• 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.

Xbox Project Scorpio Announced - 6TFlops, 320GB/s - Fall 2017

Status
Not open for further replies.

c0de

Member
Graphical costs are exponential so measurements of power need to be relative. That is, 40% is 40% regardless of how much the actual gap is. So yes the gap is about the same.

If it was exponential, than this would need to be more than the linear increase in computing power but perhaps you can explain in detail which parts have which computational complexity.
Perhaps you can make a list of each relevant algorithm and its respective runtime, in o-notation.
 

onanie

Member
If it was exponential, than this would need to be more than the linear increase in computing power but perhaps you can explain in detail which parts have which computational complexity.
Perhaps you can make a list of each relevant algorithm and its respective runtime, in o-notation.

When comparing systems, relative measures are more relevant to their current capabilities and requirements than absolute numbers.
 

c0de

Member
When comparing systems, relative measures are more relevant to their current capabilities and requirements than absolute numbers.

This depends on what you are trying to say. If we go back to the first post of this chain, it's no wonder aceofspades is using the number that he thinks is best for showing that Scorpio is no big deal while in fact it's one ps4 better in computing power.
In the end, it isn't even important at all as we don't know what the difference on screen will be, to be honest.
Also, of Scorpio has a better CPU, you won't see that in a raw number but the system will leave the ps4p easily behind.
 
Everytime I see this article posted, I make a remark saying it's clickbait and explain why, no-one replies to me, then it keeps on being reposted.

There's no direct quote, the roadmap is a 12 month roadmap which aligns to Scorpio being massed produced and AMD will really want new semi-custom business on their new designs, it's good for investment.

That article is FUD.

I could see this happening if not Zen, although the price comes with die space and the size of the box could just be similar to the OG X1. With the Surface team now responsible for hardware, it wouldn't surprise me if they put some passive liquid cooling in like they've done for SP4, Surface Book.

Either way, I'm excited to see what this thing will look like and what will be inside. I'm hoping for the best.
Puma = Jaguar with less leakage in the same process(28nm).So no.You could even call Jaguar at 16nm as Cheetah that is like AMD used to call it.
 

onanie

Member
This depends on what you are trying to say. If we go back to the first post of this chain, it's no wonder aceofspades is using the number that he thinks is best for showing that Scorpio is no big deal while in fact it's one ps4 better in computing power.
In the end, it isn't even important at all as we don't know what the difference on screen will be, to be honest.
Also, of Scorpio has a better CPU, you won't see that in a raw number but the system will leave the ps4p easily behind.

It does not depend on what anyone is trying to say. One could say the PS4 was 3 PS3s, or 96 PS2s, or 1000 PS1s more powerful than the Xbone, but none of that is any more informative than saying that the PS4 is 0.6TF more powerful than the Xbone.

The difference is 40%, and that is immediately understood in the context of the relevant generation.
 

c0de

Member
It does not depend on what anyone is trying to say. One could say the PS4 was 3 PS3s, or 96 PS2s, or 1000 PS1s more powerful than the Xbone, but none of that is any more informative than saying that the PS4 is 0.6TF more powerful than the Xbone.

The difference is 40%, and that is immediately understood.

It is understood but it's not telling the whole story, no matter if you think it does or not. A system can't be thoroughly described by one number and comparing two systems, broken down to one number, and calculating the relative difference is even more at least misleading, if not even wrong.
 

onanie

Member
It is understood but it's not telling the whole story, no matter if you think it does or not. A system can't be thoroughly described by one number and comparing two systems, broken down to one number, and calculating the relative difference is even more at least misleading, if not even wrong.

No it is not.
 

mrklaw

MrArseFace
Jaguar doesn't pack the punch needed for Oculus/Vive.

It packs the punch needed for PSVR which can run at 120Hz for some games. Then the extra GPU can be used for downsampling to clean up the image, or drive the higher resolution oculus/vive displays (or both)
 

Theonik

Member
If it was exponential, than this would need to be more than the linear increase in computing power but perhaps you can explain in detail which parts have which computational complexity.
Perhaps you can make a list of each relevant algorithm and its respective runtime, in o-notation.
Many effects will incur costs proportional to the current load. A frame that requires 33ms to render needs at least 2x the rendering power to finish in half the time. Running a PS4 game at 2x the framerate needs an exponential increase in performance. Similarly increasing the resolution has a similar effect. This is of course a reductive argument and there is many other factors that affect this comparison but the actual increase doesn't really tell you much, while relative increase does.
 

etta

my hard graphic balls
It packs the punch needed for PSVR which can run at 120Hz for some games. Then the extra GPU can be used for downsampling to clean up the image, or drive the higher resolution oculus/vive displays (or both)
PSVR is custom built with the Jaguar in mind.
Go and take a look at minimum specs for Oculus and Vive. i5 and GTX 970. PSVR is weaker than both.
 

Theonik

Member
PSVR is custom built with the Jaguar in mind.
Go and take a look at minimum specs for Oculus and Vive. i5 and GTX 970. PSVR is weaker than both.
The games are made with the headset in mind. Not the other way around. If Scorpio got support there is nothing stopping developers from treating it as an optimisation target. In fact it's exactly what will happen.
 

etta

my hard graphic balls
The games are made with the headset in mind. Not the other way around. If Scorpio got support there is nothing stopping developers from treating it as an optimisation target. In fact it's exactly what will happen.
Nope.
Oculus/Vive don't have reprojection so 90fps is the minimum and they also have higher resolution screens, 1200p instead of PSVR's 1080p. i5 4590/GTX 970 equivalent is a must.
 

Theonik

Member
Nope.
Oculus/Vive don't have reprojection so 90fps is the minimum and they also have higher resolution screens, 1200p instead of PSVR's 1080p. i5 4590/GTX 970 equivalent is a must.
Reprojection can and probably would be done on the system itself. Moreover, not every game needs to work.
 

quest

Not Banned from OT
Reprojection can and probably would be done on the system itself. Moreover, not every game needs to work.

I think basically every game would need to work to make it worth while. I can imagine someone who had a Scorpio bought a Oculus only to find out only 20-30% of the games work on it. And all those that do work are basically simple tech demos since the CPU power is not there to do anything else.
 

etta

my hard graphic balls
Reprojection can and probably would be done on the system itself. Moreover, not every game needs to work.
No since that would add delay, PSVR does reprojection in the headset after it receives the signal (unless I am misremembering it).
 

Theonik

Member
I think basically every game would need to work to make it worth while. I can imagine someone who had a Scorpio bought a Oculus only to find out only 20-30% of the games work on it. And all those that do work are basically simple tech demos since the CPU power is not there to do anything else.
But again. Making your entire argument around 'Microsoft would never disappoint people like that!' is not a very solid argument.
 

quest

Not Banned from OT
But again. Making your entire argument around 'Microsoft would never disappoint people like that!' is not a very solid argument.

Who cares about MS the Oculus people are not going to let it happen unless almost every game works. The last thing they want is bad publicity from people buying it and hardly any games work. I also imagine developers would not want it to happen either. Since they would take part of the blame for a horrible CPU not supporting VR.
 
No since that would add delay, PSVR does reprojection in the headset after it receives the signal (unless I am misremembering it).

No, it still does this on the GPU itself. Basically, immediately before the system sends off the frame, it will update and warp the image at the very last chance it can.

The breakout box does a tiny amount of processing for 3D audio, but it also takes the frame from the PS4 and will undistort it for display on the TV. It does not perform the timewarp though, and neither does the headset.

Oh shit, nice. So it takes 60fps to 120fps for the games that don't meet 90fps?

Yes. Except the Vive runs at 90Hz. So what it does, is when it takes the game is failing (key word) to hit 90Hz, it will force the game to run at 45Hz and then "reproject" or "timewarp" up to 90Hz. With PSVR, it's constantly doing this reprojection since most games will run at 60Hz and "reprojected" to 120Hz

Basically, both Vive and PSVR handle this problem in nearly the same way. They cut the framerate to an even multiple of the refresh rate and double it. They don't support arbitrary framerates in between (unlike Oculus which I'll get to in a second). They both support rotational timewarp only, but because the refresh rates are high, it's not very noticeable. This means that the game itself is the only time position is taken into account, so during reprojection, it does not modify for the latest positional data, only the rotation of the headset. The idea here is to keep frame times consistent so the effects of the timewarp don't continuously come and go. For most PSVR titles, it's going to be continuously reprojecting anyway from 60->120. For Vive, they have developers target 90, and 45->90 reprojection is there as a safety cushion.

Oculus on the other hand does Asynchronous Timewarp / reprojection. They continuously are updating the final frame with both rotational and positional information, but don't affect the game's frame rate. If a frame is late, then it will reproject the last frame it got, but it doesn't have the game render any differently. This means your game can be running at 80Hz just fine, and Oculus will take your frames as soon as they are ready and reproject any frames which didn't make it on time. Now, the downside here is that objects in the scene may appear to judder unpredictably (going from 90fps to 80 to 75 back to 90, etc etc), whereas on Vive/PSVR, the objects won't judder but you will notice them rendering at a lower framerate on the Vive. With PSVR, only if it falls below 60 will you notice this, but since it defaults to using reprojection mainly, you're not going to go from 120 down to 60, but instead always run at 60.

In all cases, if you move your head very fast, you'll potentially notice artifacts because there's only so much the system can do to warp your image before it runs out of frame and you notice either black bars or maybe it does something fancy like a smart color fill based on the edges of the frame instead.

The ELI5 is this. Imagine you have a drummer who's tapping out a consistent beat (representing the game).
- PSVR always has this drummer tapping every other note, and someone fills in the missing beats (representing 60 -> 120 reprojection)
- Vive has the drummer tapping at full speed, but if it misses some beats, the drummer will start tapping at half the beat, and someone fills in the missing beats. (representing game running at 90, and then 45->90 reprojection taking over)
- Rift has the drummer tapping at full speed, and anytime there is a missing beat, they fill it in whenever necessary but the drummer can continue tapping at whatever speed it can do (representing game running at 90 and game failing to hit 90 and async timewarp taking over)
 

etta

my hard graphic balls
No, it still does this on the GPU itself. Basically, immediately before the system sends off the frame, it will update and warp the image at the very last chance it can.

The breakout box does a tiny amount of processing for 3D audio, but it also takes the frame from the PS4 and will undistort it for display on the TV. It does not perform the timewarp though, and neither does the headset.



Yes. Except the Vive runs at 90Hz. So what it does, is when it takes the game is failing (key word) to hit 90Hz, it will force the game to run at 45Hz and then "reproject" or "timewarp" up to 90Hz. With PSVR, it's constantly doing this reprojection since most games will run at 60Hz and "reprojected" to 120Hz

Basically, both Vive and PSVR handle this problem in nearly the same way. They cut the framerate to an even multiple of the refresh rate and double it. They don't support arbitrary framerates in between (unlike Oculus which I'll get to in a second). They both support rotational timewarp only, but because the refresh rates are high, it's not very noticeable. This means that the game itself is the only time position is taken into account, so during reprojection, it does not modify for the latest positional data, only the rotation of the headset. The idea here is to keep frame times consistent so the effects of the timewarp don't continuously come and go. For most PSVR titles, it's going to be continuously reprojecting anyway from 60->120. For Vive, they have developers target 90, and 45->90 reprojection is there as a safety cushion.

Oculus on the other hand does Asynchronous Timewarp / reprojection. They continuously are updating the final frame with both rotational and positional information, but don't affect the game's frame rate. If a frame is late, then it will reproject the last frame it got, but it doesn't have the game render any differently. This means your game can be running at 80Hz just fine, and Oculus will take your frames as soon as they are ready and reproject any frames which didn't make it on time. Now, the downside here is that objects in the scene may appear to judder unpredictably (going from 90fps to 80 to 75 back to 90, etc etc), whereas on Vive/PSVR, the objects won't judder but you will notice them rendering at a lower framerate on the Vive. With PSVR, only if it falls below 60 will you notice this, but since it defaults to using reprojection mainly, you're not going to go from 120 down to 60, but instead always run at 60.

In all cases, if you move your head very fast, you'll potentially notice artifacts because there's only so much the system can do to warp your image before it runs out of frame and you notice either black bars or maybe it does something fancy like a smart color fill based on the edges of the frame instead.

The ELI5 is this. Imagine you have a drummer who's tapping out a consistent beat (representing the game).
- PSVR always has this drummer tapping every other note, and someone fills in the missing beats (representing 60 -> 120 reprojection)
- Vive has the drummer tapping at full speed, but if it misses some beats, the drummer will start tapping at half the beat, and someone fills in the missing beats. (representing game running at 90, and then 45->90 reprojection taking over)
- Rift has the drummer tapping at full speed, and anytime there is a missing beat, they fill it in whenever necessary but the drummer can continue tapping at whatever speed it can do (representing game running at 90 and game failing to hit 90 and async timewarp taking over)
Shieet, thanks for this info.
Shouldn't have mentioned anything, clearly I was misinformed.
 

Biggzy

Member
Hmm. GPU wise, that's a difference of almost a WHOLE PS4? Significant!

Yep. It's pretty crazy if you think about it like that. Which is why I would not be too disappointed if developers use a checkerboard like technique and use the remaining power to really crank up the visual quality.
 

onQ123

Member
Remember when I was crazy for thinking that MS might be using the FP16 number?


znLrZmU.png


http://videocardz.com/63700/exclusive-first-details-about-amd-vega10-and-vega20

Don't seem so crazy now do I?


I don't think they will because they said that it's 4.5X the power of the Xbox One but this is why I said you can't just go by the 6TF number without more context.
 

Pif

Banned
Explain how a .5 TF difference is the same as as a 1.8 TF difference.
Bigger bells and whistles cost more.

Which one is the bigger difference for you:

Between 0.5 and 1.5Tflops = 1Tflop difference

Or

Between 10 Tflops and 13 = 3Tflops difference.

Think really good about it. Because it's the first one.
 
Remember when I was crazy for thinking that MS might be using the FP16 number?


znLrZmU.png


http://videocardz.com/63700/exclusive-first-details-about-amd-vega10-and-vega20

Don't seem so crazy now do I?


I don't think they will because they said that it's 4.5X the power of the Xbox One but this is why I said you can't just go by the 6TF number without more context.

MS is not going to sell a 3TF SP box in 2017, especially after the launch of a 4TF PS4 Pro.

Though the talk of native half precision in Vega is interesting, makes me wonder what the FP16 capabilities the PS4 Pro and Scorpio are. Would be interesting if the Scorpio has a more advanced architecture with various FP16 fuse instructions and if the PS4 Pro does not...
 

dr_rus

Member
Remember when I was crazy for thinking that MS might be using the FP16 number?


znLrZmU.png


http://videocardz.com/63700/exclusive-first-details-about-amd-vega10-and-vega20

Don't seem so crazy now do I?


I don't think they will because they said that it's 4.5X the power of the Xbox One but this is why I said you can't just go by the 6TF number without more context.
This rumour is pointing us to Scorpio using Vega 11 if anything. If it's half Vega 10 then it'll be about 6-8TF (8 on PC where a higher TDP is possible), a prefect fit for Scorpio's spec.
 

onQ123

Member
Sorry..can you elaborate please?

A while ago I tried to explain to people that we can't just go by the 6TF number that MS gave out because the bigger GPUs was moving towards using FP16 like the mobile GPUs & that they could be using the peak number which would be the FP16 number & now Vega 10 is being shown with 24TF FP16 it's peak number. Everyone is expecting MS to be using Vega so if the smaller card also has a higher FP16 number they could be using the bigger number.

MS is not going to sell a 3TF SP box in 2017, especially after the launch of a 4TF PS4 Pro.

Though the talk of native half precision in Vega is interesting, makes me wonder what the FP16 capabilities the PS4 Pro and Scorpio are. Would be interesting if the Scorpio has a more advanced architecture with various FP16 fuse instructions and if the PS4 Pro does not...

They would never have to address it as a 3TF SP box they could just use the higher peak number which is the FP16 number if the GPU is Vega.

This rumour is pointing us to Scorpio using Vega 11 if anything. If it's half Vega 10 then it'll be about 6-8TF (8 on PC where a higher TDP is possible), a prefect fit for Scorpio's spec.

Half of Vega 10 would be 6TF on PC if it's clocked as high as Vega 10
 
Remember when I was crazy for thinking that MS might be using the FP16 number?


znLrZmU.png


http://videocardz.com/63700/exclusive-first-details-about-amd-vega10-and-vega20

Don't seem so crazy now do I?


I don't think they will because they said that it's 4.5X the power of the Xbox One but this is why I said you can't just go by the 6TF number without more context.

What was crazy wasn't the possibility of using half precision flops for having seemingly higher specs. What's crazy is to say that the 6tf was the half precision figure.

And still is, that image changes nothing.
 
Everytime I see this article posted, I make a remark saying it's clickbait and explain why, no-one replies to me, then it keeps on being reposted.
I have replied to you twice, in two separate threads. The linked post is indeed fluff, but it's not baseless. Behind it is an article from the respectable German IT site ComputerBase. They reported that AMD's Lisa Su indirectly suggested Zen APUs wouldn't be ready until 2018.

You're right to dismiss the specific page, but you've repeatedly neglected to point people to the real info. (Which doesn't give us a definitive answer either, but deserves to be part of the conversation.)

Notebook Zen in time for Scorpio?
Notebook Zen is very unlikely to be used in Scorpio. Microsoft will almost certainly be using an APU again; changing to separate CPU and GPU would increase design complexity and cost.
 

MCD

Junior Member
Seeing what Playground Games can do with only 1.3TF made me believe in MS's claim of Native 4K for their games.
 

Xcell Miguel

Gold Member
Seeing what Playground Games can do with only 1.3TF made me believe in MS's claim of Native 4K for their games.

Well, my 980 Ti OC has a bit less than 6 TF and can easily do native 4K at 30 FPS in most games where 60 is a problem (I'm getting 90-120 in Forza 6 Apex in 4K though, and 60-80 in Halo 5 Forge, but in The Division I get 55-80 FPS in 1440p and much less in 4K).

So the Scorpio would be able to easily do native 4K30 in most games, now let's see if the CPU can handle 60 FPS in most games in 1080p.
 

onQ123

Member
What was crazy wasn't the possibility of using half precision flops for having seemingly higher specs. What's crazy is to say that the 6tf was the half precision figure.

And still is, that image changes nothing.

24TF is the half precision flops for Vega 10 so why wouldn't MS use this number if they are using Vega?


I'm not saying that they are doing this but they could be doing this.


Maybe they will use Vega 11 which is said to replace Polaris 10 & if that remains a 36 CU card for the top card it would need to be clocked at 1310Mhz to reach 6TF that seem pretty high for a console, a 40 CU card would need to be clocked at 1174Mhz still seem kinda high but I guess that could happen, but I have a feeling that ESRAM is still needed for Xbox One games so you will need room for that on the GPU which make a 40 CU GPU seem less likely but if ESRAM isn't needed I guess they could have a 40 or 44 CU card clocked at 1070Mhz but that still seem like it would cause some heat inside of a console.
 

dr_rus

Member
Half of Vega 10 would be 6TF on PC if it's clocked as high as Vega 10

Smaller chips usually have higher clocking limit so something along the lines of 7-8TF from half a Vega 10 is possible. 6TF would be rather conservative considering 480 is 5,85 on PC. I'm pretty sure that AMD won't produce a completely custom GPU for Scorpio so re-using Vega 11 seems quite possible, in the same way as Polaris 10 is being re-used for Neo.
 
24TF is the half precision flops for Vega 10 so why wouldn't MS use this number if they are using Vega?


I'm not saying that they are doing this but they could be doing this.


Maybe they will use Vega 11 which is said to replace Polaris 10 & if that remains a 36 CU card for the top card it would need to be clocked at 1310Mhz to reach 6TF that seem pretty high for a console, a 40 CU card would need to be clocked at 1174Mhz still seem kinda high but I guess that could happen, but I have a feeling that ESRAM is still needed for Xbox One games so you will need room for that on the GPU which make a 40 CU GPU seem less likely but if ESRAM isn't needed I guess they could have a 40 or 44 CU card clocked at 1070Mhz but that still seem like it would cause some heat inside of a console.

Because if they were using half precision numbers they would say 12 or bigger, just like Amd is saying 24 which is like a lot more than what's on the market today.

Ms releasing a 3tflop console next year is beyond absurd.

As for the number of CUs, they will go well over 40, as you noted yourself, anything in that range requires a clock way higher than a console can afford to have.
 

leeh

Member
I have replied to you twice, in two separate threads. The linked post is indeed fluff, but it's not baseless. Behind it is an article from the respectable German IT site ComputerBase. They reported that AMD's Lisa Su indirectly suggested Zen APUs wouldn't be ready until 2018.

You're right to dismiss the specific page, but you've repeatedly neglected to point people to the real info. (Which doesn't give us a definitive answer either, but deserves to be part of the conversation.)
I'm not ignoring you, I'm just in Asia so I just end up missing the threads and then they're not on the first page.

She said Zen with new graphics. That doesn't mean Zen is delayed, it means Vega is delayed which he had a thread on. There is no indication of Zen being delayed at all. When I constructed my reply in one of the other threads, I actually stated that Zen was fine and Vega may be delayed based on what they've said.
 
Status
Not open for further replies.
Top Bottom