NGGP: NextGen Garbage Pile (aka: No one reads the topics or stays on topic) *spawn*

[Deleted member 13524]

Can we please stop comparing the GPU to specially binned mobile parts?

The desktop 7750 is a 55W* card. The 7870M is 45W, not 40W. Durango has 20% more CUs. In addition to the CPU, it also has DMEs, scalers, 32MB of ESRAM, etc. and probably an ARM security core from the rumors. It's going to be at least 120W.

Can we please stop assuming that all mobile GPUs are ultra-specially-binned and hand-picked chips? That's nonsense.

For low and mid-end parts, AMD probably even sells a lot more laptop chips than their desktop counterparts. AMD or nVidia would never be able to meet the supply demands for their mobile parts if they were all "specially binned".

The "mobile" chips are put into PCBs optimized for lower consumption, they're downclocked and undervolted as well as the memory chips.
There's no "special-sauce" going into mobile GPUs so please stop this urban myth about everything mobile being "specially binned" and impossible to use in a console.



The Cape Verde occupies 123mm^2 at 28nm. Even if we only assume a reduction to 33% of its original size during a 28->14nm shrink, we're talking about a 41mm^2.

41mm^2 is a lot smaller than AMD's current lower end chip at 40nm - Caicos (59mm^2) - which has a TDP of 9W in its 800MHz form.


In Durango, we're looking at:

1 - 8-core Jaguar that in 2013 - 28nm - are already going into tablets in a 4-core version, probably even at the same 1.6GHz base clocks.
2 - Main memory with a bandwidth that can be covered by a 4-channel LPDDR4 setup in 2017.
3 - A GPU+eDRAM combo that at 14nm could be as small as the current 40nm Caicos, which at 800MHz has a TDP of 9W.
4 - The SoC with CPU+GPU+eDRAM combined - which we're assuming it may hit as much as 300m^2 - should occupy less than 100mm^2 using 14nm.
What ~100mm^2 SoCs do we know with ~1.6GHz CPU cores right now? Tegra 3 at 40nm, Exynos 4412 at 32nm?

If all of this can be made with a 25W total TDP in 14nm, why wouldn't it fit a tablet form factor, given that we already have tablets in the market with that total TDP?

 

[TheChefO]

Hey Chefo, we actually don't KNOW anything for certain.

But taking the VG leak at face value we know it is not PVR but GCN. Just look at the cache design and the configuration.

Btw, the real wild card is... price.

All bets are off if MS can fit this all into a $249 box.

The die space allocated to the CPU and GPU in Durango is a lot less than the Xbox 360 (which from a HW perspective was not overly complex or expensive once supply channels flowed). Xenos was 2 chips (about 182mm^2 and 80mm^2) at 90nm at launch and Xenon was about 176mm^2 (each CPU was about 30-40mm^2 and the remainder of die space was the 1MB of shared L2 cache and bus). That is about 438mm^2, give or take, of total die space.

Cape Verde (7770) is a 123mm^2 GPU at 28nm. Durango is fairly close to Cape Verde units + 32MB of ESRAM (which sounds like a variant of 1T-SRAM in which case it will be in the 20-50mm^2 range?). So the total GPU budget is shrinking about about 100mm^2.

The CPU side is similar. Jaguar cores are about 3mm^2 each and the L2 cache is 3-5mm^2 (give or take) so the CPU is going to fall under 60mm^2. Again, about 100mm^2 less than the Xbox 360 budget for the CPU.

Durango is looking to cut out 200mm^2 of silicon footprint--about 40% reduction in budgets.

The main memory this go around is also going on the cheap side; the Xbox 360 had for the time relatively fast GDDR3 memory; Durango is going with bog standard DDR3 (8GB of DDR3 can be scored easily for $50 at retail).

Depending on the cost of Kinect 2, packed in software, any HW BC, and the cost of the controller it looks like MS will be in a much better position on cost than last go around.

And then there is the crazy stupid Cell Phone pricing agreement plan--expect to see $99 Xbox 3's with 2-3 year contracts of XBL for $15/mo.

Which will position MS where they want to be: a small, affordable console that is a full blown set top box/media all in one experience.

There have been quite a few hints that this isn't a bog standard GCN architecture. Whether that means PowerVR Tile based GPU or something else ... remains to be seen.

If that is all BS, then I expect mediocre sales. Regardless of price.

$200 MSRP didn't help Dreamcast.
$200 MSRP didn't help GC.

Sub-par hardware is sub-par hardware. PS2 was able to get away with it by (marketing, past success with ps1, and great 3d party support as well as ...) having radically different hardware which was difficult to see exactly how it stacked up to the competition, but if the console architecture is as close as it seems, then the difference in performance will be readily apparent.

 

[babybumb]

Thick API? Is this even a console?

I guess it makes sense. MS just put out a "no compromise" Surface with 4hours battery

 

[anexanhume]

Can we please stop assuming that all mobile GPUs are ultra-specially-binned and hand-picked chips? That's nonsense.

For the high end, they are.

For low and mid-end parts,

Which the 7870M is not.

AMD probably even sells a lot more laptop chips than their desktop counterparts.

I would bet anything it's not true for the high end. You can count the people that offer the 7870M or higher on one hand: Clevo, Samsung, Asus, MSI and Alienware.

AMD or nVidia would never be able to meet the supply demands for their mobile parts if they were all "specially binned".

Which is why you don't bin the high volume parts.

The "mobile" chips are put into PCBs optimized for lower consumption, they're downclocked and undervolted as well as the memory chips.
There's no "special-sauce" going into mobile GPUs so please stop this urban myth about everything mobile being "specially binned" and impossible to use in a console.

Those points have nothing to do with how its binned. Those are just how the binned part is clocked and ran. Moreover, you shouldn't be looking at how its binned to see whether or not it would work in a console. You should look at why it's binned. If only 10% of the parts meet the spec you need for that part, that's your bin. The point is that you can't expect to mass produce if your yield isn't going to be a majority of your good chips. Thus, you can't look at mobile high end parts to get a guesstimate for power consumption. Yes, the console part will be downclocked (and undervolted as performance allows), but that doesn't mean it fit that 10% you're reserving for mobile high end parts.

The Cape Verde occupies 123mm^2 at 28nm. Even if we only assume a reduction to 33% of its original size during a 28->14nm shrink, we're talking about a 41mm^2.

41mm^2 is a lot smaller than AMD's current lower end chip at 40nm - Caicos (59mm^2) - which has a TDP of 9W in its 800MHz form.

Which is still double what the GPU should be. Also, can you quote some 14nm leakage numbers? I guarantee you the power consumption won't scale linearly.

In Durango, we're looking at:

1 - 8-core Jaguar that in 2013 - 28nm - are already going into tablets in a 4-core version, probably even at the same 1.6GHz base clocks.

You're assuming it's a standard Jaguar core with no customizations. We don't know that yet.

2 - Main memory with a bandwidth that can be covered by a 4-channel LPDDR4 setup in 2017.

Do you know if LPDDR4 has the same latency as DDR3? Have you ever heard of a tablet with a 256 bit memory interface?

3 - A GPU+eDRAM combo that at 14nm could be as small as the current 40nm Caicos, which at 800MHz has a TDP of 9W.

Where did you start assuming you knew how big the eSRAM was? I missed that. And as I said, 9W is about twice what it should be for today's affordable tablets. Otherwise you're getting larger, more expensive (900+) and sacrificing battery life (especially with gaming!). Who wants to buy that?

4 - The SoC with CPU+GPU+eDRAM combined - which we're assuming it may hit as much as 300m^2 - should occupy less than 100mm^2 using 14nm.

Numbers as high as 450mm^2 have been floated. This potentially meshes with rumored production issues.

What ~100mm^2 SoCs do we know with ~1.6GHz CPU cores right now? Tegra 3 at 40nm, Exynos 4412 at 32nm?

Power profile of x86 and ARM cores are different. ARM cores are optimized for mobile environments.

If all of this can be made with a 25W total TDP in 14nm, why wouldn't it fit a tablet form factor, given that we already have tablets in the market with that total TDP?

Why would you want it to?

Aegies said his info is from Feb 2012, but 8 core CPU and 1.2TF number came before. First who actually had it was the guy that posted it on pastebin. I think it was Dec 2011.

He also said it's still accurate as far as he knows.

There have been quite a few hints that this isn't a bog standard GCN architecture. Whether that means PowerVR Tile based GPU or something else ... remains to be seen.

If that is all BS, then I expect mediocre sales. Regardless of price.

$200 MSRP didn't help Dreamcast.
$200 MSRP didn't help GC.

Sub-par hardware is sub-par hardware. PS2 was able to get away with it by (marketing, past success with ps1, and great 3d party support as well as ...) having radically different hardware which was difficult to see exactly how it stacked up to the competition, but if the console architecture is as close as it seems, then the difference in performance will be readily apparent.

I'm confused. At first you seem to suggest that poor hardware will affect sales regardless of price, but then you go on to quote two great pieces of hardware that were killed by marketing, not a superior machine. The Dreamcast was quite powerful for when it was released (only 3 years after N64) and the gamecube was quite a bit more capable than the PS2. As you correctly pointed out, it's marketing and 3rd party support that killed both. Nintendo lacked the 2nd, as did Sega because of poor Saturn reception. Neither Microsoft nor Sony have that issue this generation.

 

[TheChefO]

I'm confused. At first you seem to suggest that poor hardware will affect sales regardless of price, but then you go on to quote two great pieces of hardware that were killed by marketing, not a superior machine. The Dreamcast was quite powerful for when it was released (only 3 years after N64) and the gamecube was quite a bit more capable than the PS2. As you correctly pointed out, it's marketing and 3rd party support that killed both. Nintendo lacked the 2nd, as did Sega because of poor Saturn reception. Neither Microsoft nor Sony have that issue this generation.

Dreamcast was killed by marketing which suggested it was to be outclassed by superior hardware which btw was the follow-up to the most popular console the prior gen (even though 1st gen ps2 software was inferior to DC).

GC was again made out to be inferior (marketing) to ps2 (and xbox) regardless of how close the hardware actually was.

Point stands - Nobody (not many) wants the "inferior version" regardless of cost differences. If in fact the Durango is producing inferior results compared to Orbis, a low price won't help (much). Neither will marketing, as these days buyers are better informed and if the architectures are as close as suggested, there will be no "magic tricks to unlock if given enough time" as was the message with ps2 and ps3.

This will be game, set and match rather quickly.

 

[Aeoniss]

Dreamcast was killed by marketing which suggested it was to be outclassed by superior hardware which btw was the follow-up to the most popular console the prior gen (even though 1st gen ps2 software was inferior to DC).

GC was again made out to be inferior (marketing) to ps2 (and xbox) regardless of how close the hardware actually was.

Point stands - Nobody (not many) wants the "inferior version" regardless of cost differences. If in fact the Durango is producing inferior results compared to Orbis, a low price won't help (much). Neither will marketing, as these days buyers are better informed and if the architectures are as close as suggested, there will be no "magic tricks to unlock if given enough time" as was the message with ps2 and ps3.

This will be game, set and match rather quickly.

Providing there is nothing that counteracts the apparent inferiority of the hardware- e.g. the "uniqueness" factor that lead the Wii (in combination with a comparatively low price) steal the show last generation.

Kinect 2.0 may be what MS is banking on.

Still, I would personally favor a more robust and powerful gaming machine over one that is going well out of its way to be the most unique 'special' option out there- at the cost of hardware potency.

 

[anexanhume]

Dreamcast was killed by marketing which suggested it was to be outclassed by superior hardware which btw was the follow-up to the most popular console the prior gen (even though 1st gen ps2 software was inferior to DC).

GC was again made out to be inferior (marketing) to ps2 (and xbox) regardless of how close the hardware actually was.

Point stands - Nobody (not many) wants the "inferior version" regardless of cost differences. If in fact the Durango is producing inferior results compared to Orbis, a low price won't help (much). Neither will marketing, as these days buyers are better informed and if the architectures are as close as suggested, there will be no "magic tricks to unlock if given enough time" as was the message with ps2 and ps3.

This will be game, set and match rather quickly.

I'll repeat a point made on GAF: the "most powerful" console hasn't won since the SNES. I don't think the PS2 ever claimed to be most powerful, just more powerful than the DC and more supported than GC or Xbox.

 

[Deleted member 13524]

For the high end, they are.

Which the 7870M is not. I would bet anything it's not true for the high end. You can count the people that offer the 7870M or higher on one hand: Clevo, Samsung, Asus, MSI and Alienware. Which is why you don't bin the high volume parts.

Those points have nothing to do with how its binned. Those are just how the binned part is clocked and ran. Moreover, you shouldn't be looking at how its binned to see whether or not it would work in a console. You should look at why it's binned. If only 10% of the parts meet the spec you need for that part, that's your bin. The point is that you can't expect to mass produce if your yield isn't going to be a majority of your good chips. Thus, you can't look at mobile high end parts to get a guesstimate for power consumption. Yes, the console part will be downclocked (and undervolted as performance allows), but that doesn't mean it fit that 10% you're reserving for mobile high end parts.

All this is based off the awfully wrong assumption that in 2017 a mobile GPU with the performance of a HD7870M, shrinked to 14nm therefore with the size of a current Caicos, would still be high-end.

Flash news: it wouldn't.

You're assuming it's a standard Jaguar core with no customizations. We don't know that yet.

So now you think there's this special sauce in Durango's Jaguar cores that radically increases its TDP?

Do you know if LPDDR4 has the same latency as DDR3?

If we go by the previous LPDDR iterations then yes, the latencies should be similar.
Unless you think Durango is using some ultra-low latency (and therefore ultra-expensive) DDR3, when they already have embedded RAM for low-latency demands.

Have you ever heard of a tablet with a 256 bit memory interface?

And you think I was talking about 64bit channels? You think that chart from JEDEC is referring to 64bit channels?

Have you ever heard of a LPDDR implementation that uses 64bit channels?

Where did you start assuming you knew how big the eSRAM was? I missed that. And as I said, 9W is about twice what it should be for today's affordable tablets. Otherwise you're getting larger, more expensive (900+) and sacrificing battery life (especially with gaming!). Who wants to buy that?
(...)
Why would you want it to?

Gee.. I wonder who would ever pay $1000 for a tablet that could triple as a windows 8 laptop and a home+portable gaming console that has a 4-year old portfolio of dedicated AAA games with console exclusives...
Microsoft and others would be really stupid in releasing anything resembling that.

Numbers as high as 450mm^2 have been floated. This potentially meshes with rumored production issues.

It's still only some 150mm^2 total size. Still smaller than an A5X, which goes into a... tablet.

Power profile of x86 and ARM cores are different. ARM cores are optimized for mobile environments.

Did you stop reading about low-power x86 solutions in 2008?
There are x86 smartphones in the market (clocked above 1.6GHz BTW), and the solutions with Jaguar cores are, according to AMD, aimed at low-power PCs, netbooks and tablets.

 

[TheChefO]

... I don't think the PS2 ever claimed to be most powerful...

Are you kidding???

Sony had the thing banned in certain countries as it could be used for "Super Computer R&D" ...

That was Sony's angle from day 1 with ps2, regardless of the reality of the situation.


Without veering too off-topic on this, I'll say that in instances where the most powerful console didn't win, it had other mitigating factors which pulled it down.

NeoGeo - expensive console and ungodly expensive games ($200per)
N64 - no optical media, poor 3rd party relations and support
xbox - poor brand reception, limited 3rd party support, battling a well entrenched competitor

These days I'd put Sony and MS as near equals in brand reception (edge to Sony), 3rd party support (edge to Sony for internal devs), media should be the same, architecture seems to be the same, online should be near the same.

All else equal, the more powerful box will win.

 

[DuckThor Evil]

I'll repeat a point made on GAF: the "most powerful" console hasn't won since the SNES. I don't think the PS2 ever claimed to be most powerful, just more powerful than the DC and more supported than GC or Xbox.

PS2 didn't launch at the same time as those other 2 though. N64 also launched a lot later than PS1 and cartridges killed it.

 

[TheChefO]

Providing there is nothing that counteracts the apparent inferiority of the hardware- e.g. the "uniqueness" factor that lead the Wii (in combination with a comparatively low price) steal the show last generation.

Kinect 2.0 may be what MS is banking on.

Still, I would personally favor a more robust and powerful gaming machine over one that is going well out of its way to be the most unique 'special' option out there- at the cost of hardware potency.

Perhaps, but from what I understand, move 2.0 looks to be roughly on equal footing with kinect 2.0 so I wouldn't expect much of an edge there ...

It then boils down to IP content for casuals, and next to Nintendo, Sony is in a much stronger position to capture the casual audience than MS.

Again - Assuming the leaked specs are accurate.

 

[Deleted member 11852]

Durango Move Engines

VGleaks have now published some information on Durango's Move Engines.

The Durango GPU includes a number of fixed-function accelerators. Move engines are one of them.

Durango hardware has four move engines for fast direct memory access (DMA)

This accelerators are truly fixed-function, in the sense that their algorithms are embedded in hardware. They can usually be considered black boxes with no intermediate results that are visible to software. When used for their designed purpose, however, they can offload work from the rest of the system and obtain useful results at minimal cost.

http://www.vgleaks.com/world-exclusive-durangos-move-engines/

So, assuming info correct of course, we know what these are. I must say they don't look ground-breaking but this sort of hardware can sometimes be used in interesting ways.

 

[shiznit]

You guys are missing one crucial point about the PS2s dominance:

It was the first console that could play DVDs, and it came out at the perfect time for the VHS to DVD transition, which was a once in a century upgrade similar to going from horses to cars and it took full advantage.

There are no parallels between the reasons for the PS2 victory and the coming gen, just because the PS2 won despite not being the most powerful doesn't mean Durango can.

 

[dumbo11]

It was the first console that could play DVDs, and it came out at the perfect time for the VHS to DVD transition, which was a once in a century upgrade similar to going from horses to cars and it took full advantage.

There are no parallels between the reasons for the PS2 victory and the coming gen.

Personally, I'd suggest anyone reading the thread to go back and look at the original durango design brief... (that funny powerpoint presentation)

It explains the leaked design of Durango quite well, along with how MS expect to sell it.

 

[anexanhume]

All this is based off the awfully wrong assumption that in 2017 a mobile GPU with the performance of a HD7870M, shrinked to 14nm therefore with the size of a current Caicos, would still be high-end.

Flash news: it wouldn't.

You're now completely conflating two different topics: yield as it refers to part binning with part performance classification at a completely different time. These two topics don't have anything to do with another in this case. How a 7770 is regarded in 3 to 4 years has nothing to do with how it yields at a specific performance target today.

So now you think there's this special sauce in Durango's Jaguar cores that radically increases its TDP?

No, I'm pointing out that you're operating on incomplete info, yet you're using vanilla jaguar cores to make your comparison. It's probably valid, but may not be.

If we go by the previous LPDDR iterations then yes, the latencies should be similar.
Unless you think Durango is using some ultra-low latency (and therefore ultra-expensive) DDR3, when they already have embedded RAM for low-latency demands.

How can you guarantee performance if you can't guarantee the same latency? I've never heard of a console changing memory types during its lifetime yet you seem to be suggesting that you can make part substitutions with no required change to game code, or no noticeable performance impacts. How can you guarantee that?

And you think I was talking about 64bit channels? You think that chart from JEDEC is referring to 64bit channels?

Have you ever heard of a LPDDR implementation that uses 64bit channels?

Where did I say you were talking about any size channel. I was merely pointing out what the console will likely require and asking you to provide precedent for such a configuration in a mobile/tablet environment.

Gee.. I wonder who would ever pay $1000 for a tablet that could triple as a windows 8 laptop and a home+portable gaming console that has a 4-year old portfolio of dedicated AAA games with console exclusives...
Microsoft and others would be really stupid in releasing anything resembling that.

Surface RT is selling poorly and there's nothing to suggest the market is clamoring for the Pro. It also only gets 4 hours battery life doing normal windows stuff. How long do you expect to game? It won't be more than 2 hours in a similar setup.

It's still only some 150mm^2 total size. Still smaller than an A5X, which goes into a... tablet.

Which had noted heat concerns and was phased out in 6 months. Also, along with the upgraded display, it required a 70% larger battery.

Did you stop reading about low-power x86 solutions in 2008?
There are x86 smartphones in the market (clocked above 1.6GHz BTW), and the solutions with Jaguar cores are, according to AMD, aimed at low-power PCs, netbooks and tablets.

Conflating again. I said you can't compare x86 and ARM when it comes to die size, TDP and frequency. I said nothing about whether they were or could be existing low power x86 implementations that could go in tablets.

 

[Deleted member 13524]

You're now completely conflating two different topics: yield as it refers to part binning with part performance classification at a completely different time. These two topics don't have anything to do with another in this case. How a 7770 is regarded in 3 to 4 years has nothing to do with how it yields at a specific performance target today.

Nah, you just refuse to admit that a GPU with as many transistors as a HD7770 today is a GPU that could fit into a mobile SoC in 4 years using 14nm.

No, I'm pointing out that you're operating on incomplete info, yet you're using vanilla jaguar cores to make your comparison. It's probably valid, but may not be.

Okay, so then you agree that it could happen (which was my argument from the beginning), instead of flat-out insisting that it's completely impossible to happen, as you did a couple of posts ago.

Good.

How can you guarantee performance if you can't guarantee the same latency? I've never heard of a console changing memory types during its lifetime yet you seem to be suggesting that you can make part substitutions with no required change to game code, or no noticeable performance impacts. How can you guarantee that?

As long as latency is the same or lower and bandwidth is the same or higher, there's no reason why it wouldn't work or even result in any noticeable difference in performance.

Actually, there are cases where not just the memory but even the CPU has changed. DS -> DSi, for example.

Where did I say you were talking about any size channel. I was merely pointing out what the console will likely require and asking you to provide precedent for such a configuration in a mobile/tablet environment.

When you assumed that a 4-channel setup would result in a 256bit bus.
No point in making excuses now.

Surface RT is selling poorly and there's nothing to suggest the market is clamoring for the Pro. It also only gets 4 hours battery life doing normal windows stuff. How long do you expect to game? It won't be more than 2 hours in a similar setup.

No, more like 3.5-4 hours.
Which is how much a 3DS and a Vita can do, actually.

Which had noted heat concerns and was phased out in 6 months. Also, along with the upgraded display, it required a 70% larger battery.

Heat concerns? How many ipad 3s were returned due to heat concerns? Please elaborate.
The ipad 4 with a less consuming SoC came to prove that the upgraded battery is largely a consequence of the more powerfull LED backpanel required to light the higher-density display.

Conflating again. I said you can't compare x86 and ARM when it comes to die size, TDP and frequency. I said nothing about whether they were or could be existing low power x86 implementations that could go in tablets.

No, you said "ARM are optimized for mobile" as if there couldn't be x86 that are equally optimized for mobile.
But even after changing your argument, you're still wrong. Medfield is comparable to most 2011/12 ARM SoCs in size, TDP and frequency.
Jaguar, coming from Brazos, shouldn't be too far from a Cortex A15 or a 2 GHz Medfield.

 

[anexanhume]

At this point, I don't see any point in continuing the discussion. You're either purposely ignoring my points or shifting to different topics that get away from the original intent of the discussion. I stand by my initial assertions: Durango will not be less than a 120W system and there will be no Durango tablet.

 

[bkilian]

Those days are long gone. Nvdidia 1D shader FLOPS were superior to AMD 5D and 4D shader FLOPS but GCN has 1D shader, too:

AMD Radeon HD5870 (VLIW5 architecture, 5D shader) has 2.7 TFLOPS theoretical peak performance
AMD Radeon HD7850 (GCN architecture, 1D shader) has 1.7 TFLOPS theoretical peak performance

Nevertheless the HD7850 is easily faster than the HD5870 in real world scenarios. So it definitely makes sense to use a Nvidia 1D shader GPU in a 2011 devkit for a console that will be using 1D shaders in 2013.

There was never a nvidia GPU in any devkit from MS that was released after the 360 launched, including rumored alpha devkits.

 

[Averagejoe]

You guys are missing one crucial point about the PS2s dominance:

It was the first console that could play DVDs, and it came out at the perfect time for the VHS to DVD transition, which was a once in a century upgrade similar to going from horses to cars and it took full advantage.

There are no parallels between the reasons for the PS2 victory and the coming gen, just because the PS2 won despite not being the most powerful doesn't mean Durango can.

Is different in many ways.

While all people think the PS2 was less powerful and is true,it was also the first to launch,the xbox and GC did not launch on early 2000,they launch on late 2001.

The PS2 was the graphic king for 20 months that is almost 2 years,when the PS2 launched there wasn't a more powerful console than it,and it remain like that for 20 months.

This generation is different both units will release the same year apparently and one will be more powerful.

Is hard to tell considering what happen this generation.

 

[dobwal]

http://develop.scee.net/files/presentations/gdc2009/DeferredLightingandPostProcessingonPS3.ppt

See slide 11.

Explain to me how those slides invalidate the argument below.

No it doesn't. A cpu with gpu like processors helping out a gpu leads to better gpu performance. It doesn't mean it leads to a performance level greater than if those gpu like processors actually sat on the gpu.