Tom Warren: XSS GPU to be 20CU running at 1.55Ghz ≈ 3.97TF

how do you know? people doubted the one x to do 4k and it has titles that justify the claim of “true 4k” . With historical evidence, we can clearly see they can use these words for marketing and will achieve these items with whatever they want. is it right? probably not, but expect that to be done as we seen it done in the past is what im trying to say.

Plus Ms already markets the series x as 8k ready.

XSX can support up to 8k yes, but the volume of games that will support 8k resolution you’ll be able to count on 1 hand so using it as a marketing strategy would be extremely misleading. The same can probably be said for XSS, it seems like this will be marketed towards 1080p for most titles rather than 4k, again, they don’t want to mislead customers about what the capabilities of each device are.

As someone with a 4k TV, if they market them as 4k and 8k, why would I get the 8k model? I wouldn’t, but then the XSS isn’t the device for me as it won’t take full advantage of my 4k TV.

Sorry!! Haven’t seen it …

I don’t expect the “hardware acceleration” is nearly as performant as NVIDIA’s tensor flow cores. Be curious how much GPU resources the upsampling would take.

1 Like

The cost of Super Resolution scale pretty linear with the number of source pixels. So upscaling from 720p to 1080p will need less inference ressources than upscaling from 1440p to 4k.

Does anyone remember when and where the 4TF rumor came from? I’ve been following the consoles and I don’t recall an actual source for this like some of the other leaks. From what I remember it was just a guess based on a supposed leak of the SX, but maybe there is something more concrete I missed.

I think the original leak came from @Jez and Window Central.

There’s a good chance we won’t be seeing 1X enhancements supported on XSS. If anything, it’ll likely run games at the upper bounds of resolution and frame rate for XBOS versions.

For one it isn’t targeting 4K, but also the RAM is much faster so even though it has less it can push the data through faster so it might need to store less data in RAM to run a One X profile game on it.

They do this right now with the Xbox one x. So I’m implying the same strategy will be used for series s and x. We already see the tag lines already being used in every publication for the series x.

I mean people thought series x wouldnt have any 120fps games. Where are they now? Oh yeah, saying there aren’t many people with 120hz sets lol.

Kekekekekekekkeke

Hopefully MS explains how this is going to work, because that’s not how programs run. If it loads X GB of data to RAM, it doesn’t matter how fast ithe RAM is, it needs that exact amount available.

I’m not personally interested in the lockhart S but I assume MS has market research saying it has an audience. Like I know a guy who just recently got a PS4 and was still gaming on a PS3 up until he got it because he doesn’t have a lot of money and that’s all he could afford. Will be interesting to see what the market for this will be. Devs can say they won’t bother but if it starts selling, they will start making games for it.

The tensor cores are severely underutilized because the theorical figure assumes you are using matrix x matrix multiplication and that does not happen that often, specially in the inference stage (but even for learning it’s not actually used that much).

Though in the end all boils down to performance. Pc will always have the edge compared to consoles, high or low end, but doesn’t mean it the consoles won’t be able to pull off some limited stuff.

1 Like

I guess the bigger question would be: is it supposed to run the x1x version of games or have its own set of auto enhacements from base? (Like forcing 1080p plus doubling the framerate)

They said the velocity architecture allows for the total addressable memory to be bigger than the physical memory. Maybe that’s how they would be able to do the X1X version of games, like RDR2 in 4K.

Yeah that’s true, I just find it interesting there is 20% less ram but that it is 70% faster (not counting the RAM reserved for the OS on either system.) I am just curious if MS engineers have something on the hardware level cooked up for that.

The gpu - cpu balance this gen gen was way off and everybody knows that cpu was heavily lacking. So this sounds like a more balanced approach. I still expect this machine to be a 1080p/60fps box and these specs sound like they would do just that.

If they can just enable SFS for example with no input from the dev it would be possible, but I think that the game will need to use the apis to make use of that

Yup, this won’t apply to BC titles.