Can anyone with the Slim and the 'new' DS4 confirm whether the sticks are better than before ?
They're the same. Too low.
Can anyone with the Slim and the 'new' DS4 confirm whether the sticks are better than before ?
Yeah, what about all of the others GPU ?
Whoops, it's like AMD offers a wiiiiide range of GPUs, going from 4 CU to 64 CUs. There's no "AMD made a GPU for PS4, XB1, PS4 Pro and used them for PC". AMD and Nvidia both have scalable architectures which they deploy in a big range of configurations. Sony and Microsoft just shop what they want for their own business.
So an Elite controller would divide the user base more than a Pro console would?
Yeah, what about all of the others GPU ?
Whoops, it's like AMD offers a wiiiiide range of GPUs, going from 4 CU to 64 CUs. There's no "AMD made a GPU for PS4, XB1, PS4 Pro and used them for PC". AMD and Nvidia both have scalable architectures which they deploy in a big range of configurations. Sony and Microsoft just shop what they want for their own business.
And RX480 supports DX12. How about that ?
Weren't you saying PS4 Pro was using two GPUs btw ?? :")
lol...in the case of the PS4P it's a simple fact...that GPU is not a 480...period lol
I highly doubt the Scorpio will be running an off the shelf GPU either. In fact I'd be shocked if it wasn't customized in some way
Sony & AMD work together in creating the GPU & it's not a one & done situation.
I wonder if a few years back AMD actually created the Rx 480 with its odd SP/CU count to coincide with Sony's upcoming PS4's Pro GPU? You see what im wondering? If we look at OG PS4's CU count...and double it, its exactly what the Rx 480 is. Just throwing that out there i guess...but the Rx 480's SP/CU count is exactly double what OG PS4's gpu is. Makes me wonder if Rx 480 was designed with PS4 Pro in mind a few years back.
It's 8.4TF FP16 it also has help rendering geometry & the ID buffer is going to help a lot once devs find more useful things to do with the data that it's providing.
I'm not sure why people are arguing with facts just because they don't get it. 8.4TF is the number of floating point operations per second that PS4 Pro GPU is capable of.
I think it was more in line of Sony looking at AMD's roadmap and picking the right chip that fits their needs.
Obviously. There's no magic chip or some super secret experiment or team work.
AMD has their GCN architecture, they refine it over iterations, and they deploy it from a wide range of cores, from embedded to high end. From 2CU to 64CUs.
Polaris was meant to be a midrange lower powered chipset, the full chip is 36CU. Sony was shopping for a GPU and decided it fitted their needs.
Yeah I know, I was just explaining that to the person I quoted =)
If I count in half-eggs, the carton in my fridge right now actually has 24 eggs in it. Most don't think about the possibilities of half-eggs.
Microsoft said they're releasing a carton of 18 eggs next year, but how do we know they're not talking about half-eggs? I'm just saying.
Hold on. You can't choose both.
PS4 Pro is two GPUs, so it's not RX 480.
RX 480 is PS4 Pro GPU so it's one GPU.
I recall you that PS4 GPU is 20 CU with two disabled.
That RX480 is the full Polaris 10 chip and that it is DX12.
Did you read the interview? Things that Sony help create end up in the AMD road map & things from the AMD road map also end up in the console before they hit the market. this stuff have to be tested they don't just toss it together in one shot & hope that it all work,
Obviously. There's no magic chip or some super secret experiment or team work.
AMD has their GCN architecture, they refine it over iterations, and they deploy it from a wide range of cores, from embedded to high end. From 2CU to 64CUs.
Polaris was meant to be a midrange lower powered chipset, the full chip is 36CU. Sony was shopping for a GPU and decided it fitted their needs.
Yeah I know, I was just explaining that to the person I quoted =)
If I count in half-eggs, the carton in my fridge right now actually has 24 eggs in it. Most don't think about the possibilities of half-eggs.
Microsoft said they're releasing a carton of 18 eggs next year, but how do we know they're not talking about half-eggs? I'm just saying.
In fairness, you guys are completely ignoring what Cerny actually said about the relationship and how it works...
They sit down, decode how many CUs they need and then take a look at the roadmap for various other features they can pick and choose from...such as the features they have pulled from beyond Polaris for the Pro
Cerny talks about how during the design process for the PS4 it was him that pushed hard for the ACE's and asynchronous compute...which eventually AMD incorporated into their discreet GPUs...
These two things aren't incompatible. That's the whole point of semi-custom: Taking an existing product, refine it according to the customer needs.
im sorry but youve gotta be an absolute idiot to believe this
they would get ripped to shreds by the media if they attempted this, imo it would be worse than original plan xbone
Hell, why not market it as an 16.8 trillion 8-bit FPO/s machine!
FP16 is only useful in a (very) limited amount of cases.
im sorry but youve gotta be an absolute idiot to believe this
they would get ripped to shreds by the media if they attempted this, imo it would be worse than original plan xbone
It's 8.4TF FP16 it also has help rendering geometry & the ID buffer is going to help a lot once devs find more useful things to do with the data that it's providing.
I'm not sure why people are arguing with facts just because they don't get it. 8.4TF is the number of floating point operations per second that PS4 Pro GPU is capable of.
So by your logic, there's a very good chance that Scorpio will be able to achieve 12TF then.
That's pretty decent and falls in line with my theory that Microsoft most likely wants Scorpio as the "base model" Xbox for the next decade.
That will all depend on the full specs of the Xbox Scorpio which we don't know yet.
Well, logic would dictate that Scorpio's GPU should at the very least be capable of a similar feature set as the PS4 one yes?
So, theoretically, the Scorpio coukd also use FP16 to help improve performance. So if we assume Microsift are calling it a 6TF based on standard FP32 measurement, then again, according to your very own logic, 12TF should be achieveable.
onQ123 said:*ranting*
If sony and AMD hold the patent on that tech in a console, MS won't be able to use it though will they?
This depends on the contract but double rate FP16 and the advanced working distrubutor are "standard" IP blocks in development by AMD, they will be available for everybody.If sony and AMD hold the patent on that tech in a console, MS won't be able to use it though will they?
Apparently, I need to school you again.
GPU production is based on the same basic process as any other silicon production, but GPUs are unique in that they are made of many components that are repeated numerous times throughout (hence GPU core counts in the thousands).
In fact the differences between the GPU from one year and the GPU from the next commonly comes from two groups. Either transistor node size, architecture, or both.
When transistor node size is reduced the physical size of transistors is shrunk allowing for more transistors in the same space or less power used to perform the same task.
A change in architecture is a change in how transistors are arranged in order to perform their expected tasks.
In the design and testing of new GPUs it's commonplace to shrink your existing architecture to a new node then see what architectural improvements you can make in the newly allotted overarching space.
For a GPU on a new node you essentially just want more of the same results, any further improvements from architecture are just gravy.
Since the components of a GPU are so massively reproduced, it becomes effective to "copy, shrink, & paste" the design a few times.
The PS4P is taking the GPU from the PS4, shrinking the node to 14nm, and "adding polaris+ achitecture". At this point it is obvious that the PS4P GPU is moreso a new GPU than the old one simply "mirrored".
Anyway, about this whole half vs. single precision.
The bottom line is that a 4.2TFLOP 32FP chip could do up to 8.4TFLOP 16FP, wherein one much more versatile 32FP operation is given up for two much simpler 16FP operations. But a chip designed to to do 8.4TFLOP 16FP wouldn't be capable of any 32FP operations which are required to run modern games. Mixed precision is great for reducing constraints, it doesn't magically double performance.
I didn't say use the same GPU. I said have a similar feature set.
This depends on the contract but double rate FP16 and the advanced working distrubutor are "standard" IP blocks in development by AMD, they will be available for everybody.
The ID-Buffer seems to be a custom enhancement with Sony, AMD and Microsoft might not incorporate the same for future products.
With FP16 being part of that set? If so what if thats off the table for MS.
However it works out i'm looking forward to seeing my Pro in action in a few weeks, and also very intrested what Scorpio will physically be able to put out, i'm not sure what a 1.8tf advantage is going to buy extra at 4k ad i'm still thinking it'll be a case of both consoles sharing the same asset detail levels but Pro being 1800p upscaled and Scorpio native.
Whatever transpires it's intresting times in the console world and opefully they can keep each other on their toes, as thats whats best for consumers.
I'm just gonna get an adapter to let me use the Xbox elite controller on my ps4. it's a shame he's outright dismissing it on that shaky reasoning.
Isn't FP16 a standard thing in modern GPUs?
With FP16 being part of that set? If so what if thats off the table for MS.
However it works out i'm looking forward to seeing my Pro in action in a few weeks, and also very intrested what Scorpio will physically be able to put out, i'm not sure what a 1.8tf advantage is going to buy extra at 4k ad i'm still thinking it'll be a case of both consoles sharing the same asset detail levels but Pro being 1800p upscaled and Scorpio native.
Whatever transpires it's intresting times in the console world and opefully they can keep each other on their toes, as thats whats best for consumers.
I don't think it'll just be a ~2 tflop advantage for the gpu. If it has 12gb of 320 GB/s ram with a new modern CPU
the difference will be large enough that devs will do things on a different level
So it is basically MS's new gen in all but name?
Pretty much...they say no one left behind, but they would have done something closer to PS4 if that was the case.So it is basically MS's new gen in all but name?
This depends on the implementation, it's a jungle out there.Modern GPU's have needed their own die space for FP16 previously haven't they?
This is a unified thing from what i can make out, so each one can run either one FP32 or two FP16.
This depends on the implementation, it's a jungle out there.
There are separate FP16 and FP32 ALUs and mixed-precision ALU designs.
Some ALUs are capable of one FP16 operation or one FP32 operation, some are always using the same space for the data storage, some only need the half if you use FP16 and some can issue two FP16 operations at once, instead of one FP32.
There are always some compromises you have to take on die-space, power-efficiency and so on.
This guy on another forum is going crazy now.. after I explained to him that PS4 Pro has this mixed double rate FP16 feature.. he blew up saying that Cerny doesn't know anything... he's being paid to say stupid shit.. He's saying that FP16 'doesn't exist in any meaningful capacity'.. saying that game engines need to be FP16 and that it wont work. Says this feature will mean nothing to PS4 Pro. Then proceeds to go on about how Cerny hasn't been relevant since 2003 and knows nothing about programming, nothing about designing consoles, and if he did, Knack wouldn't have been one of the least technically impressive games of the PS4 launch... LMAO
I think I hit a nerve, because I told him a long while back that Scorpio would have features that desktop Polaris didn't have currently. He got mad. :/
I'm just gonna get an adapter to let me use the Xbox elite controller on my ps4. it's a shame he's outright dismissing it on that shaky reasoning.
According to who?
I have a slightly different perspective of this from most developers, as I work on mobile games, but I'd argue the exact opposite of what you are saying. Precision is a fairly complex topic, but it suffices to say that some calculations and some techniques require higher precisions than others. Sometimes FP32 is not really enough, very often it is more than enough. It is used more because it's a good compromise, because it is sufficient for 99% of things, not because it is necessary for even 50% of things. I write shaders for a living which utilise not just FP16 but even lower precisions.
Open GL ES, the main API used for iOS and Android development (Metal and Vulkan are starting to take over now) specifies lowp floats that are typically implemented using just 8 bits. The API doesn't specify this and the manufacturers don't implement it for a laugh, they do it because there are massive performance gains to make, something as important for mobile gaming as it is for high end PC and console gaming.
-
Having said all of that, it is also completely silly to suggest the machine is at 8.4 TFLOPs for the simple reason that the name float, without clarification, should always be taken to mean a 32-bit float. FP16 doesn't change the fact that PS4 Pro has a 4.2TF GPU, but it should be a significant upgrade, and certainly not one that should be dismissed as irrelevant (in all honesty, I'm tending to assume that those dismissing this have absolutely no idea what they're talking about).
Thanks for this post, I learnt something today.According to who?
I have a slightly different perspective of this from most developers, as I work on mobile games, but I'd argue the exact opposite of what you are saying. Precision is a fairly complex topic, but it suffices to say that some calculations and some techniques require higher precisions than others. Sometimes FP32 is not really enough, very often it is more than enough. It is used more because it's a good compromise, because it is sufficient for 99% of things, not because it is necessary for even 50% of things. I write shaders for a living which utilise not just FP16 but even lower precisions.
Open GL ES, the main API used for iOS and Android development (Metal and Vulkan are starting to take over now) specifies lowp floats that are typically implemented using just 8 bits. The API doesn't specify this and the manufacturers don't implement it for a laugh, they do it because there are massive performance gains to make, something as important for mobile gaming as it is for high end PC and console gaming.
-
Having said all of that, it is also completely silly to suggest the machine is at 8.4 TFLOPs for the simple reason that the name float, without clarification, should always be taken to mean a 32-bit float. FP16 doesn't change the fact that PS4 Pro has a 4.2TF GPU, but it should be a significant upgrade, and certainly not one that should be dismissed as irrelevant (in all honesty, I'm tending to assume that those dismissing this have absolutely no idea what they're talking about).
So it is basically MS's new gen in all but name?
What? Cheap 3rd party controller's don't have extra buttons To give you an advantage.
You don't have to take your thumbs off the analog. If people feel they need a advantage cool for them. I personally don't want any edge over the competition.
that 8.4tflops sounds like some misterxmedia tales