Saved texture in paint.net reduced the size by 50% and it's 1/3rd (ish)Just dropping texture resolution to half would drop bandwidth needed to 1/4th.
Saved texture in paint.net reduced the size by 50% and it's 1/3rd (ish)Just dropping texture resolution to half would drop bandwidth needed to 1/4th.
Saved texture in paint.net reduced the size by 50% and it's 1/3rd (ish)
Yep.
And those .05ms may be a bit too optimistic for a lot of consumer drives but surely a lot of them can hit 0.1 ms, which already makes them 50-100 times faster than a HDD. So access latency is imperceptible, and the system has "instantaneous" access to the data.
For example, the 970 Evo M2, when performing 4K Random Reads
.3 ms
DirectStorage is not going to convert anybody's PC into a device that can perform I/O as fast as PS5. If anybody thinks it will, they are lacking a basic understanding of PC architecture which is bunch of components connected across a number of buses. A Software API cannot have your GPU access your SSD directly - it will still have to go over the equivalent of north and south bridges with the CPU driving I/O transfers and possibly unpacking and decompressing data before it even gets to the GDDR.PC would be at big disadvantage here until DirectStorage is available. Sony solution DMA's data from controller to ram. in PC the data might flow through multiple sw layers and even IPC(kernel to user space). I know many datacenter solutions work around these kind of issues but regular apps probably don't.
DirectStorage is not going to convert anybody's PC into a device that can perform I/O as fast as PS5. If anybody thinks it will, they are lacking a basic understanding of PC architecture which is bunch of components connected across a number of buses. A Software API cannot have your GPU access your SSD directly - it will still have to go over the equivalent of north and south bridges with the CPU driving I/O transfers and possibly unpacking and decompressing data before it even gets to the GDDR.
Oh, I'm 100% sure that Microsoft, along with Intel and AMD, have looked at options for changing the I/O arrangement in future hardware and future iterations of Windows. What will not appeal to anybody, and which really is necessary if you want to approach the I/O efficiencies of nextgen consoles, is allowing the GPU to tap the SSD directly - effectively cutting around/through all the layers (and protection) inherent of the Windows microkernel.im actually wondering if directstorage will require hardware support from the cpu chipset and motherboard.
Let's think:Have a larger batch of system RAM reserved. System RAM to VRAM is faster than SSD to Unified RAM, and it has less latency.
PC games requiring 16 GB of RAM would be a very expected next generation difference I think.
Let's think:
Assming a next-gen open world map has 100GB data. On PS5 you can warp to any location instantly with the SSD.
On PC you need to put all of the 100GB into system memory.
We can say that if we want to the same degree of freedom in designing PC games we need 64GB system RAM REQUIRED and 128GB RECOMMENDED.
Which one? Would you explain why?Nope.
Which one? Would you explain why?
Can we please not use devices definitions like "PC" as they are very ambiguous. Do you mean "HDD", or "current Windows 10 SSD" or "Future SSD PC's contemporaneous with PS5"?Assuming a next-gen open world map has 100GB data. On PS5 you can warp to any location instantly with the SSD.
On PC you need to put all of the 100GB into system memory.
PCs will have faster NVME drives and DirectStorage API. PC's also have GPUs with dedicated 8GB of Video RAM in addition to general system memory; PCs have ability of having more resources loaded at any given time.
Can we please not use devices definitions like "PC" as they are very ambiguous. Do you mean "HDD", or "current Windows 10 SSD" or "Future SSD PC's contemporaneous with PS5"?
If we want to design a game for PC, with 100GB world map and we can warp to anywhere in the map, we should have a storage with at least 3~4GB/s of real in-game speed. Or you should have 100GB system RAM.
Do we have any PC SSD which has in-game read speed 3~4GB/s (NOT just the speed test)? To the best of my knowledge even using the fastest NVME SSD the loading time is still quite long for PC games.
going by the graphs and comments kraken achieves 3:1 compression on geometry. For these types of massive geometry new gen games, that would balloon size by 3x potentially. It would also mean, depending on the format, that ps5 potentially achieves 16.5GB/s effective bandwidth when streaming the billions of nanite triangles.Yes but GPU's deal with the common compressed texture formats natively so they will remain compressed in the PC space as they are now.
The difference here is the full data compression that the consoles are offering on top of that in the form of Kraken/BCPACK which is on average no more than 2:1.
where did you hear kraken is only for textures? kraken has been compared to zlibBCPACK and Kraken is for textures.. Zlib is for everything: textures, audio, geometry, probably some scripts too, etc. It's a zip file, and every game has its assets inside zip files (regardless of file extension).
going by the graphs and comments kraken achieves 3:1 compression on geometry. For these types of massive geometry new gen games, that would balloon size by 3x potentially. It would also mean, depending on the format, that ps5 potentially achieves 16.5GB/s effective bandwidth when streaming the billions of nanite triangles.
I imagine it loads both triangles and textures, unless it is some form of geometry texture or some other unknown format. But given the data in whatever format should probably have similar underlying structure(i imagine since it is the same geometry data only in another format or shape), I imagine the geometry component would load relatively fast.Untextured triangles?
We don't know that until we know how virtualisation works. Remember, 100 objects with 4k textures requiring loads of data (~5 GBs raw 24 bit colour) can be represented with 64 MBs of textures when healthily virtualised. The actual amount of geometry needed to render a frame is as many triangles as there are pixels in a perfect case, so 8 million triangles for 4K. The amount to transfer will be somewhere between that 8 million vertices and all the geometry for all the object. Until we know what that mid-point is, we don't know what minimum speed is needed on storage.If we want to design a game for PC, with 100GB world map and we can warp to anywhere in the map, we should have a storage with at least 3~4GB/s of real in-game speed. Or you should have 100GB system RAM.