we literally have road to PS5 and Microsoft explaining how the velocity architecture works on top of Sampler feedback sharing to all talk about memory management. We do know.Cool, you're a dev with practical hands on experience then, awesome.
we literally have road to PS5 and Microsoft explaining how the velocity architecture works on top of Sampler feedback sharing to all talk about memory management. We do know.Cool, you're a dev with practical hands on experience then, awesome.
The 970 is a poor example since it was shown that when it needed all 4GB of ram performance dropped.Are you suggesting that the XSX will randomly allocate things on the two pools of RAM, without taking into account the bandwidth that each tasks requires? We already have examples of this with the GTX 970.
The 970 is a poor example since it was shown that when it needed all 4GB of ram performance dropped.
Yep. That's how it works in Nvidia GPUs so I'd expect the same to be true for RDNA2.The rumour is that RDNA2 raytracing performance is tied directly to the number of CUs. Didn't the Forza promo video have more complex RT reflections than the GT7 demo? Not quite a far comparison though since GT7 showed what looked like actual gameplay.
And I'm saying that it also showed that when it needed more than 3.5GB of ram for that given task it would get an overall performance hit. It comes down to how much RAM developers have access to and how it can be used.I specifically mentioned the GTX 970 because it is exactly the same situation. This is to show that tasks that require less bandwidth can be allocated on the slower portion of RAM and this is what we can expect to happen with the XSX. Ray Tracing can be done on the high speed portion of the RAM, the OS can be on the slower portion.
I see people saying this a lot, but didn't they show us this Gears Demo running on the hardware all the way back in March? Is there something I'm not getting, as to why this doesn't count?
I specifically mentioned the GTX 970 because it is exactly the same situation. This is to show that tasks that require less bandwidth can be allocated on the slower portion of RAM and this is what we can expect to happen with the XSX. Ray Tracing can be done on the high speed portion of the RAM, the OS can be on the slower portion.
Pretty sure there is no game which uses machine learning for denoising at the moment, all use normal filtering methods.The XSX has more CU, ray tracing consumes a lot of memory bandwidth. The PS5 has 448GB/s and the XSX has 560 GB/s. Another big factor that we need to find out is if the PS5 has dedicated hardware to accelerate machine learning. If not, the denoising step should be more efficient on the XSX. Current ray tracing solutions don't shoot enough rays to cover the whole scene. You can see on the image below how it looks, with black spots where no rays reached. Machine learning is necessary to fill out the blank spaces.
It...doesn't have ray tracing so it doesn't count for showing off the Series X ray tracing?
And I'm saying that it also showed that when it needed more than 3.5GB of ram for that given task it would get an overall performance hit. It comes down to how much RAM developers have access to and how it can be used.
Pretty sure there is no game which uses machine learning for denoising at the moment, all use normal filtering methods.
This video makes a big point that it has no ray tracing, because the entire game was built around the old lighting system, and would require redoing every level's lighting to use ray tracing. But also, at that point they had only spent two weeks working on the Series X version, basically taking the PC version in Ultra and using some new screen-space effects from the newer version of UE4. In the PC version, they had cranked the screen-space reflections and shadows all throughout the game.The poster I was replying to was asking about games running on Series X, not specifically ray-tracing.
The X|S build of Gears 5 does have ray-tracing, not sure if it was in the tech demo shown in March. I didn't have time to rewatch this video.
That didn't happen on the 970 however. As soon as something went above 3.5GB memory use, performance tanked to unplayable.
No, that is not correct. Like I said, not all types of data require the same amount of bandwidth. Obviously new games will require more and more RAM, this is just how things work, but it doesn't have to do with filling up the RAM, only when the 3.5GB of RAM is full and that is not enough to run all of the tasks that require high bandwidth, is when we see a drop in frames.
"In conclusion, we went out of our way to break the GTX 970 and couldn't do so in a single card configuration without hitting compute or bandwidth limitations that hobble gaming performance to unplayable levels. We didn't notice any stutter in any of our more reasonable gaming tests that we didn't also see on the GTX 980, though any artefacts of this kind may not be quite as noticeable on the higher-end card - simply because it's faster. In short, we stand by our original review, and believe that the GTX 970 remains the best buy in the £250 category - in the here and now, at least."
Nvidia GeForce GTX 970 Revisited
When the GTX 970 launched last year, the tech press - Digital Foundry included - were unanimous in their praise for Nvi…www.eurogamer.net
While not all types of memory access require the same bandwidth, the 970 didn't manage the memory access in such a way that the 0.5GB of lower bandwidth memory could be effectively used.
I owned a 970 and analyzed the performance. Whatever testing they said they did, it was trivial to bring a 970 to its knees.
The 970s management was effectively a mismanagement then, because in real world performance that upper RAM crippled the card. Doesn't bode well for Xbox Series... :/
Denoising is not a hardware feature, it's software.
Currently games have used traditional compute path for denoising. (Usually separate denoisers for reflections/shadows/specular/diffuse.. etc.)
We have machine learning denoisers in offline renderers and preview windows of offline renderers.
You cannot ignore reality.
"But so far with this new information we have been unable to break the GTX 970, which means NVIDIA is likely on the right track and the GTX 970 should still be considered as great a card now as it was at launch. In which case what has ultimately changed today is not the GTX 970, but rather our perception of it."
GeForce GTX 970: Correcting The Specs & Exploring Memory Allocation
www.anandtech.com
What the heck is this? Dude I've tested this myself. Can you say the same? You just keep parroting vague reviews.
Will not reply again on this. I have first hand experience testing and observing this slowdown. I have nothing more to say.
No, the full 4GB of RAM could be occupied. This was not necessarily a problem with filling up the RAM, but situations could happen where a game could need more fast RAM than the 3.5GB that is available, only in those situations we would see problems. On a closed system like the XSX memory allocation can be even more efficient than the GTX 970. On PC games are designed to run on hundreds of GPU models with different amounts of RAM. Let's not also derail the main subject here, which is to make clear that the additional bandwidth that the XSX has, will help with Ray Tracing.
"Nvidia's driver automatically prioritises the faster RAM, only encroaching into the slower partition if it absolutely has to. And even then, the firm says that the driver intelligently allocates resources, only shunting low priority data into the slower area of RAM."
Nvidia GeForce GTX 970 Revisited
When the GTX 970 launched last year, the tech press - Digital Foundry included - were unanimous in their praise for Nvi…www.eurogamer.net
Wow sarcasm.Cool, you're a dev with practical hands on experience then, awesome.
Automatic VRAM allocation has been a thing on PC GPU's for a very long time. No one has to worry about "hundreds of different GPUs'" when developing for PC. 🙄
I'm aware of that. I was referring to how when a developer makes a game they have to take into account the different types of RAM sizes and speeds. On PC you also have the option to change the settings to levels that your GPU cannot support at a good framerate. There is no developer out there that will create a game that needs 30GB of GPU RAM for example. The point here is that it's easier to optimize for a console. We can be sure that most developers will be making sure not to go over budget on the high speed XSX RAM and kill the performance by loading a tasks that requires high bandwidth on the slow memory pool.
It has a slightly stronger GPU so it should have better RT but the question is how much better will it actually look in comparisons ?
It would also help if Microsoft showed some next games running on the damn console at this point.
I doubt memory speed is even much of a factor for 99% of devs out there. Bandwidth just translates to performance. PC devs tend to focus on one or two specs, and just build scalability from there.
Also that " We can be sure that most developers will be making sure not to go over budget on the high speed XSX RAM and kill the performance" Is probably not a safe bet. considering the performance issues of most console titles released the past couple of generations. We cna probably count on first party devs not to though.
OK, so I should expect for most developed that are making games for the XSX not care about going over budget with the fast RAM and willingly drop performance by putting tasks that require high bandwidth on the slow portion of RAM?
It won't be willful. It'll just be somehting that happens. The level editor dude wanted that tree there, or the script editor brought in the wrong LOD model for this sequence, etc, etc. Most developers don't have the time to optimize to a level where everything is perfect, even on fixed hardware.
I mean did the From Software guys purposely decide to have so many issue running their darksouls games on consoles? I doubt it.
Tools and engines are getting better at automating this stuff though, so who knows.
It...doesn't have ray tracing so it doesn't count for showing off the Series X ray tracing?
more shaders and more number of CU should put Series X on top.
That's the formula to calculate the giga rays, but this doesn't proof that clock are as important than more CU. We have to wait and see, but most people think CU scale better with RT and even if those would scale the same, Xbox still has a big bandwidth advantage for RT tasks.No ms showed the amd formula for raytracing in hotchip clip.
4 × number of CU × clock of CU
4× 52× 1850=380 giga rays
So u dont need rumor. Both clock and cu number are as important as one another
Minecraft Path Tracing demo.We should prob w8 too talk about this until we see some more (any) games running on XSX hardware.
I don't think we have seen any gameplay so far on XSX with raytracing?
It's actually a misinformed take (like so much around these consoles, especially the PS5), since Ray Tracing scales roughly linearly with resolution, so there would actually be no difference in Ray Tracing quality between XSX and PS5 versions of games despite the performance delta, rather just in resolution.