I am hanging out to see what MS show us on the 17th August at the Hot Chips show. A deep dive into the system, and hopefully alot of our questions are answered.
Now does anyone know how we can watch it?
2 Likes
Is there much we don’t know at this point?
I’d be more interested in the Series S hardware. Will there be some kind of upscaling tech line DLSS involved?
I think DirectML will be for that.
2 Likes
I was told by a dev there is. I guess we’ll see how much they were “holding back” soon.
1 Like
Honestly, between this and the potential unveiling of the Series S, I am quite excited !
I would think the XSS would have the same 4/8-bit integer support the XSX has.
4 Likes
Both XSS/XSX will have ML hardware. There are still some lingering questions related to XVA.
1 Like
This is what i suspected is still out there to talk about. They’ve said very little about ML to this point. My bet is they show it off using BC games.
I think there was Playground Games who used that tech on Forza Horzion 3.
I doubt they show any games in some major way. Don’t think HotChips is really the kinda show for that. They MIGHT give a peak how it is used in a game or two tho. If I recall, their X1 HotChips presentation included RYSE. I think.
There was already a very big technical article which goes deep enough with loads of great infos inside if you want to read more about why it is a gamechanger.
Here is just a glimpse
ML Super Sampling (left) and bilinear upsampling (right)
9 Likes
Really cool article. No question ML will help development. Leveraging cloud compute for neural networks, they could offer developers tools that continue to evolve all gen. Question is to what level does local hardware play in some of the features they talked about? The upsampling for example.
They also haven’t discussed physics one bit which I’m surprised about. They do own Havok after all and rebranded it a few years ago. Zen 2s should be capable of a lot more than Jaguar cores for physics calculations.
4 Likes
I think we could learn more about the RDNA 2 architecture just by itself. How does RT work on XSX? What can we truly expect from ML on XSX? How exactly does XVA have 100gb of virtual RAM? What customizations did they add to the GPU and CPU over stock AMD parts? I’m really interested to see what they reveal.
3 Likes
This is something that stuck out to me as well. We have seen a lot of games this gen avoid collision with foliage for example. I can see how the world reacts to the player be taken to a new level next gen.
1 Like
Honesty I think we won’t see most of those. It’s probably more about the internal buses (interconnects), coherency, MMU, memory, and some heavy coverage on the decompression stuff.
Looks like the 1X talk is available here so people can get an idea of what to expect (9 mins in)
3 Likes
There is a lot on a hardware level we don’t know specifically the implementation of some things. While I doubt we will get an answer from the presentation there is for example how RT is implemented into the design. What modifications they did to CPU & GPU that are not standard. How they end up being much smaller than anticipated compared on how structure sizes from Zen 2 and RDNA GPUs on the market. Those kind of information.
2 Likes
I’m interested to learn how the 76 MB of SRAM has been appropriated across the chip, especially how much cache the CPU has.
1 Like
So anyone know where we can watch it? If we can watch it.
1 Like
It’s a paid industry event so you’d have to register (and pay the registration fee). Other wise you can get information from press write ups fairly quickly.
1 Like