Gamers are disappointed and developers are excited. Weird eh?
So did Mark Cerny went crazy or just forget how to design a console?
No, he just choose a different approach. The most used philosophy in console design is to go all in for the GPU of the system.
I think that is clear that on PlayStation 5 the heart of the system is the SSD and then the APU.
Why? Why developers are so excited for the SSD? OK, SSD storage is new tech for consoles but on PC we have this option for many years now.
Nearly no loading times of fastest streaming of assets is exciting but this is only a fraction of the possibilities from a very fast SSD.
This is why developers are excited and I think that even them they really don't know for sure the possibilities of the SSD as the main storage solution.
We had some case studies from some studios in the past (nearly 10 years ago) for the possibilities that a SSD gives on gaming development.
For anyone that is interested here is a
GDC 2011 presentation from Iron Galaxy Studios on this subject. Keep in mind that is an old case study with a regular Intel SATA SSD. Spoiler alert. They blew away from the results.
So why did Sony went all in for the SSD? PCs and Xbox Series X have SSD storage, why Sony's solution is any different?
Unique custom SSD setup:
Xbox Series X has a NVMe SSD with 2.4GB/s, PC NVMe PCI 3.0 can reach 3.5GB/s and PCI 4.0 up to 7.0GB/s.
PlayStation 5 can reach 5.5GB/s. So it is fast but not the fastest. Or is it?
These speeds are from the maximum bandwidth of the SSD but they are not representative of the final actual speed.
On every system (PC or console) there is a lot of overhead and other bottlenecks that affects a lot the real benefit from a fastest storage option:
It can't be so bad. Can be? PC benchmarks shows that the bottleneck issues are very real:
The total custom SSD setup on PlayStation 5 was created to resolve these bottlenecks so the system can take full advantage of the raw SSD speed:
We can't know for sure before real tests but I believe that the storage on PlayStation 5 will be a lot faster even in comparison to expensive NVMe PCI 4.0 SSD.
OK, so maybe PlayStation 5 will have the best loading times or even nearly no loading times as Mark Cerny said. That's it?
No, the advantages are many, many more.
The obvious:
- Boot times
- Loading times
- Update times
- No duplicate data
- Really fast streaming of assets
These are some of the advantages that I think Sony is looking to have the best performance with their SSD:
- Highest possible resolution textures
- Textures variety
- Assets variety
I believe these points are the most interesting for next gen games.
Mark Cerny maths:
Mark Cerny said yesterday, on a scenario that the SSD could be so fast to load necessary data on the fly that, 4GB of compressed data to load for every moment of gameplay seems about right for a next gen game.
They estimated that a player in a game is turning in 0.50 seconds so this is the time target they had for the SSD speed.
PlayStation 5 can load 4GB in 0.54 seconds. In comparison PlayStation 4 would need 80 seconds for the same data to load.
Data management is going to have a revelation with SSD. A former Naughty Dog technical art director
tweeted about that yesterday:
The SSD is so fast that PlayStation 5 can store data for the next 1 second of gameplay in the RAM. In comparison PlayStation 4 must store for the next 30 seconds of gameplay. This is going to have a huge impact in the quality of textures, etc.
APU:
OK, the SSD is not so bad. How about the GPU? Something must render and render fast all these data from the SSD.
The frequency of the GPU is absolutely insane for a console or for any known GPU to be honest.
The GPU frequency is capped at 2.23GHz. Not up to 2.23GHz, not overclocked but
capped to this max frequency.
Obviously this is nearly at the maximum frequency the RDNA 2 chip can safely operate. Great job from AMD on these chips.
Sony went with a small and fast GPU. Microsoft went with the more conventional big and slow GPU option.
So why Mark Cerny went with this direction? Its unconventional and for sure harder option for a console.
As Mark Cerny said the high frequency has some advantages and the disadvantages are less.
If we aim for a certain TFLOPS target we have two GPU options. Big and slow or small and fast. The percentage difference (X%) in the frequency of the smaller and faster GPU can give the following advantages:
- Rasterization is X% faster
- Processing of the command buffer (example of command buffer) is X% faster
- L caches on the GPU have X% more bandwidth
Also important that less CU (compute units) can be fully utilized easier on a GPU.
The faster rasterization probably was the key advantage behind the direction Mark Cerny went with the GPU.