• 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.

Inside the Scorpio Engine: the processor architecture deep dive

Space_nut

Member
Isn't he just talking about what kind of games you expect Scorpio to run at native 4k?
(Xbox One @4k? More complex games? Less etc)

Bet mostly all the major big games we see at E3 will be 4k native on Scorpio with enhanced 4k assets. Destiny 2, BF2, Cyberpunk, etc.
 
Double rate FP16 won't come close to making up for a ~2TF, higher clock speed, 2.5GB RAM and +100 GB/s bandwidth advantage. The Scorpio will be approx 50% more powerful regardless.
 

onQ123

Member
Why did they include checkerboard, if the gpu is 8.4tf?

Because TF/TFLOPS stand for trillion floating-point operations per second & that does not tell you if the floating-point operations are 32-bit, 24-bit 16-bit & so on.


Nothing can be said to change the fact that PS4 Pro peak performance is 8.4tf fp16.
 

D4Danger

Unconfirmed Member
The stuff about the improved DVR sounded cool. Didn't​ really understand half the other stuff but it sounds like they learned from their mistakes and are just focused on making a powerful box for developers not stuff like kinect.
 
Why did they include checkerboard, if the gpu is 8.4tf?
The PS4 Pro GPU definitely is 8.4TF at FP16, no "if" necessary. The open question is how much game code can use that speed. The answer is definitely not "all", so a comparable 8.4TF will never be achieved. But it's also definitely not "zero", so Pro will be somewhere above 4.2TF compared to Scorpio's 6TF. The array of estimates for effective power we have so far is 4.6TF to 5.6TF; we'll have to see how accurate any of these might be.

The stuff about the improved DVR sounded cool. Didn't​ really understand half the other stuff but it sounds like they learned from their mistakes and are just focused on making a powerful box for developers not stuff like kinect.
Yeah, the DVR sounds very well designed and executed. This is one area where Scorpio will undoubtedly have a very big advantage over Pro.
 

twobear

sputum-flecked apoplexy
Saying 'it's 6TFLOPS FP32 vs 8.4TFLOPS FP16' is frankly a bit meaningless (or misleading, rather) because you're comparing...literally different things. If PS4 Pro is doing calculations with the same precision as Scorpio it can do 4.2TFLOPS vs Scorpio's 6. No matter how you cut it or spin it or obfuscate it, Scorpio is a more powerful machine.

Yes, some of the time the fact that PS4 is using FP16 will result in indistinguishable (though perhaps not identical) results, or very similar results, or whathaveyou. Either way it isn't 'the same'.

This whole argument--especially with nobody with actual development experience with working with FP16 in real-world examples--smacks of 'ESRAM's secret sauce is lower bandwidth' conversations from Xbone's release.
 

Colbert

Banned
Isn't he just talking about what kind of games you expect Scorpio to run at native 4k?
(Xbox One @4k? More complex games? Less etc)

The whole conversation derived from the FP16 double rate feaature not in the Scorpio and that the PS4 Pro 8.4 FP16 TF spec is technically correct. So I assumed the question was meant to be in that area of subject.

Edit:
And seeing no more questions about it, it now seems to be answered.
 

Rodelero

Member
Saying 'it's 6TFLOPS FP32 vs 8.4TFLOPS FP16' is frankly a bit meaningless (or misleading, rather) because you're comparing...literally different things. If PS4 Pro is doing calculations with the same precision as Scorpio it can do 4.2TFLOPS vs Scorpio's 6. No matter how you cut it or spin it or obfuscate it, Scorpio is a more powerful machine.

Is anyone, anywhere in this thread, suggesting that Scorpio isn't more powerful than the PlayStation 4 Pro?
 

twobear

sputum-flecked apoplexy
Is anyone, anywhere in this thread, suggesting that Scorpio isn't more powerful than the PlayStation 4 Pro?

I mean, the pretty direct implication of saying 'it's 6TFLOPS FP32 vs 8.4TFLOPS FP16' is that you are trying to give the impression that PS4 Pro is more powerful or capable. You're trying to directly compare two not wholly comparable specs.
 
Well, he's not saying anything other than PS4 Pro's peak theoretical performance at FP32 is 4.2TF, and at FP16 it's 8.4TF. He's not wrong and not misunderstanding things.

Posting a number that has so far been used to quantify the graphical horsepower of a console or graphics card without the context of what it actually means is deliberately misleading.

If I was trying to mislead people I would post the numbers without context. each & every time that I have posted about PS4 Pro peak performance being 8.4TF I let it be known that it is fp16.

You are trying to mislead because the vast majority of GAF posters have no idea what FP16 means and you know this full well. Many of them do know that a teraflop is a measure of graphical horsepower and that more is better. You are taking advantage of people's ignorance on the matter.

To be fair, he's not talking about the relevance of FP16, just that the Pro is 8.4TF FP16

That is precisely why his posts are bad. He doesn't explain what it means and why it matters, he just posts a number without any context whatsoever.

I honestly don't know why so many people keep falling for the same schtick over and over again, not just by posters such as OnQ123 but by the console makers themselves. From Sony's turbocharged PC architecture to Microsoft's GTX 1070-beating customizations, it's always the same deal. Companies hyping their next product as the best thing since sliced bread and customers lacking the knowledge to understand technical information yet hyping themselves anyway. There's nothing wrong with being excited about a product but there has to be some degree of critical thinking.
 

Rodelero

Member
I mean, the pretty direct implication of saying 'it's 6TFLOPS FP32 vs 8.4TFLOPS FP16' is that you are trying to give the impression that PS4 Pro is more powerful or capable.

You're putting quotation marks around that, but I'm pretty sure you are trying to quote this:

"Actually it's 6tf fp32 vs 4.2tf fp32/8.4tf fp16" from OnQ, which, as you can see, is quite different, and completely accurate.
 

Taggen86

Member
Still not completely convinced that it will be able to run all titles in native 4k. 6 tflops is still only 6 tflops. Not even my OC GTX 1080 (10 tflops) can run the new mass effect or deus ex in 4k locked 30 fps with ultra settings and without drops. While it is clearly more powerful than ps pro, I do not expect it to perform like a GTX 1070 or 1080 rig with a i6 or i7, especially with the cpu they are using. They should have used a more graphically demanding title than Forza when demonstrating the power of the hardware. Show a 900p/30 fps title such as mass effect or ac unity running in native 4k and I would be really impressed. Forza, a 1st party game which runs in 1080p/60 fps on a 1.2 tflop xbox one running in 4k/60 on the scorpio says little about how 900p/30fps third party titles will run on it. Maybe even ps pro can run Forza in native 4k/60 fps or 1800p/60fps with xbox one settings, who knows. We need to see more games running on scorpio before drawing conclusions about its ability to run games in native 4k.
 

Colbert

Banned
Posting a number that has so far been used to quantify the graphical horsepower of a console or graphics card without the context of what it actually means is deliberately misleading.



You are trying to mislead because the vast majority of GAF posters have no idea what FP16 means and you know this full well. Many of them do know that a teraflop is a measure of graphical horsepower and that more is better. You are taking advantage of people's ignorance on the matter.



That is precisely why his posts are bad. He doesn't explain what it means and why it matters, he just posts a number without any context whatsoever.

I honestly don't know why so many people keep falling for the same schtick over and over again, not just by posters such as OnQ123 but by the console makers themselves. From Sony's turbocharged PC architecture to Microsoft's GTX 1070-beating customizations, it's always the same deal. Companies hyping their next product as the best thing since sliced bread and customers lacking the knowledge to understand technical information yet hyping themselves anyway. There's nothing wrong with being excited about a product but there has to be some degree of critical thinking.

Well said.
 

blu

Wants the largest console games publisher to avoid Nintendo's platforms.
The whole conversation derived from the FP16 double rate feaature not in the Scorpio and that the PS4 Pro 8.4 FP16 TF spec is technically correct. So I assumed the question was meant to be in that area of subject.
The question was exactly in that area of subject, and I get no idea what got you to deviate in describing what an example pipeline does. For example, nothing in that picture says '6 TFLOPS will suffice'. So, back to the question: what content do you envision will be handled by Scoprio so that it would never need an extra flop, ever, at 4k?
 

jroc74

Phone reception is more important to me than human rights
Is anyone, anywhere in this thread, suggesting that Scorpio isn't more powerful than the PlayStation 4 Pro?

I havent seen it yet.

Posting a number that has so far been used to quantify the graphical horsepower of a console or graphics card without the context of what it actually means is deliberately misleading.



You are trying to mislead because the vast majority of GAF posters have no idea what FP16 means and you know this full well. Many of them do know that a teraflop is a measure of graphical horsepower and that more is better. You are taking advantage of people's ignorance on the matter.



That is precisely why his posts are bad. He doesn't explain what it means and why it matters, he just posts a number without any context whatsoever.

I honestly don't know why so many people keep falling for the same schtick over and over again, not just by posters such as OnQ123 but by the console makers themselves. From Sony's turbocharged PC architecture to Microsoft's GTX 1070-beating customizations, it's always the same deal. Companies hyping their next product as the best thing since sliced bread and customers lacking the knowledge to understand technical information yet hyping themselves anyway. There's nothing wrong with being excited about a product but there has to be some degree of critical thinking.

I will take OnQ over posters claiming MS did this and that to the cpu, gpu and Sony didnt do anything.

He's at least posting facts.
 

timlot

Banned
MS said they need over 300gbps memory bandwidth to do native 4K with a 6TF GPU. Wouldn't 8.4TF GPU need more than 218gbps memory bandwidth? Or does FP16 double memory bandwidth too? Does Pro game memory doubles from 5GB to 10GB with FP16? That would need to all calculate right?
 

onQ123

Member
Posting a number that has so far been used to quantify the graphical horsepower of a console or graphics card without the context of what it actually means is deliberately misleading.



You are trying to mislead because the vast majority of GAF posters have no idea what FP16 means and you know this full well. Many of them do know that a teraflop is a measure of graphical horsepower and that more is better. You are taking advantage of people's ignorance on the matter.



That is precisely why his posts are bad. He doesn't explain what it means and why it matters, he just posts a number without any context whatsoever.

I honestly don't know why so many people keep falling for the same schtick over and over again, not just by posters such as OnQ123 but by the console makers themselves. From Sony's turbocharged PC architecture to Microsoft's GTX 1070-beating customizations, it's always the same deal. Companies hyping their next product as the best thing since sliced bread and customers lacking the knowledge to understand technical information yet hyping themselves anyway. There's nothing wrong with being excited about a product but there has to be some degree of critical thinking.

I actually made the mistake of thinking that Gaf would understand , but I'm not the problem the problem is people who don't understand making more noise than the ones that do. so to say that I'm trying to mislead people with facts is one of the silliest things I have ever read.
 
I actually made the mistake of thinking that Gaf would understand , but I'm not the problem the problem is people who don't understand making more noise than the ones that do. so to say that I'm trying to mislead people with facts is one of the silliest things I have ever read.

Do you really not remember implying that MS's 6TF number could be referring to FP16? That's what your posts in the speculation thread were obviously implying.
 

Paradicia

Member
This might seem like a stupid question but what will project scorpio actually be called? I'm surprised we don't have a leak of what it's called at this point.
 
This whole argument--especially with nobody with actual development experience with working with FP16 in real-world examples--smacks of 'ESRAM's secret sauce is lower bandwidth' conversations from Xbone's release.
ESRAM was supposed to give higher bandwidth, not lower...and in fact it did that, according to actual developers of actual Xbox One games.

As far as real development examples of FP16, there were a couple Frostbite technical papers posted earlier (from as far back as PS3!). In addition, here's some 2014 talk about FP16 usage (pro and con) that includes developers. And here's a couple examples from the last year, which list some actual implementations, as well as obstacles.

MS said they need over 300gbps memory bandwidth to do native 4K with a 6TF GPU. Wouldn't 8.4TF GPU need more than 218gbps memory bandwidth? Or does FP16 double memory bandwidth too? Does Pro game memory doubles from 5GB to 10GB with FP16? That would need to all calculate right?
My understanding is that packed math does effectively double memory bandwidth/capacity, because you're fitting two floating point values into the space previously taken by one. I'm sure the reality has many caveats and details that this glosses over, but in general the improved efficiency the whole reason behind using FP16. (Some relief on bandwidth and memory happens even when the hardware doesn't run FP16 at double rate.)

Also, Pro will never be a "8.4TF GPU" in comparison to Scorpio's 6TF. That's Pro's theoretical maximum if absolutely every calculation used FP16. But we know for certain that can't happen.
 

Colbert

Banned
The question was exactly in that area of subject, and I get no idea what got you to deviate in describing what an example pipeline does. For example, nothing in that picture says '6 TFLOPS will suffice'. So, back to the question: what content do you envision will be handled by Scoprio so that it would never need an extra flop, ever, at 4k?

Don't twist my words, what are you trying to reaching for?

I said (paraphrasing): MS did not see a need for the FP16 double rate feature as they think they will be fine with what they have (GPU spec, mem amount and speed) to achieve native 4K resolution with the supposed (and researched) render pipelines in those games.

If you ask for a specific 4K render pipeline image where it says "that are our 4K render passes for Scorpio" in it, I have none. Please ask Ubisoft, Dice, Microsoft, Monolith, Playgound, Turn10 or whatever studio for one. But the one I showed can be used with a 4K frame buffer as well. But you may already know that.

What content will be handled by Scorpio? What do you thing I expect on a gaming console? Games of course!

I also expect the majority of games in native 4K on Scorpio. Not all of course and I am talking about new released games. Already released games is another story as they need to be patched for 4K frame buffers.
 

vpance

Member
So did Sony influence AMD into incorporating 2xFP16 in Vega (like they pushed more ACE for Aync compute) or was in the works already?
 

Colbert

Banned
This might seem like a stupid question but what will project scorpio actually be called? I'm surprised we don't have a leak of what it's called at this point.

It will definitely not be called "Xbox One Scorpio" as Albert Penello mentioned that in the last podcast on majornelson.com . That's why they named the SOC "Scorpio Engine" because they wanted to preserve the scorpio code name somehow with the product (as a fan service)
 
Because TF/TFLOPS stand for trillion floating-point operations per second & that does not tell you if the floating-point operations are 32-bit, 24-bit 16-bit & so on.


Nothing can be said to change the fact that PS4 Pro peak performance is 8.4tf fp16.
**which is limited to only certain tasks as there will be 0 games that are completely writen for fp16 on consoles.**

This is really no different than eSRAM claims that others have brought up. Could eSRAM reach a higher bandwidth than GDDR5? Yes it could. Only very limited in what was actually usable. Same kinds of claims.
 

Leyasu

Banned
But considering that Pro and Scorpio are separated by less than 50% overall, Pro alone getting even a 25% boost (or 20, or 15...) is very significant.


No, not really. As previously posted, FP16 has already been in use for plenty of stuff for years, because it provides some benefits even without packed math. Those elements will now be capable of providing even more performance boost on Pro (and Switch).

The sauce is no longer secret. Microsoft waited a year for nothing. Pro already was/is the world's most powerful console. Microsoft better be careful with their marketing.
 

JaggedSac

Member
The array of estimates for effective power we have so far is 4.6TF to 5.6TF; we'll have to see how accurate any of these might be.

Just a quick estimate, but wouldn't an effective power of 5.6tf require somewhere around 20% of all the instructions being executed to be fp16.
 
FP16 is to Pro what 60 customizations/DX12 in the command processor is to Scorpio. I'm sure they'll provide some improvements but the numbers are the numbers.
 

Paradicia

Member
It will definitely not be called "Xbox One Scorpio" as Albert Penello mentioned that in the last podcast on majornelson.com . That's why they named the SOC "Scorpio Engine" because they wanted to preserve the scorpio code name somehow with the product (as a fan service)

Yeah, I gathered as much!

My guess is Xbox One X.
 

twobear

sputum-flecked apoplexy
ESRAM was supposed to give higher bandwidth, not lower...and in fact it did that, according to actual developers of actual Xbox One games.

As far as real development examples of FP16, there were a couple Frostbite technical papers posted earlier (from as far back as PS3!). In addition, here's some 2014 talk about FP16 usage (pro and con) that includes developers. And here's a couple examples from the last year, which list some actual implementations, as well as obstacles.

Sorry, I mistyped, I meant lower latency not bandwidth.
 

onQ123

Member
ESRAM was supposed to give higher bandwidth, not lower...and in fact it did that, according to actual developers of actual Xbox One games.

As far as real development examples of FP16, there were a couple Frostbite technical papers posted earlier (from as far back as PS3!). In addition, here's some 2014 talk about FP16 usage (pro and con) that includes developers. And here's a couple examples from the last year, which list some actual implementations, as well as obstacles.

Also this demo from the Vega tech talk https://youtu.be/bDl6xJJqIAU?t=22m14s

LyxjuZg.jpg
 

blu

Wants the largest console games publisher to avoid Nintendo's platforms.
Don't twist my words, what are you trying to reaching for?
I'm trying to find out what makes you so confident with your '6TFLOPS must suffice for everybody' (paraphrased).

I said (paraphrasing): MS did not see a need for the FP16 double rate feature as they think they will be fine with what they have (GPU spec, mem amount and speed) to achieve native 4K resolution with the supposed (and researched) render pipelines in those games.
And I asked: native 4k resolution doing what? (not meaning 'what is a pipeline?' I can assure you I know what a rendering pipeline is).

If you ask for a specific 4K render pipeline image where it says "that are our 4K render passes for Scorpio" in it, I have none. Please ask Ubisoft, Dice, Microsoft, Monolith, Playgound, Turn10 or whatever studio for one. But the one I showed can be used with a 4K frame buffer as well. But you may already know that.
Ergo my surprise when you posted a sample pipeline diagram.

What content will be handled by Scorpio? What do you thing I expect on a gaming console? Games of course!
I suspected you meant spreadsheets. /s

I also expect the majority of games in native 4K on Scorpio. Not all of course and I am talking about new released games. Already released games is another story as they need to be patched for 4K frame buffers.
Majority of what games? Current games and their sequels rendered at 4k? Or something more ambitious? For instance, Epic's original SVOGI @ 4K might not run at 6TFLOPS, if we are to trust Sweeney's own words.
 

timlot

Banned
So Pro is going to be like PS3 cell processor. Once developer get the hang of it that's when the magic is going to start happening. Checkerboarding will be a thing of the past. Sony always thinking about inefficiency vs MS short term brute force approach. Guess its that Japanese vs US mind set.
 
Microsoft coming to the party a year late with an underpowered machine at a higher price?

What happened to learning from past mistakes? Phil 'hot air' Spencer just doesn't get it...The disdain they continue to show towards gamers is baffling.

They are even losing the software sales race to Switch in the UK, what is it going to take for them to wake up and see sense...Sad and sorry sight from the company that once put out bleeding edge tech and set trends in the gaming world.
 
So Pro is going to be like PS3 cell processor. Once developer get the hang of it that's when the magic is going to start happening. Checkerboarding will be a thing of the past. Sony always thinking about inefficiency vs MS short term brute force approach. Guess its that Japanese vs US mind set.

And yet the lead architect of the PS4 Pro is American ;)
 

Marmelade

Member
So Pro is going to be like PS3 cell processor. Once developer get the hang of it that's when the magic is going to start happening. Checkerboarding will be a thing of the past. Sony always thinking about inefficiency vs MS short term brute force approach. Guess its that Japanese vs US mind set.

I can't even tell which posts are supposed to be serious or not in this thread anymore
I'm gonna go with "not serious" on this one
 

vpance

Member
No, it is still a big leap over PS4 Pro. Most modern graphics calculations require FP32 and will not be able to work to the same result in FP16.

From sebbbi on B3D

All existing games (except a few HDR games) output image at 8 bits per channel (RGB8). Input textures are also commonly 8 bit per channel (and BC compressed = lower quality as 8 bit).

As your input and output data has only 8 bit precision you don't need to calculate all intermediate math at 32 bit. Games don't store intermediate buffers as 32 bit floats either. Rgba16f is used commonly for HDR data and rgb10 and rgba8 for other intermediate buffers. 16 bit float processing is fine for most math in games. Results cannot be distinguished by naked eye from full 32 bit float pipeline, as long as the developer knows what he/she is doing. Especially if temporal AA is used.

Unfortunately writing good mixed fp16/fp32 code requires good knowledge about floating point math behaviour and some basic numeric range analysis (inputs/outputs and intermediate values). It is possible to write math in a way that minimizes floating point issues, allowing you to use fp16 more often. Of course if you use fp16 in a wrong way, you get banding and other artifacts.

Sounds ok to me. Devs will eventually get better at using it.
 

Detective

Member
So Pro is going to be like PS3 cell processor. Once developer get the hang of it that's when the magic is going to start happening. Checkerboarding will be a thing of the past. Sony always thinking about inefficiency vs MS short term brute force approach. Guess its that Japanese vs US mind set.

Yup, you figured it out.
 

Inuhanyou

Believes Dragon Quest is a franchise managed by Sony
So Pro is going to be like PS3 cell processor. Once developer get the hang of it that's when the magic is going to start happening. Checkerboarding will be a thing of the past. Sony always thinking about inefficiency vs MS short term brute force approach. Guess its that Japanese vs US mind set.

I almost got baited. Good one
 

geordiemp

Member
So Pro is going to be like PS3 cell processor. Once developer get the hang of it that's when the magic is going to start happening. Checkerboarding will be a thing of the past. Sony always thinking about inefficiency vs MS short term brute force approach. Guess its that Japanese vs US mind set.

Sony just saving 100 bucks on cost (vortex cooler + extra Memory costs and more complex arrangement for bandwidth / board costs).

Whether MS pass on the extra cost or not to consumers is another discussion.

It is intersting. Even though I have a pro, I think more expensive is the better option personally....
 

Green Yoshi

Member
Customizing the gpu to handle latency, bandwidth, and memory extremely more than a basic model is way better for graphics than fp16 I would bet. How's fp16 working for the switch? ;)
Lego City: Undercover runs in 1080p on Switch and with a lower resolution on Xbox One. ;-)
 

twobear

sputum-flecked apoplexy
So Pro is going to be like PS3 cell processor. Once developer get the hang of it that's when the magic is going to start happening. Checkerboarding will be a thing of the past. Sony always thinking about inefficiency vs MS short term brute force approach. Guess its that Japanese vs US mind set.

Joke post but this is 100% what some posters want you to believe.
 

Fredrik

Member
I would go back to Sunset Overdrive instantly if this ends up happening for that game.
Oh wow I totally forgot about that, Sunset Overdrive at 60fps would be nuts! And what about Forza Horizon 2,3? Could they hit 60fps at 1080p or is the open world too much to handle for the modified jaguar if 60fps is the target?
 
Top Bottom