Baseless Next Generation Rumors with no Technical Merits [post E3 2019, pre GDC 2020] [XBSX, PS5]

Status
Not open for further replies.
Its not very likely its laptop APU for single fact that it uses fastest possible GDDR6 chips Samsung can provide. 16GB of GDDR6 as system memory? That is suicidal and actually bad for laptops (latency, costs, TDP, all around pretty bad)

I agree. The memory is either being misreported by the benchmark or it is some other device like a console, not necessarily PS5 or Xbox. Although I doubt it would be a low cost Chinese console as well. You don't put 16GB GDDR6 in a low cost part so this definitely seems like a mid to high tier part.

Maybe AMD is going to release a gaming laptop with this type of setup, but we have not heard a single rumor about something like this coming from AMD.
 
Is there really an issue with latency? That's what the caches are for. If PS4 had DDR instead of GDDR at equivalent BW, would it run faster due to lower latency?
I think you would probably have less issues with latency in consoles as per default. My thinking is, putting GDDR6 in laptops, especially as these seem to be very fast chips (>16Gbps?), would make it completely counterproductive.

I guess only product where you would want to use such high amount of RAM at such bandwidth is either console or actual GPU (Quadro 5000). Putting it in laptop defeats the purpose.
 
I agree. The memory is either being misreported by the benchmark or it is some other device like a console, not necessarily PS5 or Xbox. Although I doubt it would be a low cost Chinese console as well. You don't put 16GB GDDR6 in a low cost part so this definitely seems like a mid to high tier part.

Maybe AMD is going to release a gaming laptop with this type of setup, but we have not heard a single rumor about something like this coming from AMD.
Yes, I agree. It's possible.
 
Its not very likely its laptop APU for single fact that it uses fastest possible GDDR6 chips Samsung can provide. 16GB of GDDR6 as system memory? That is suicidal and actually bad for laptops (latency, costs, TDP, all around pretty bad)

Suicidal? Look up gaming laptops using the RTX 2070, for example. They use dedicated GDDR6 memory to feed the GPU.

It's not exactly rare or even uncommon for gaming laptops to have dedicated graphics memory to feed the GPU cores with a separate memory pool to feed the CPU cores...just like the one on Userbench that has since been deleted. Laptop makers have been doing this for well over a decade now.

Regards,
SB
 
Suicidal? Look up gaming laptops using the RTX 2070, for example. They use dedicated GDDR6 memory to feed the GPU.

It's not exactly rare or even uncommon for gaming laptops to have dedicated graphics memory to feed the GPU cores with a separate memory pool to feed the CPU cores...just like the one on Userbench that has since been deleted. Laptop makers have been doing this for well over a decade now.

Regards,
SB
Can you link me laptop with 16GB of GDDR RAM? Not GPU GDDR + dedicated DDR4 RAM as system RAM, completely unified 16GB of RAM.
 
Can you link me laptop with 16GB of GDDR RAM? Not GPU GDDR + dedicated DDR4 RAM as system RAM, completely unified 16GB of RAM.

You first claimed there's no way a laptop would use GDDR6 because of latency, TDP and costs. There are plenty of laptops with GDDR6 right now, and most probably 2 years from now it'll be hard to find a laptop with discrete GPU that has no GDDR6.
So the TDP and costs argument is out the window.
As for latency, there is indeed a custom PC that runs the regular Windows 10 distribution made by Subor that uses GDDR5 as main memory. The effects of higher latency can be downsized by buffers.

The advantage of regular DDR over GDDR is simply production cost. It's definitely not power-per-bandwidth, nor absolute performance. It's not even much better on memory density nowadays.
A single chip of 14GT/s GDDR6 (56GB/s) would provide more bandwidth than eight chips of DDR4 3.2GT/s (51GB/s).
 
You first claimed there's no way a laptop would use GDDR6 because of latency, TDP and costs. There are plenty of laptops with GDDR6 right now, and most probably 2 years from now it'll be hard to find a laptop with discrete GPU that has no GDDR6.
So the TDP and costs argument is out the window.
As for latency, there is indeed a custom PC that runs the regular Windows 10 distribution made by Subor that uses GDDR5 as main memory. The effects of higher latency can be downsized by buffers.

The advantage of regular DDR over GDDR is simply production cost. It's definitely not power-per-bandwidth, nor absolute performance. It's not even much better on memory density nowadays.
A single chip of 14GT/s GDDR6 (56GB/s) would provide more bandwidth than eight chips of DDR4 3.2GT/s (51GB/s).
First of all, no, I never claimed that there is no GDDR6 memory in laptops, thats absurd. I claimed there is no way laptop will be APU based and use SINGLE pool of 16GB GDDR6 RAM because it is counterproductive and makes 0 sense. Especially while using, what appears to be, fastest and most expensive GDDR6 chips that are not yet in mass production (18Gbps).

So no, its not out of the window.

Second point, you are mixing APU (which this benchmark was) and discrete GPU laptops and claiming "There will be hard to find laptop with discrete GPU that has no GDDR6". I agree, laptops with discrete GPUs such as 2070 will continue to use GDDR6 memory, which they always have, but system memory will not be 16GB of GDDR6 unified RAM and thats my entire point.

APUs in laptops have always been used as cost effective solution to even more underpowered Intel HD graphics. Not one AMD APU in laptops had anything remotely similar to their entry level PC discrete GPU, let alone top of what they are offering with double the RAM. This is why this leak makes 0 sense for laptop based APU.
 
I claimed there is no way laptop will be APU based and use SINGLE pool of 16GB GDDR6 RAM because it is counterproductive and makes 0 sense.
(...)
Not one AMD APU in laptops had anything remotely similar to their entry level PC discrete GPU, let alone top of what they are offering with double the RAM. This is why this leak makes 0 sense for laptop based APU.

Kaveri has a 256 bit DDR3/GDDR5 controller, SK Hynix was developing with JEDEC the GDDR5M standard for implementation in SO-DIMMs (= laptops) and Elpida was starting production on those, before it went bankrupt.

GDDR5M didn't come through because AMD's bulldozer ended up being terrible so OEMs avoided the company's CPUs and APUs, and Intel didn't care for higher performing iGPUs until very recently.
With AMD being on OEMs' good graces again thanks to Zen, I wouldn't be surprised if we saw GDDR6M being developed for future use with 35-45W Zen APUs.

As for being 16GBytes instead of 12 or 8, it's simply a capacity that is forward-thinking.
 
Kaveri has a 256 bit DDR3/GDDR5 controller, SK Hynix was developing with JEDEC the GDDR5M standard for implementation in SO-DIMMs (= laptops) and Elpida was starting production on those, before it went bankrupt.

GDDR5M didn't come through because AMD's bulldozer ended up being terrible so OEMs avoided the company's CPUs and APUs, and Intel didn't care for higher performing iGPUs until very recently.
With AMD being on OEMs' good graces again thanks to Zen, I wouldn't be surprised if we saw GDDR6M being developed for future use with 35-45W Zen APUs.

As for being 16GBytes instead of 12 or 8, it's simply a capacity that is forward-thinking.
So, there is no laptop available with single pool of GDDR6 memory, am I right?

And Flute is actually 18Gbytes GDDR6 with much higher TDP then 35-45W (I assume, judging by CPU and GPU code that matches Gonzalo, at least 180W). Gonzalo is APU that scored 20K in FS...That is 5700XT + 3700X territory.
 
So, there is no laptop available with single pool of GDDR6 memory, am I right?

And Flute is actually 18Gbytes GDDR6 with much higher TDP then 35-45W (I assume, judging by CPU and GPU code that matches Gonzalo, at least 180W). Gonzalo is APU that scored 20K in FS...That is 5700XT + 3700X territory.

First you say GDDR6 wouldn't make sense in a laptop because of TDP and latency. Then you're proven wrong (GDDR6's power efficiency per bandwidth is much larger than DDR4's) and you change to "there are no APUs with GDDR because it makes no sense". Then you're proven wrong (Kaveri has a GDDR5 controller) and now it's "there's no laptop available with a single pool of GDDR6".

No, there is no laptop available with a single pool of GDDR6 in the market.
Your previous arguments were always on the side of "it makes no sense for a future APU to use GDDR because XYZ". When all your XYZ arguments were defeated you changed to "b-but there's no actual laptop in the current market with it".
Well no shit, there's no current laptop with it, yes we can all go to wikipedia and see the formal list of AMD APUs to check that.

Flute is a future APU, not a current one. It's a future, unreleased and unnanounced APU that will be using future unreleased GDDR6 chips on one or more future unreleased platforms and people have been evaluating the possibility of this being a laptop solution.
Yes, it's probably a console APU and no one has questioned that probability. Regardless, you haven't presented any valid technical argument as to why this "wouldn't make sense" in a laptop. It could very well be a chip for a premium laptop with premium graphics, akin to Intel's 100W Kaby Lake G.
The gaming laptop market is flooded with solutions carrying 45W Intel CPUs and 65-100W nVidia GPUs, so a ~80-100W Zen+Navi solution using 16GB of GDDR6 to compete on that market wouldn't be crazy.
 
First you say GDDR6 wouldn't make sense in a laptop because of TDP and latency. Then you're proven wrong (GDDR6's power efficiency per bandwidth is much larger than DDR4's) and you change to "there are no APUs with GDDR because it makes no sense". Then you're proven wrong (Kaveri has a GDDR5 controller) and now it's "there's no laptop available with a single pool of GDDR6".
You're being unfairly argumentative here. It was very clear to me what AbsoluteBeginner meant from his initial remark.

Its not very likely its laptop APU for single fact that it uses fastest possible GDDR6 chips Samsung can provide. 16GB of GDDR6 as system memory? That is suicidal and actually bad for laptops (latency, costs, TDP, all around pretty bad)
Three reasons were cited for why GDDR6 as system memory, the only unified memory pool in the device, was a bad idea. The ensuing arguments should focus on that, as you end your post with. Implications of moving goalposts aren't valid here; people seem to have just missed the original premise and presented cases AbsoluteBeginner has ended up arguing tangentially such as mixed memory systems.
 
First you say GDDR6 wouldn't make sense in a laptop because of TDP and latency. Then you're proven wrong (GDDR6's power efficiency per bandwidth is much larger than DDR4's) and you change to "there are no APUs with GDDR because it makes no sense". Then you're proven wrong (Kaveri has a GDDR5 controller) and now it's "there's no laptop available with a single pool of GDDR6".

No, there is no laptop available with a single pool of GDDR6 in the market.
Your previous arguments were always on the side of "it makes no sense for a future APU to use GDDR because XYZ". When all your XYZ arguments were defeated you changed to "b-but there's no actual laptop in the current market with it".
Well no shit, there's no current laptop with it, yes we can all go to wikipedia and see the formal list of AMD APUs to check that.

Flute is a future APU, not a current one. It's a future, unreleased and unnanounced APU that will be using future unreleased GDDR6 chips on one or more future unreleased platforms and people have been evaluating the possibility of this being a laptop solution.
Yes, it's probably a console APU and no one has questioned that probability. Regardless, you haven't presented any valid technical argument as to why this "wouldn't make sense" in a laptop. It could very well be a chip for a premium laptop with premium graphics, akin to Intel's 100W Kaby Lake G.
The gaming laptop market is flooded with solutions carrying 45W Intel CPUs and 65-100W nVidia GPUs, so a ~80-100W Zen+Navi solution using 16GB of GDDR6 to compete on that market wouldn't be crazy.
Nah, I think I was being very clear on what I meant and you just tried to twist my posts how you see fit. My argument was very simple, we could have stopped at entire point of an APU and concluded that benched APU is not laptop, but there are literally 5 other things that point in that direction.

I have no problem with people arguing 16GBs of absolute fastest RAM, yet to be found in mass products, will be used as single pool of RAM in laptop. Nor do I have a problem with people arguing laptop APU will be made out of 16 threaded Zen2 and Navi GPU clocked at 1.8GHZ. I have a problem with people twisting my posts on B3D and presenting them like I don't know discrete GPUs in laptop use GDDR memory.

You could have said "No, there are XY amount of laptops that use 16GB of GDDR and APUs made out of top of the shelf AMD GPUs and CPUs" but you chose to dissect my argument into

a) Laptops actually having GDDR memory which feed discrete GPU
b) Back in 2014 GDDR5m memory controller being developed by AMD

Which both have nothing to do with laptops designed with APU in mind.

Entire point of AMDs APUs is bringing cost effective solution on single die vs Intels HD graphics, which are what most people are actually buying. They never aimed at top performing discrete laptops with their APUs because, technically, they make 0 sense. Thats all. Economically it makes very little sense because these chips would be expensive, hard to cool and would bring console drawbacks (single pool of high latency RAM) into environment where they are absolutely not needed.
 
How large is this gaming laptop market? I can get 3rd parties creating such products for enthusiasts but I have serious problems imaging this is huge enough for AMD to design specific APUs here.

Personally I have always despised gaming on a laptop just for the heat and noise factor with the MacBook pros I've used over the years so I couldn't even imagine to buy one with even more heat/noise for gaming.
 
My argument was very simple, we could have stopped at entire point of an APU and concluded that benched APU is not laptop, but there are literally 5 other things that point in that direction.
No, we cannot conclude that because it's not a valid conclusion at all.


a) Laptops actually having GDDR memory which feed discrete GPU
I wasn't the only or even the first person to make that statement.
Which by the way is a valid statement because of your claims of TDP limitations. If anything, using GDDR in an APU would decrease power consumption and total heat output for the same bandwidth, or much more bandwidth for the same power/thermals.
And it's something you apparently still didn't get.


b) Back in 2014 GDDR5m memory controller being developed by AMD
No, it's not "being developed". Kaveri is a fully matured product that released 5 years ago. It's a x86 APU with an embedded GDDR5 controller.


Which both have nothing to do with laptops designed with APU in mind.
There's an entire lineup of laptop Kaveri chips that denies this claim.
Plus, had you read the IEEE article I linked to you'd know that GDDR5M was developed for use in SO-DIMMs which go into laptops, not desktops.


Economically it makes very little sense because these chips would be expensive, hard to cool and would bring console drawbacks (single pool of high latency RAM) into environment where they are absolutely not needed.
1 - Expensive chips can go into premium laptops which are plenty. That's why Kaby-G and Vega 12 laptops exist, with expensive HBM2 stacks and interposers/EMIB.

2 - GDDR chips are "hard to cool"? Please do provide a even a single reference to this.

3 - Honestly, you'd need to actually know how the real latency is calculated to make such an assumption, which I suspect you don't.
The suggestion that GDDR would be terrible for the system memory in PCs because latency is mostly a poorly informed internet myth created by people who think the "CAS latency" number means the end-all of memory latency.
The effect of CAS latency cycles on actual latency depends directly on command rate. Typical DDR4 CAS latency is 16 cycles. The same value for DDR2 was 4 cycles. You don't see developers protesting to get DDR2 back in the market. The real latency has been steadily going down because of command rates going up.
This was properly clarified many times back in 2013-2014 when people started claiming the PS4 with GDDR5 would suffer from horrible latency compared to the XBOne with DDR3, yet I guess the myth still persists.




How large is this gaming laptop market? I can get 3rd parties creating such products for enthusiasts but I have serious problems imaging this is huge enough for AMD to design specific APUs here.

Personally I have always despised gaming on a laptop just for the heat and noise factor with the MacBook pros I've used over the years so I couldn't even imagine to buy one with even more heat/noise for gaming.
At the end of 2018 the gaming laptop market was worth $12B/year and it has grown over 10x in the last 5 years:

GoXm4Wt.png


It's a huge market right now, to the point of every major to medium laptop maker having their own gaming brand.
So yes, it's very well worth the effort of developing a custom SoC for high-end laptops.
Even more considering that AMD charged Subor around $65M for their Z-Plus SoC IIRC.
 
Entire point of AMDs APUs is bringing cost effective solution on single die vs Intels HD graphics, which are what most people are actually buying. They never aimed at top performing discrete laptops with their APUs because, technically, they make 0 sense. Thats all. Economically it makes very little sense because these chips would be expensive, hard to cool and would bring console drawbacks (single pool of high latency RAM) into environment where they are absolutely not needed.
You raise a number of points still not justified effectively IMO. You say cost is a concern - what's the price difference between an APU using two different memories with two different buses versus using one pool of the more expensive stuff? Is that clearly quantitatively of the cards. You state high-end laptops are a no-go; ToT's link suggests very much otherwise, and shows a growth that AMD would have been aware of when (if) designing a high-end APU. And you state latency is a problem again. Is there any actual evidence that it is? Have you a link to an investigation of a PC OS running on GDDR versus DDR with significant penalties versus the benefits of the increased bandwidth?
 
No, we cannot conclude that because it's not a valid conclusion at all.
Agree to disagree.


wasn't the only or even the first person to make that statement.
Which by the way is a valid statement because of your claims of TDP limitations. If anything, using GDDR in an APU would decrease power consumption and total heat output for the same bandwidth, or much more bandwidth for the same power/thermals.
And it's something you apparently still didn't get.
No, you are again dissecting my argument and picking and choosing what to make of it. If GDDR as single pool of memory is such a novel and great idea, why hasnt such laptop been released? I am waiting this list because I suspect its non existant at this point. There is a reason you see gaming laptops with 4-8GB of RAM and 16-32GB of DDR4 and that is why there is not a SINGLE laptop you can show me with gaming console grade APU and single pool of GDDR6 RAM.

Not only is 16GB of GDDR6 memory relatively expensive, it also uses considerably more watts then DDR4 memory and gives NO benefits in overall system design.


No, it's not "being developed". Kaveri is a fully matured product that released 5 years ago. It's a x86 APU with an embedded GDDR5 controller.
There's an entire lineup of laptop Kaveri chips that denies this claim.
Plus, had you read the IEEE article I linked to you'd know that GDDR5M was developed for use in SO-DIMMs which go into laptops, not desktops.
And yet, here we are still waiting APU that can match PS4s 2013 SOC. But yea, we are getting Navi XT + 8 core Zen2 with 16GB of GDDR6 relatively soon.

1 - Expensive chips can go into premium laptops which are plenty. That's why Kaby-G and Vega 12 laptops exist, with expensive HBM2 stacks and interposers/EMIB.

...And yet best AMD APU based laptops are 4 core parts with severly cut down Vega chips that are using dual channel DDR4 memory. Nothing, absolutely nothing close to full Navi chip paired with 16GB of GDDR6.

2 - GDDR chips are "hard to cool"? Please do provide a even a single reference to this.
16GB of DDR4 (Two modules) will use less then 6W. For 16GB of GDDR6 you are looking 30W+. You do the math.

3 - Honestly, you'd need to actually know how the real latency is calculated to make such an assumption, which I suspect you don't.
The suggestion that GDDR would be terrible for the system memory in PCs because latency is mostly a poorly informed internet myth created by people who think the "CAS latency" number means the end-all of memory latency.
The effect of CAS latency cycles on actual latency depends directly on command rate. Typical DDR4 CAS latency is 16 cycles. The same value for DDR2 was 4 cycles. You don't see developers protesting to get DDR2 back in the market. The real latency has been steadily going down because of command rates going up.
This was properly clarified many times back in 2013-2014 when people started claiming the PS4 with GDDR5 would suffer from horrible latency compared to the XBOne with DDR3, yet I guess the myth still persists.
No, I have several reasons why SINGLE POOL of GDDR memory is bad decision in laptops. First of all, you are providing far too much of fast and expensive memory for product that has no use for it. Second, because having DDR4 as system memory has only benefits in laptop space, and no drawbacks. You are getting memory with lower latency, SIGNIFICANTLY lower TDP and costs per chip for raw bandwidth that cut down Zen2 will never even need.

Dedicated PC GPUs with 8GB of GDDR6 on 256bit bus have 0 issues with memory or bandwidth at 4K resolutions, so what is the purpose of creating APU that would be paired with 16GB of GDDR6 in a laptop?

Note, I am talking strictly about APUs here. Not laptop with discrete GPUs, I am talking about APUs. Parts whose sole purpose for last 6 years was to provide acceptable 720p-1080p performance compared to integrated Intel solutions.

In the end, I rest my case. There is a reason why you haven't named one single laptop that has PC gaming grade APU paired with gobbs of GDDR as a system RAM. You haven't provided that because you are refusing to acknowledge it makes no sense to do so, from AMD nor manufacturers POV. Because that chip would be very big, very hot and very expensive for laptop, yet would still fell severely short compared to dedicated solution given by Nvidia. This is why APU playground is 720p/1080p low/medium detail gaming, and not Quadro 5000 and this is why there is no real need for such APU.

When HBM becomes cheaper and widely used as other types of RAM, then you will see raise in higher performing APUs. Until then, there is literally no need for them.
 
You raise a number of points still not justified effectively IMO. You say cost is a concern - what's the price difference between an APU using two different memories with two different buses versus using one pool of the more expensive stuff? Is that clearly quantitatively of the cards. You state high-end laptops are a no-go; ToT's link suggests very much otherwise, and shows a growth that AMD would have been aware of when (if) designing a high-end APU. And you state latency is a problem again. Is there any actual evidence that it is? Have you a link to an investigation of a PC OS running on GDDR versus DDR with significant penalties versus the benefits of the increased bandwidth?
No, I never said high performance laptops where no go. I said high performance APUs in laptops are no go, clear distinction. There are laptops out there with Quadro 5000 GPUs and 16GB of GDDR6 dedicated to graphics. There are alot of high performance laptops out there, but there are none high performing APUs and I gave the reasons why.

Creating high performance APU for laptop poses several design problems. As would be in this case, you would either have to go for very high amount of expensive, fast and power hungry RAM or you would split the memory.

Splitting the memory, like 100% of high performing laptops are doing these days, gives you ability to dedicate specific amount of fast RAM required for gaming tasks with second pool of RAM acting as system memory.

This is because apart from gaming, you will most likely spend more then 50% of your time working on actual productive things (or surfing or whatever) and design where 16GBs of GDDR6 in this case are used for 30 Chrome tabs, Netflix and Photoshop is just downright horrible. You are basically wasting fast, expensive RAM for tasks any memory will do ALONG with wasting considerably more power and pushing costs of laptop higher. Point is, DDR4 :

a) uses much less power then GDDR6 per GB
b) costs much less then GDDR6 per GP
c) sacrifices no performance in general system performance

And that is a reason why laptops have split RAM, especially in high performance gaming sector. If anything, single pool memory is used diametrically opposite, on lower end of laptop spectrum.

In case you go with single pool of GDDR6, you design yourself in a corner from beginning with higher costs and lower battery life without perceived performance benefit. In laptop environment, when Nvidia already has perf/watt advantage, that is already a lost battle before it has even started. This is why I just ask what would be the point of single pool of GDDR6 in laptop. It brings no benefits in this space and if it did, well surely there would at least be few percent of these on market. Instead, there is quite literally not a single one in existence.

Benefit of lower end APUs (which, all of them are and which is actually their main purpose) is the fact that you do not have to use fastest RAM out there, but DDR4 will do. It will give enough BW for most novice GPU tasks and plenty of battery life. Obviously, AMD is placing their APUs at lower end of spectrum because their advantage over Intel in this field is clear, but bringing APU into high performance space would lose them ALL the benefits, while further increasing their deficit in areas where they are already behind.
 
Status
Not open for further replies.
Back
Top