• Hey Guest. Check out your NeoGAF Wrapped 2025 results here!

[MLiD] Project Helix is 25 Percent Faster Than PS6

How did having the "world's most powerful console" work out for ms any other time? It'll be different this time? Somehow? Because it'll be a 1500 dollar pc?
 
The NPU can't possibly have lower latency if being used in any way for doing anything that has to do with graphic rendering. Like that's physically impossible, especially when comparing it to neural arrays that are inline or integrated within a GPU compute unit.
The NPU has much lower latency, so as I said, it's not used for graphics rendering. However, it's still part of the overall unit and does a number of unique computational tasks that it takes over from the GPU that aren't directly related to graphics. The NPU will be key for local language models.
 
We already had 3-4x times more powerful in RT for PS5 Pro vs. PS5 and this turned out to be bullshit in the real world.

For sure it doesn't translate to 3-4x higher FPS, so similar situation will be true for next gen consoles when we know what their general power will be. K KeplerL2 is right obviously.
MLiD is not wrong, when you add all the new hardware capabilities plus PSSR with neural rendering, the PS6 can do 5090 performance.

All you guys ever talk about is rasterization to gauge the PS6 performance, even though Mark Cerny already cleared that up in his PS5 Pro tech seminar.

"First, there's rasterized rendering, by which I mean the conventional rendering strategies that were all we had up through PS4 Pro or so. There's not a whole lot of growth left here. It mostly has to come from making the GPU bigger or memory faster.

Ray tracing is different. It's still early days for the technology, and I suspect we're in for several quantum leaps in performance over the next decade.

Machine learning, though, has the greatest potential for growth, and that's an area we're beginning to focus on."

The PS6 will rely more on RT and AI hardware than the previous consoles. I'm surprised that Kelper doesn't understand this but still leans on raster gauge performance.
 
MLiD is not wrong, when you add all the new hardware capabilities plus PSSR with neural rendering, the PS6 can do 5090 performance.

All you guys ever talk about is rasterization to gauge the PS6 performance, even though Mark Cerny already cleared that up in his PS5 Pro tech seminar.

"First, there's rasterized rendering, by which I mean the conventional rendering strategies that were all we had up through PS4 Pro or so. There's not a whole lot of growth left here. It mostly has to come from making the GPU bigger or memory faster.

Ray tracing is different. It's still early days for the technology, and I suspect we're in for several quantum leaps in performance over the next decade.

Machine learning, though, has the greatest potential for growth, and that's an area we're beginning to focus on."


The PS6 will rely more on RT and AI hardware than the previous consoles. I'm surprised that Kelper doesn't understand this but still leans on raster gauge performance.

So it's the same as 5070 = 4090? Am I right?
 
MLiD is not wrong, when you add all the new hardware capabilities plus PSSR with neural rendering, the PS6 can do 5090 performance.

All you guys ever talk about is rasterization to gauge the PS6 performance, even though Mark Cerny already cleared that up in his PS5 Pro tech seminar.

"First, there's rasterized rendering, by which I mean the conventional rendering strategies that were all we had up through PS4 Pro or so. There's not a whole lot of growth left here. It mostly has to come from making the GPU bigger or memory faster.

Ray tracing is different. It's still early days for the technology, and I suspect we're in for several quantum leaps in performance over the next decade.

Machine learning, though, has the greatest potential for growth, and that's an area we're beginning to focus on."


The PS6 will rely more on RT and AI hardware than the previous consoles. I'm surprised that Kelper doesn't understand this but still leans on raster gauge performance.
You're gonna be really disappointed by the PS6 then
 
MLiD is not wrong, when you add all the new hardware capabilities plus PSSR with neural rendering, the PS6 can do 5090 performance.

All you guys ever talk about is rasterization to gauge the PS6 performance, even though Mark Cerny already cleared that up in his PS5 Pro tech seminar.

"First, there's rasterized rendering, by which I mean the conventional rendering strategies that were all we had up through PS4 Pro or so. There's not a whole lot of growth left here. It mostly has to come from making the GPU bigger or memory faster.

Ray tracing is different. It's still early days for the technology, and I suspect we're in for several quantum leaps in performance over the next decade.

Machine learning, though, has the greatest potential for growth, and that's an area we're beginning to focus on."


The PS6 will rely more on RT and AI hardware than the previous consoles. I'm surprised that Kelper doesn't understand this but still leans on raster gauge performance.

you're acting like the 5090 is only using standard rasterising here too tho...

if the PS6 can "reach 5090" performance with the help of AI, then what do you think the 5090 does once you use its superior ML hardware?
it's like you saying that a VW Beetle, with the help of adding nitrous injection, can keep up with a Ferrari that's running in first gear... without acknowledging that the Ferrari can just shift up and leave it in the dust.

also do you even know the sheer amount of RT cores the 5090 has? that thing is insane. it has TWICE as many RT cores as the 5080, which itself is already beating any AMD card on the market by a huge margin in both ML and RT workloads... now double that for each, then you have a 5090.

the 5090 is so far ahead of current gen consoles that it will probably be able to play AAA games at high settings until 2035 or longer.
 
Last edited:
The NPU has much lower latency, so as I said, it's not used for graphics rendering. However, it's still part of the overall unit and does a number of unique computational tasks that it takes over from the GPU that aren't directly related to graphics. The NPU will be key for local language models.
Yes, part of the overall unit, but not used for graphic rendering stuff. Outside, co-pilot stuff, I can't even imagine how it can be used for anything frame-time related, as in that case, you will see how much latency it throws in. Well, it could be used for NTBC decompression, I guess. But so can the neural arrays, though I can see a scenario in this one case that have a separate NPU can be beneficial for NTBC data streaming.

But when it comes to actually rendering a frame, and all the things associated with that, be that denoising, ML upscaling, FG...etc, using the NPU for stuff like that will only add latency and thus, a significant increase in frametime.
MLiD is not wrong, when you add all the new hardware capabilities plus PSSR with neural rendering, the PS6 can do 5090 performance.

All you guys ever talk about is rasterization to gauge the PS6 performance, even though Mark Cerny already cleared that up in his PS5 Pro tech seminar.

"First, there's rasterized rendering, by which I mean the conventional rendering strategies that were all we had up through PS4 Pro or so. There's not a whole lot of growth left here. It mostly has to come from making the GPU bigger or memory faster.

Ray tracing is different. It's still early days for the technology, and I suspect we're in for several quantum leaps in performance over the next decade.

Machine learning, though, has the greatest potential for growth, and that's an area we're beginning to focus on."


The PS6 will rely more on RT and AI hardware than the previous consoles. I'm surprised that Kelper doesn't understand this but still leans on raster gauge performance.
5090 performance? Not possible, and that's not just talking about raster, its not possible by any metric... hell, the PS6 can't even do 4090 performance, and the 5090 is over 30% faster than the 4090.

The PS6 will land between a 5070ti and 5080. And that's still plenty and ok.
 
You're gonna be really disappointed by the PS6 then
Replying to your comment due to what you said about helix (a while ago) that regardless of its price it would be the device one would want. Minus this being an enthusiast forum and a price floor no one can predict; are there things about helix that has not been revealed yet?
 
MLiD is not wrong, when you add all the new hardware capabilities plus PSSR with neural rendering, the PS6 can do 5090 performance.
I'm a console peasant and even I'm finding this ridiculous.

5090 can do all that neural rendering too. So if we are talking future potential, it has ~3300 FP4 AI TOPS compared to the ~2400 of PS6. You could potentially get more juice out of the neural array architecture though, so may be they will be comparable in real world AI performance. But that remains to be seen.

So even if you grant that PT and AI comes within range, shading performance, fill rate, bandwidth etc are far, far behind. Which makes the comparison plain stupid as PT +AI does not equal the final game render. Rasterization/shading still plays a big part in the pipeline.

The 5080 is a good benchmark to beat for a next gen console. The 5090 would still be good step ahead in final rendering performance even in a "built for PT from the ground up" game. Whatever improvements occur in PT will likely benefit 5090 too. For example, Nvidia just published a paper to accelerate ReSTIR PT by 2x to 3x purely with algorithmic optimizations. They are presenting it next month. Every card that can do PT will benefit from it, including RDNA 5. Not a single shipped game uses ReSTIR PT yet, but that's just a matter of time when it improves in such leaps and bounds.

May be some of that gap in raster can be filled with the speculated dual issue improvements, but even that can only take you so far.

K KeplerL2 clearly understands this. MLiD probably understands it too. But he put that stupid comparison out there to sensationalize and has dug his heels in too deep to walk back. He's twisting himself up in a pretzel to still sound like he is right. He is not. He should just stay quiet about it and move on if he can't be gracious enough to acknowledge the hyperbole.

It's no different than his "easily saturate 4k 120 fps" nonsense.
 
Last edited:
I think MLID is wrong he's definitely not able to fully interpret alot of what he gets access to. I think he's misunderstanding the 10x+ figures he's seen. But still think PS6 RT will possibly reach 5080 levels if we are to believe most of the RDNA 5 hype. Just dont see it hitting 5090 levels.
 
Last edited:
:messenger_fire::messenger_fire::messenger_fire::messenger_fire:Kepler vs MLiD:messenger_fire::messenger_fire::messenger_fire::messenger_fire:

recap bachelorette GIF


And my money 100% is on K KeplerL2
 
It seems that now that Xbox and its fanboys are dead, PC fanboys have arrived and are even obsessed with "RDNA1"... Always the same frustration with the "victorious enemy" wanting to "balance things" by diminishing those who are on top.

Do you know why the PS6 won't have ray tracing equivalent to a 5090?

Because whoever says that won't have a video card equivalent to that in their PC when the PS6 comes out.
 
I find these discussions about performance increasingly tiresome and ridiculous.

With a console, the CPU and GPU are crammed into one unit.

You can't expect miracles in terms of performance.

When the PS6 is released, you shouldn't expect anything close to 5090 performance.

Performance-wise, you can expect a significantly faster CPU and a souped-up 9070XT. Path tracing at 30fps.

More isn't technically possible, even with the 4nm process, when you consider the TDP, power consumption, and the size of the chassis. It's complete nonsense to think that you can even come close to the performance of a dedicated GPU and CPU with an APU.

If we're honest, we're always one to almost two generations behind.
 
I find these discussions about performance increasingly tiresome and ridiculous.

With a console, the CPU and GPU are crammed into one unit.

You can't expect miracles in terms of performance.

When the PS6 is released, you shouldn't expect anything close to 5090 performance.

Performance-wise, you can expect a significantly faster CPU and a souped-up 9070XT. Path tracing at 30fps.

More isn't technically possible, even with the 4nm process, when you consider the TDP, power consumption, and the size of the chassis. It's complete nonsense to think that you can even come close to the performance of a dedicated GPU and CPU with an APU.

If we're honest, we're always one to almost two generations behind.
Frankly thats not enough to make a worthwhile ps6 imo. Im going to go all on123 and suggest frame Gen may come to ps5 pro next year. Let ps6 be a true generational leap in 2030. Ps5 Gen has plenty of life left in it. Just need the games
 
Last edited:
I think a helix at 1200 or 1500 could definitely sell 10 million if the rumours are true and an equivalent gaming pc is 2000 to 2500 and that helix plays your steam catalogue etc. absolutely.

If it's literally like an iPhone pro max but for the console world, absolutely.

We are not talking like a pro console we are talking about a genuinely powerful system across cpu GPU here from launch day.
I'm a big fan of the approach, if it matches the leaks, but I don't even see such a device hitting Quest 3 numbers.
 
Last edited:
I find these discussions about performance increasingly tiresome and ridiculous.

With a console, the CPU and GPU are crammed into one unit.

You can't expect miracles in terms of performance.

When the PS6 is released, you shouldn't expect anything close to 5090 performance.

Performance-wise, you can expect a significantly faster CPU and a souped-up 9070XT. Path tracing at 30fps.

More isn't technically possible, even with the 4nm process, when you consider the TDP, power consumption, and the size of the chassis. It's complete nonsense to think that you can even come close to the performance of a dedicated GPU and CPU with an APU.

If we're honest, we're always one to almost two generations behind.
Well said, but for the being generations behind part.

The real issue is that we are just in a very loud echo chamber all across the industry, and the people making the most noise are usually people who don't really know what they are talking about, aren't taking the time to at least learn, or are flat out delusional.

But on the generations thing, are we really though? I mean, look at PCs... the issue here is that people act like the absolute best PC GPU or whatever is the standard by which everything is measured, when even in the PC space that specification is most likely unattainable to the vast majority of PC gamers. Reflected in the fact that less than 0.5% of gamers have one. Even as we speak, over 40% of PC gamers are gaming on something that is equivalent to or less powerful than a PS5. And yet, we would act like somehow consoles are in some sort of stone age. I find it grossly hypocritical.
Frankly thats not enough to make a worthwhile ps6 imo. Im going to go all on123 and suggest frame Gen may come to ps5 pro next year. Let ps6 be a true generational leap in 2030. Ps5 Gen has plenty of life left in it. Just need the games
What will be out there in 2030 that would allow for a "true generational leap"?

I don't know, but PS6 is plenty generational leap... as long as you know what to actually be looking at for the... leap.

I feel the mistake you and some others are still making is that you are looking at TF and raster performance.... when you do that, then it seems like the PS6 is just like 3-4x more powerful than the PS5. But that will be a super big mistake of an assumption to make.

I'll touch on one thing that is a simple example of this.... the fact that the PS6 will have built-in and competent ML hardware, that would allow for things like PSSR, FG, Ray reconstruction...etc, already makes it a generational leap over the PS5. And that would have still been the case even if the PS6 was also a 10TF console.

Its like people forget that once upon a time we were talking about sprites and the number of polygons... those metrics of measuring performance are as redundant now as talking about TFs and raster performance. I keep saying it, and even Cerny has said it too, those things have peaked. But people seem to be ignoring or overlooking the areas that they are supposed to be looking at now, the areas that actually speak to where gaming hardware is going. Eg. People are talking about RAM quantity and bandwidth when what they should be talking about is stuff like NTC.
 
Last edited:
The tradition of consoles has always been a 10x upgrade in power, in all aspects. Now, according to some people, the PS6 can't be 10x more powerful ONLY in ray tracing and AI than the PS5?

For me, that's even a bit underwhelming... I wish the PS6 was 10 times more powerful in AI and ray tracing than the PS5 Pro.

And don't worry, if everything goes well, the AI bubble will burst and Nvidia and AMD will go back to focusing on PC gaming... Then Nvidia GPUs will return to focusing on gaming, unlike now where the focus is on AI and the leftovers are for games.

Even better is that Nvidia will dedicate itself to AI, and when the bubble bursts, AMD will have already caught up with Nvidia in gaming and both will compete on equal footing for a few years...
 
Last edited:
It's 10x faster in RT, but that doesn't mean 10x FPS when using RT because games do a lot more than just RT
So the discussion is more about performance forecasting?

Ah, then forget it... I don't think it's productive to discuss performance expectations... I'm leaving the topic :P
 
I hope PS6 and Helix are powerful enough well balanced consoles that you wont be needing a Pro console in the next 4-5 years after its release. Sony releasing a Pro console 2-3 years after the vanilla console is getting stupid.

Brand new GPU/CPU architecture, Updated SDK toolkit from MS end, enough AI and other stuff which will take a while to tap into. Games should look "next gen" like.
 
Last edited:
I hope PS6 and Helix are powerful enough well balanced consoles that you wont be needing a Pro console in the next 4-5 years after its release.

Not really a need but a want. Sony's said in the past the main reason for Pro models was to keep hardcore fans interested over the long term.

RAM prices would have to subside of course.
 
Last edited:
And it wouldn't be that hard to do?

Shrink to N2P
Up the CU count from 56 to like 80 or more?
Increase the memory bus to 256-bit and use the 4 GB chips. With 36 gpbs GDDR7 instead of 32, that'd be 80% more memory bandwidth and 2 extra GB.
Higher CPU and GPU clocks
Higher System Power
 
I mean AMD wont be just releasing RDNA 5 for Helix/PS6, For Sony they have another contender and its entry level PC gaming with future APU's:

-rumored PC Intel CPU/NVIDIA tile based APU
-PC Zen6/RDNA5 Laptops and desktops
-PC Intel Nova Lake/Battlemage

If MS can get console like UI/gaming on PC laptop/desktop with 1440p-4k with FSR 5.0, DLSS 5.0. XeSS 3.0 (or whatever version), 60-120hz, Raytracing at an affordable price range, it definitely would be enticing.

I still think Helix is just salvaging Xbox console to be a bridge/stepping stone to PC gaming. Taking old xbox games with AI upscaling, mods, and other stuff for replay value is nice, but future next gen games would be fully PC based (i hope cause xbox needs to die)
 
The NPU will not only be able to do things like co-piloting, but also real-time translation into any language and real-time lifelike voice generation from written texts. In addition, the NPU is faster, i.e. has lower latency than the GPU, so the ~120 TOPs will be enough for many interesting things.

The Helix GPU will likely have 1500-2000 TOPs INT8 and 3000-4000 TOPs INT4.
.

Bro we had arcade systems doing this with digital voice synthesizers since the 1980s 🤣
 
I'm a console peasant and even I'm finding this ridiculous.

5090 can do all that neural rendering too. So if we are talking future potential, it has ~3300 FP4 AI TOPS compared to the ~2400 of PS6. You could potentially get more juice out of the neural array architecture though, so may be they will be comparable in real world AI performance. But that remains to be seen.

So even if you grant that PT and AI comes within range, shading performance, fill rate, bandwidth etc are far, far behind. Which makes the comparison plain stupid as PT +AI does not equal the final game render. Rasterization/shading still plays a big part in the pipeline.

The 5080 is a good benchmark to beat for a next gen console. The 5090 would still be good step ahead in final rendering performance even in a "built for PT from the ground up" game. Whatever improvements occur in PT will likely benefit 5090 too. For example, Nvidia just published a paper to accelerate ReSTIR PT by 2x to 3x purely with algorithmic optimizations. They are presenting it next month. Every card that can do PT will benefit from it, including RDNA 5. Not a single shipped game uses ReSTIR PT yet, but that's just a matter of time when it improves in such leaps and bounds.

May be some of that gap in raster can be filled with the speculated dual issue improvements, but even that can only take you so far.

K KeplerL2 clearly understands this. MLiD probably understands it too. But he put that stupid comparison out there to sensationalize and has dug his heels in too deep to walk back. He's twisting himself up in a pretzel to still sound like he is right. He is not. He should just stay quiet about it and move on if he can't be gracious enough to acknowledge the hyperbole.

It's no different than his "easily saturate 4k 120 fps" nonsense.
Take a look at this.
lzPmM1XQcHSvdhob.png

The 7900 XTX (with 96CUs) does 47fps, 9070XT (with 64 CUs) does 57fps, the 5090 does 119 fps, all without upscaling.

You are telling me that PS6 can't achieve 120 fps with better RT Cores and frame generation?

Look at the RT uplift in performance of the 9070XT with only 64CUs, which easily beats the 7900XTX with 96 CUs. RDNA5 Radiance Cores is said to be an even bigger uplift.

We have enough benchmarks out there, like come on guys, this is Tommy Fisher all over again.
 
Take a look at this.
lzPmM1XQcHSvdhob.png

The 7900 XTX (with 96CUs) does 47fps, 9070XT (with 64 CUs) does 57fps, the 5090 does 119 fps, all without upscaling.

You are telling me that PS6 can't achieve 120 fps with better RT Cores and frame generation?

Look at the RT uplift in performance of the 9070XT with only 64CUs, which easily beats the 7900XTX with 96 CUs. RDNA5 Radiance Cores is said to be an even bigger uplift.

We have enough benchmarks out there, like come on guys, this is Tommy Fisher all over again.
The 9070XT would also be 120 FPS with FG in your example? You can't compare one GPU with FG and another without and claim they are the same performance
 
The 9070XT would also be 120 FPS with FG in your example? You can't compare one GPU with FG and another without and claim they are the same performance
So why you don't believe the PS6 with AI can't match the 5090 raw performance?

Or are you jealous of MLiD?

Or leak something new.🤷‍♂️
 
Last edited:
Top Bottom