playsaves3
Banned
799 isn’t happening it’s 699 max reason why is cause it’s being sold without a disc drive
-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
PS5 Pro Specs Leak are Real, Releasing Holiday 2024(Insider Gaming)
- Thread starter jm89
- Start date
Quantum253
Gold Member
That would make sense to follow the slim line. So an over/under of $50 649/699. Or 599/649. I just think 100 difference is low for the hardware costs
Topher
Identifies as young
I can’t see it without a drive. And I won’t buy it without one either.
playsaves3
Banned
It will support the detachable drive the slim uses
Topher
Identifies as young
Hopefully
playsaves3
Banned
Oh no I agree below 599 is also delusion I agree with your prices by the way
DevilHunterX0
Member
I would pay "Tree Fiddy"
James Sawyer Ford
Member
It’ll be 549 or 599
They won’t go above 599
They won’t go above 599
bitbydeath
Member
Not even in the ballpark, it’s the same rubbish others tried pushing here. The base is 33.5TF, if dual issue is used it goes up.
Imagine thinking AMD would add a feature that halves their card’s performance with the possibility of gaining some back.
Last edited:
Celcius
°Temp. member
It also won't be cheaper than $599
StreetsofBeige
Gold Member
So 67TF FP16, 33TF FP32, so what's the cut in half 16-17TF people had been talking about lately? Thats a different measurement?
bender
What time is it?
PS4 Pro was north of 10% of the PS4 sales. Cutting off BC presents technical challenges (PSN Store which would be somewhat trivial) but you'd also have a messaging issue of selling a more technically capable device that offers less functionality. Sony did that with the PS3 but that was explainable as a cost savings measure and there was no benefit to PS3 games on any model. Further, I think PS4 functionality is important even when PS6 hits. Even if a fraction of the userbase is still around, that could still be 20 million customers. That userbase would be useful to indie devs, niche game developers, or developers targeting the Switch 2 not to mention having that back catalogue could bolster PS+ offerings. We still get a ton of PS4 releases that have no native PS5 version. I know I'd barely use my PS5 if t weren't for BC. Maybe I'm alone in that usage but Sony has that data. I'm not saying that BC is trivial architecturally, but it seems like a challenge that can be overcome and has a ton of benefits.
playsaves3
Banned
Don’t believe the cpu leaks right now
bender
What time is it?
At least he learned well as a marketing shill. The psychological difference between seeing 599 and 600 is huge.
playsaves3
Banned
It’s funny we are discussing both the 7800 xt and 6700xt since that is basically the difference here
ReBurn
Gold Member
Probably double precision floating point math. That uses 64 bits.
SlimySnake
Flashless at the Golden Globes
PS4, PS4 Pro, PS5, X1, X1X, XSX and XSS were all based on tflops calculations that assumed 2 instructions per clock. With RDNA3, AMD now claims that they can run 4 instructions per clock. That allows them to claim DOUBLE the theoretical flops, but in reality, they were not able to double the performance.
So that 33 tflops number is nonsense and the real number is the 16 tflops number which you can compare against the PS5 tflops to get an idea of its actual performance. If it was 33 RDNA2 tflops, you would be getting 3.3x the performance, but in reality, you will get 1.6x at best.
With that said, thanks to the extra ray tracing power in the PS5, it will perform way better than a 16 tflops RDNA2 GPU in ray traced games. Probably something like a 23 tflops 6900xt.
SportsFan581
Member
Maybe they can deactivate a shader engine and run at PS5 base clocks or run the full chip at a slightly lower clock? Maybe a consideration for the capabilities of the cooler?
Or maybe such a small difference doesn't even change anything to begin with.
bitbydeath
Member
Not even in the ballpark, it’s the same rubbish others tried pushing here. The base is 33.5TF, if dual issue is used it goes up.
Imagine thinking AMD would add a feature that halves their card’s performance with the possibility of gaining some back.
playsaves3
Banned
These people are hating you but I was expecting 60-80% raster improvement not 45% maybe the leak is wrong though or clock speeds aren’t finalized so let’s have some hope
SportsFan581
Member
That might just be a general thing for console hardware going forward now, at least in the first few years. This long cross-gen phase won't be forgotten and is sure to alter some buying decisions because some people are now going to have the mindset that it might be a loooong time before any new hardware gets properly utilized. But, new players are always coming in and for them it is the difference between spending X on base PS5 or X on the Pro. We'll have to see what it offers vs. what it costs.
This is the main reason I don't think there will be a lot of cross-gen with the upcoming Switch. I think they will move entirely to that platform with their key software right out of the gate. Just because I think they have confidence in their ability to get crazy attach rates at launch and they trust their software to sell over the long-term.
Last edited:
SlimySnake
Flashless at the Golden Globes
Alex is already in pain. He's downplaying the 2-4x improvements in ray tracing as just theoretical that wont show up in the final performance gains. Sony simply said the speed is 2-4x faster, but he's started planting seeds of doubt on whether or not it will translate into actual performance. Classic skepticism from the same guy who claimed PS5 wont have dedicated RT hardware and then after cerny confirmed it was dedicated, he started spouting it would only support rt shadows.
Ashamam
Member
Even when they have the actual machine in hand they don't really give full credit to the holistic design. (referring original PS5 and console bespokeness in general). They had an entire Cerny presentation to refer to and pretty much proceeded to view the PS5 in purely PC terms. I don't expect DF to be particularly insightful with just a handful of leaked specs. Maybe they could touch on what a 2ms dedicated hardware upscaler brings to the table as compared to FSR2. I don't think they have ever shown the budget for FSR2 on PS5 hardware have they? For example if you offload to PSSR do you just gain the differential in ms, or can you use the resources during the old FSR window to be doing async stuff that runs outside of the frame budget.
The Pro is going to be a complicated beast to analyse, lots of pluses, but if the CPU/memory is close to the PS5, lots of potential bottlenecks that could kneecap some of the GPU/ML advances.
Anyway, I appreciate DF for the basic analysis but they rarely go particularly deep, I think its because they can't, just watch John when he is interviewing actual developers or mentions he has booted up UE5 engine as if that makes him knowledgeable. Alex is better but there is a still a LOT of hand waving at points about being unsure of reason etc and quickly moving on. Obviously they are journalists, but I think it would be beneficial if they could do more with actual dev guests. That way there could be a bit of critical feedback to DF which at times I think is sorely needed.
Ashamam
Member
To be fair hes not wrong in this case. Its obvious to everyone here I guess that the fps isn't going to jump by 2-3x, that would be ridiculous, but some people might see it that way.
Audiophile
Member
FP16 Half Precision = 67TF
FP32 Single Precision Dual-Issue = 33.5TF
FP32 Single Precision = 16.75TF
FP64 Double Precision = ~1TF
AMD started reporting/advertising the Dual-Issue FP32 as their standard Teraflop number on RDNA3.
Dual-Issue only applies to specific operations/workloads which can be optimised for it.
The base PS5's FP32 Single Precision compute is 10.28TF and the rumoured PS5 Pro's is 16.75TF; so that's the fair baseline comparison at this point (while also taking into account the architectural efficiency gains of RDNA3/4 over RDNA2).
It's hard to say at this point roughly what proportion of real world, gaming workloads will be able to run with Dual-Issue and to what degree devs may optimise for it (or if Sony can have the system automatically run conducive operations in Dual-Issue).
If, purely for eg. an average of a little over 10% of workloads can run in Dual-Issue then that'd give an equivalent of 15TF of Single-Issue FP32 + 3.5TF of Dual-Issue for 18.5TF. Still, these are purely theoretical as with any flop number, in reality a Dual-Issue Add operation might be 80% faster or a Multiply-Add might be 20% faster; rather than 100% faster.
One thing' for sure, this isn't a case of 10.3TF vs 33.5TF and a subsequent ~3.3x power increase in compute.
Last edited:
James Sawyer Ford
Member
Who is claiming 2-3x better frames?
It should be more stability of frames (less tanking) and/or higher quality of the RT itself (higher rez or more ubiquitous)
Last edited:
Self_Destructive
Member
He's not downplaying anything... he's simply stating facts... lmao wtf.Alex is already in pain. He's downplaying the 2-4x improvements in ray tracing as just theoretical that wont show up in the final performance gains. Sony simply said the speed is 2-4x faster, but he's started planting seeds of doubt on whether or not it will translate into actual performance. Classic skepticism from the same guy who claimed PS5 wont have dedicated RT hardware and then after cerny confirmed it was dedicated, he started spouting it would only support rt shadows.
"RT performance" is not "framerate"....
Bojji
Member
It's hard to find good info about this. Ada has massive increase in tops number compared to ampere but it doesn't translate to real world gaming performance in any measurable way (it's mostly used for DLSS anyway).
End performance will be a sum of both raster and RT calculations. RT part might run 2 or 4x faster but raster won't go above 63%, end result vs. PS5 will depend on how much RT calculations were used compared to raster.
Rocco Schiavone
Member
Otherwise it doesn't make sense if the upgraders can't play their physical PS5 games on the Pro, right??
About the price it will be 599€ with the option to add a drive.
zeroluck
Member
Per Tensor core perf has remained the same going from Turing to Ada, per RT core perf saw 50% bump(28% from Turing to Ampere, 18% Ampere to Ada)
Last edited:
Self_Destructive
Member
There are people out there acting like with RT they're expecting those kinds of multipliers to the overall framerate in RT games..
Higher resolutions require.... more rays per frame! Meaning that with increased resolutions the demand of ray tracing also increases. So right there you're losing a lot of that "RT performance" to simply having to trace rays in a higher resolution scene. Then if you want to actually improve and add more RT to the game.. you're adding even more strain to it.
So that 2x-3x improvement basically becomes a relatively modest bump in framerates when it's all said and done.
Last edited:
Codeblew
Member
Yeah, this guy has a smug, punchable face but he isn't wrong this time.
Audiophile
Member
Would be surprised if they managed to get a reasonably sized chip if they did go for N6 again. Perhaps they've done the calculations and found a much bigger N6 chip + additional cooling provisions is still cheaper than a smaller N4P chip.
Thinking about it this could be where Zen 2 and the lower GPU clock come in and start making sense.
If they were going to go with the N5/N4P family then Zen4/4c would make a load more sense, you're already building a distinct, new chip and why would you bother to port Zen 2's topography over to a new process node when Zen 4 already exists on it; and would likely be much easier to integrate and could perform much better at lower clocks?
As for 2180MHz GPU clock, that could suggest a hard limit for the bigger chip (though at that point I'd think it would have made more sense to go down to 54/60CU and get back that extra 53MHz just to simplify BC).
Perhaps it came down to:
N6 + Zen 2 CPU + 2180MHz GPU + heavy cooling = expensive
N5/N4P + Zen 4C CPU + ~2500MHZ GPU + moderate cooling = a little more expensive (due to node cost)
I still think the latter would be more elegant if they could do everything to keep that die as small as possible. Go for a curve ball and drop to a 192-Bit bus but use G6X or the absolute lowest binning GDDR7 cast offs. Keep on-die cache small but stack 3D V-Cache etc. lol As well as of course, 54/60CU rather than 60/64CU (though I'm unsure if the additional Shader Engine logic may offset the die space reduction from 4 less CUs).
Then again, after going through this. The RDNA3 chips this will likely be based on will be on N5 and (and RDNA4 on N4P or N3). So they'll be porting that topography up to N6 which seems like more palaver.
Thinking about it this could be where Zen 2 and the lower GPU clock come in and start making sense.
If they were going to go with the N5/N4P family then Zen4/4c would make a load more sense, you're already building a distinct, new chip and why would you bother to port Zen 2's topography over to a new process node when Zen 4 already exists on it; and would likely be much easier to integrate and could perform much better at lower clocks?
As for 2180MHz GPU clock, that could suggest a hard limit for the bigger chip (though at that point I'd think it would have made more sense to go down to 54/60CU and get back that extra 53MHz just to simplify BC).
Perhaps it came down to:
N6 + Zen 2 CPU + 2180MHz GPU + heavy cooling = expensive
N5/N4P + Zen 4C CPU + ~2500MHZ GPU + moderate cooling = a little more expensive (due to node cost)
I still think the latter would be more elegant if they could do everything to keep that die as small as possible. Go for a curve ball and drop to a 192-Bit bus but use G6X or the absolute lowest binning GDDR7 cast offs. Keep on-die cache small but stack 3D V-Cache etc. lol As well as of course, 54/60CU rather than 60/64CU (though I'm unsure if the additional Shader Engine logic may offset the die space reduction from 4 less CUs).
Then again, after going through this. The RDNA3 chips this will likely be based on will be on N5 and (and RDNA4 on N4P or N3). So they'll be porting that topography up to N6 which seems like more palaver.
Last edited:
Mr.Phoenix
Member
Just tossed away 0.75TF like it was nothing. Funny thing is if he was even trying to round down it should be 16.5TF. If rounding to the nearest whole number, it definitely should be 17TF. Guess fo his agenda 16Tf just sounds better.
On another note, I don't know how I feel about this 33.5TF being plastered about. I did call it though that it would be a 16-17Tf machine and they would be talking up 30TF+ due to the nonsense Nvidia and AMD has started doing with regards to the whole dual issue compute thing.
My issue with that is that we are yet to see a single GPU that supports this whole dual issue thing show a clear performance advantage over GPUs that don't have it. So unless we are hoping or expecting that somehow when it comes to the S, devs would fully utilize it, I think it would be more practical (and maybe even honest) to just call this thing a 17TF machine. And if the RT stuff is true, then we should expect it to perform like a 17TF GPU from Nvidia. Or whatever GPU that fits that power gap from AMD based on the architecture the PS5pro uses.
So expect performance to land somewhere in the 3080 - 4070 super range.
asdasdasdbb
Member
The Jaguar BC mode, lol.
Actually... I think it's more likely that while Zen 4 is small, their custom Zen 2 is even smaller. I agree it's a mistake.
SlimySnake
Flashless at the Golden Globes
The more efficient the GPU is at doing ray tracing the more rendering performance it will be able to offer thus resulting in higher fps or more pixels. It is literally why the Nvidia GPUs offer more frames than equivalent AMD GPUs because they are able to process ray tracing faster.
And its not like the PS5 doesnt support ray tracing. So its not like they are starting at zero. The Pro has 60% more tflops and offers roughly 45% more performance. So right off the bat, they will get 1.5x more frames. If the ray tracing hardware inside each CU is improved, that means it will affect not just those extra 60% tflops, but also the existing 10 tflops in the PS5. You will easily get 2x more frames in ray traced games.
Alex is just a downer when it comes to this stuff. He claimed for the longest time that the console could only do rt shadows. Turns out the PS5 can do rt reflections, rt shadows, RTGI, rt translucent effects. Forget Avatar, look at what callisto was doing on consoles. They werent just using reflections for metals and reflective surfaces, they were able to give even stone and wooden objects reflective properties that mimicked rtgi bounce lighting in a way. on top of reflections and shadows, they had translucent reflections in each zombie's skin giving every enemy this gory glossy look you would see in cg movies.
The RDNA 2 rt hardware can do anything the nvidia hardware could do, just slower. It came at a cost and that either reflected in resolutions or frames. That cost is now going to be on par with whatever it costs nvidia to run rt effects on their GPUs.
So yeah, if they have made it 2x more efficient and added 60% more gpu power, you can easily get 100% more frames or 100% more pixels because those first 10 tlfops are also running ray tracing at higher speeds.
P.S Unless of course they become cpu bound but im assuming they have given the cpu a decent clockspeed boost. They gave the PS4 Pro CPU a 30% clock speed boost, if they do the same here, we will get a 4.5 ghz CPU that should remove most bottlenecks.
StreetsofBeige
Gold Member
Cool thanks.
I remember 16-17TF more, but then when this 33TF posts started coming out (I dont read or follow these PS5 Pro threads closely) I thought it might be now triple the power of PS5 at 33.
Mr.Phoenix
Member
To play devil's advocate here even though I agree with the point you are making)
The rumour suggests 3x RT performance compared to the PS5. That comparison is all the context that is really needed if looking at it from a purely technical point of view. What that translates to is that in situations where the PS5 RT pipeline can take up 6ms in a frame, the PS5por would complete the same task in 2ms as an example.
We simply can't be picking and choosing what part of the spec leaks we want to get overly anal about, because at the end of the day the system should be taken in its entirety and not singling out any one part. Eg. Whatever version of RDNA its using would be better than what was in the og PS5, or AI accelerated reconstruction could not just yield better IQ results, and still be faster compared to whatever the were doing on the PS5. Mem bandwidth would be higher too...etc.
The point is, it doesn't matter which part(s) of the PS RT pipeline is faster, when the reference comparison is that it has RT that is 3-4x faster than the og PS5. Only question is, do they mean 3-4x faster than an OG PS5 with a 10TF GPU? Or 3-4 times faster than the og PS5 with a 16.7Tf GPU? The latter would mean that the actual RT improvement compared to the og PS5 would be more like 2.xx or so. Oh... and better RT performance does directly translate to better framerates.
Last edited:
Fake
Member
Probably monday as their DF weelky show happen.
Unless they can't wait to discuss and release sunday.
Bulletbrain
Member
For most people the discussion doesn't need to be so complex.
A current ps5 raster game running at average 30fps - expected pro performance average 43fps (based on 0 45x performance uplift.
A current ps5 game with full RT features running at average 30fps - expected pro performance average XXFps? certainly it won't be 60-90fps all of a sudden. I believe the ceiling is that 0.45 rendering uplift, logically speaking. So it will run it somewhere between 30-43fps on average.
SlimySnake
Flashless at the Golden Globes
Yes
No. The RT performance of the ENTIRE GPU has improved. Not just the 45%. There are still 10 tflops on the PS5 that are also going to get these IPC gains. Just like how the PS5 gained the 50% IPC gains going from the base PS4's GCN architecture to Pro's Polaris architecture and finally the RDNA1 IPC gains. Thats why the 10 tflops PS5 behaves more like a 15 tflops PS4 when running games built on the PS5 sdk with all those IPC gains.
The same thing will happen here. The entire GPU will get the 2x IPC gains which in turn will help you get those extra frames. I think 90 is pushing it, but you will get 60 fps and then they will use framegen to get you those 30 extra fake frames everyone loves so much on PC. They have smartly invested enough in those AI accelerators to do what nvidia does in the 4000 series GPUs.
Ashamam
Member
The real benefit here will be in the performance mode. A higher internal resolution, locked 60 and a better upscaling solution should result in a less comprimised experience. The RT improvements I think will tie in more to performance modes than pushing the quality modes IQ. But thats because I think the entire focus of the machine is around 60fps rather than pushing visuals at 30fps quality mode. So if I'm wrong, who knows.
bender
What time is it?
If that's true, it will be interesting to see how the market responds to this advertising push, but maybe it's just me who thinks how this will be advertised is the most interesting facet.
James Sawyer Ford
Member
That's false.
The rendering uplift has nothing to do with RT. They are two separate metrics.
If you are only getting 45% more performance in frames, you could possibly boost the quality of the RT in the game by a much more meaningful degree.
64bitmodels
Reverse groomer.
there's less reason? you'd be getting 1500/2000 dollar PC performance for 700-800. that sounds like a pretty large reason for me.
64bitmodels
Reverse groomer.
btw do people here really think the ps5 pro will be 600 bucks?
700 at the very least. the current ps5 is 50 bucks away from the pro, not a big enough price jump for the specs we are getting here
700 at the very least. the current ps5 is 50 bucks away from the pro, not a big enough price jump for the specs we are getting here
James Sawyer Ford
Member
The PS5 Digital is $450
Many of the components will not really be increasing much in price, since they are using roughly the same memory, power, cpu as the base version. The only thing that's really getting expanded is the GPU.
I think it's reasonable to expect that the increase in cost from a BOM perspective is probably less than $100 versus the basic PS5. So add a margin on top of that and you get $599 being reasonable. I don't think they will go higher than that. And it's also possible that they make a decent margin at $549.
I think if we look back at the PS4 vs PS4 Pro, the difference in price was roughly $100, wasn't it? $399 vs $299
bender
What time is it?
I think it will be $100 over the current slim models. 4 years later and $100-150 (or I guess $50-200 depending what disc path you are going) dollars over the launch units feels about right.
Last edited:
dano1
A Sheep
I would guess they are going to do direct selling like they did with the PSVR2 for the first month. At least I hope so. Scalpers are just plain scum.