 22.19% faster, so other parts of the GPU, the cache, the rasterizer etc run faster.
PS5 GPU has fewer CUs (in exchange for a higher clock), it's easier to fill fewer CUs with useful work in practise.
The PS5 SSD performs on average at 8-9GB/s (20GB/s peak) compressed vs Xbox at 4.8GB/s compressed.
Microsoft have been sparse on the SSD details, their 4.8GB/s suggests 37% better compression than Sony, but as far as we know, their hardware compression is zlib and so it should be 10% worse than the PS5's Kraken, so take the 4.8GB/s with a BIG pinch of salt, using Sony's ratios minus 10% you'd expect to see 3.76GB/s on the Xbox, or alternatively apply Microsoft's ratios to the PS5 + 10% and you get 11.55GB/s.
PS5 SSD has 16 priority levels vs the standard 2, making it easier for developers to have the data they need when they need it, potencially reducing the number of wasted cycles waiting for data or poping/loading times.
PS5 has a bespoke cache scrubber on the GPU, when RAM is updated with SSD data, its cache is scrubbed, improving its performance by wasting less GPU cycles.
Xbox achieves its performance partly through split speed GDDR, 10GB is faster than the PS5, whilst 6GB is slower, this will take developers more effort to optimise for as they decide how to split their usage of the RAM.
I'm sure there are plenty of details we still don't know about both consoles, but I think once you consider some of these architectural differences, the picture looks much less clear cut that simple teraflop and GHz numbers would have you believe.
I do also believe that extra SSD speed can go a long way. The Xbox SSD is fast enough to speed up the loading time of classic games, but we're about to enter an era of games with no limits, where even as your character turns around new textures are being loaded, allowing for endless draw distances, no popping, higher detail, more variety, more players, no loading passagways etc. In that world, a faster SSD is the difference between how fast your character will be allowed to turn or the number of players on a map all casting different spells etc.
Agreeing with the comments, he wasn't exactly going to say he felt bad, and it is still early days in reveals, not everything has been announced on either side yet and for the most part, many games from each side's reveal will come to the other at some stage.
I do remain unconvinced that there's truly a hardware advantage. OK maybe it'll push a few extra frames here and there on a classic game but the PS5 is likely going to excel in audio, haptics, loading times and open worlds. I think it's very tit for tat this gen, there's ups and downs to each.
Comments 4
Re: Xbox Boss 'Felt Good' After Watching Sony's PS5 Show
@BlueOcean Going back to the power pool, Sony's plan is pretty clear, it's to make the console cheaper and keep the fan quiet. They expect most early games won't be able to keep the CPU and GPU both close to 100% utilisation for a long time, and that has traditionally been the case with most consoles and meanwhile they've made it easier to get there as soon as possible by keeping the system balanced and easy to work on. By the time developers start hitting the upper limits that cause notable clock drops, we'll already be looking to the mid-gen refresh i.e. PS5 Pro and Xbox equivalent, for anyone who cares about maximum performance.
Re: Xbox Boss 'Felt Good' After Watching Sony's PS5 Show
@BlueOcean So some of this isn't quite the case, it's true that Sony has a common power pool, however it's not actually true that you can't run the GPU and CPU at maximum clock at the same time. Power consumption isn't just linked to clock, it's also driven by how many transistors are working at any one point in time. For example if you're performing heavy AVX calculations on the CPU, it's going to consume a lot more power than working through a normal pipeline. As Sony puts it, it'll be working at maximum clock the vast majority of the time, if needed; it's when you perform certain combinations of actions that use a lot of energy, it'll drop the clock, but only by a few percent, as a small drop in clock gives a big improvement in power consumption. These changes are 100% predictable, repeatable and thus profilable.
The PS5 CPU also does have SMT, Sony just never pointed it out because the cases where you wouldn't use SMT are pretty-much non-existent, it's largely marketing blurb to allow a higher clock speed to be quoted. SMT is just AMD's version of hyper threading, which comes as standard in all of their CPUs; in a nutshell, the PS5 has 8 cores, with 16 hardware threads (SMT).
Quite rightfully as you say you could use some RAM for the GPU, some for the CPU and if used correctly this would yield a performance boost, but there are a lot of challenges and choices a developer might face. What if they want to use more than 10GB for the GPU? How can they write their engine to somehow manage the difference in speed for the extra data, is there going to be an algorithm, will that itself eat up performance? What if they want to transfer a lot of data between CPU and GPU, will that data always have to be limited in speed or do they transfer it over to fast RAM? But that takes bandwidth. What if suddenly the CPU becomes the bottleneck, do they have to rethink how all this will work? And remember, what the PS5 does have going for it here is a much faster cache for prefetching from RAM. Also, most developers are 3rd party and coming from PS5 or PC to Xbox or going there after Xbox, they don't want to have to deal with this customisation.
The amazing thing about the SSD on both the Xbox and PS5 is that other than asking for a file to be loaded, there's no CPU intervention at all, they really do run at full speed, all the time, loading data directly into RAM. Subtle differences in performance on the CPU/GPU aren't going to impact that - you load the model into RAM, the GPU renders it and as I mentioned Sony actually reduces a bottleneck here specifically with cache scrubbing. I haven't really spoken about it but the new Geometry shaders on both consoles are also a key component to this, they can handle vertex manipulation traditionally done by the CPU and they can cull any non-visible vertexes, allowing you to have large levels of model detail at a very low cost to bandwidth and performance (this was key to the UE5 PS5 demo).
Re: Xbox Boss 'Felt Good' After Watching Sony's PS5 Show
@BlueOcean I think the theory actually conveys a fairly level playing field, it's the reality I'm not sure about, here's why:
On the face of it:
PS5 GPU Teraflops: 10.28, CPU (SMT): 3.5GHz
Xbox GPU Teraflops: 12.115, CPU (SMT): 3.6GHz
Difference: 17.85% / 2.85%
However, once you dig deeper:
we know, their hardware compression is zlib and so it should be 10% worse than the PS5's Kraken, so take the 4.8GB/s
with a BIG pinch of salt, using Sony's ratios minus 10% you'd expect to see 3.76GB/s on the Xbox, or alternatively
apply Microsoft's ratios to the PS5 + 10% and you get 11.55GB/s.
potencially reducing the number of wasted cycles waiting for data or poping/loading times.
performance by wasting less GPU cycles.
this will take developers more effort to optimise for as they decide how to split their usage of the RAM.
I'm sure there are plenty of details we still don't know about both consoles, but I think once you consider some of these
architectural differences, the picture looks much less clear cut that simple teraflop and GHz numbers would have you believe.
I do also believe that extra SSD speed can go a long way. The Xbox SSD is fast enough to speed up the loading time of classic games,
but we're about to enter an era of games with no limits, where even as your character turns around new textures are being loaded, allowing
for endless draw distances, no popping, higher detail, more variety, more players, no loading passagways etc. In that world, a faster SSD
is the difference between how fast your character will be allowed to turn or the number of players on a map all casting different spells etc.
Re: Xbox Boss 'Felt Good' After Watching Sony's PS5 Show
Agreeing with the comments, he wasn't exactly going to say he felt bad, and it is still early days in reveals, not everything has been announced on either side yet and for the most part, many games from each side's reveal will come to the other at some stage.
I do remain unconvinced that there's truly a hardware advantage. OK maybe it'll push a few extra frames here and there on a classic game but the PS5 is likely going to excel in audio, haptics, loading times and open worlds. I think it's very tit for tat this gen, there's ups and downs to each.