AMD May Not Be Trying To Compete With Nvidia

[Prysin]

43 minutes ago, FirstArmada said:

 

Or maybe AMD rapes Nvidia market share , Nvidia goes bankrupt , Intel buys them out and voila

 

( ͡° ͜ʖ ͡°) 

lol nope. Not happening.

Nvidia has the cash reserve to survive for a decade or two with negative revenue before needing to file for chapter 11

[MEC-777]

I highly doubt we have seen the end of AMD in the high-end GPU market. The development of mainstream and high-end is usually staggered now. 6 months to a year after the 1080, we will see the 1080ti and I bet AMD will have something to succeed the Furry X down the road, just the same.

[patrickjp93]

1 hour ago, Prysin said:

EVGA given Radeon tech. Nvidia wouldnt stand for that for even a moment. We know what happened when XFX tried to do that.

MSI and ASUS, sure, but both of these companies would have put out teasers by now. They are tight lipped, but both ASUS and MSI loves to tease.

 

To be honest, the only tight lipped companies in this industry is probably PNY and Sapphire, and even they doesnt keep everything under a tight lid.

 

Intel entering the dGPU market in any segment is but a farfetched dream. They are nowhere near ready and wont be for YEARS. Remember, they are not battling their ability to produce, they are battling ESTABLISHED BRAND NAMES.

 

By admitting to use Radeon tech, the market would be in turmoil and consumers would be confused. If Intel made "better"products then Radeon, it would sink the RTG, if they made worse, it would cause Intels project to flop.

 

Honestly, i think Intels usage of Radeon technology is going to be in HPC or atleast not outside of SoC stuff.

EVGA has no exclusivity binding to Nvidia, and even if it did, then Intel could talk to Sapphire.

 

Intel is as established as you get. It's a different segment of the same IC industry. No one batted an eye when Intel said it would make accelerators. Why should we bat an eye at the idea it could make dGPUs when licensing AMD graphics IP?

 

No, consumers wouldn't be confused at all. Using IP is not the same as using the competition's architecture. And Intel is a partner in a ton of open standards. If Intel backs GPUOpen, there's no confusion and no turmoil. And Intel just has to make the biggest enthusiast dies. It doesn't have to mess with AMD's mainstream, medium-die goals.

 

lol nope. Not happening.

Nvidia has the cash reserve to survive for a decade or two with negative revenue before needing to file for chapter 11

That's up to the shareholders and board of directors to decide, not Huang. If Huang has an utter failure and Intel offers a buyout, I'm sure the board would leap at it. Intel has low volatility and lots of market growth potential with the IP injection.

[Tomsen]

On 28/4/2016 at 2:54 PM, Adored said:

The point I was making in the video is that AMD could have released a card to compete with GP104.  It could have had HBM, possibly even HBM2, and it could have cost the same as what the GTX 1080 is gonna cost (think north of $600).

 

It would have been faster, but most of you would still have bought the 1080 instead.  This has been seen time and time again in the past with cards like the 560 Ti vs the 6950 and the 460 vs the 6850.  Both times where AMD was faster yet completely outsold anyway.  The 460 outsold the 6850 by 4-1.

 

AMD is a company out to make money.  Nvidia is too and they make more money by selling to their high-end buyers over and over.  680, 690, Titan, 780, 780 Ti, Titan Z, 980, Titan X, 980 Ti...now the 1080 another 25% faster than the 980 Ti.

People just keep lapping up those 30% increases at more $$$ but that's a market that will eventually run out of legs.  It's unsustainable.

Nvidia is bottlenecked by wafer allocation at TSMC so they have to pick and choose which cards get made first.  Those gigantic P100 Teslas at $10K a piece are way ahead of any gamer GPU and using up wafers like nobodies business.  AMD will gain a lot of market share in mobile while Nvidia does this, it's really very simple strategy.

First of; Why can't polaris 10 compete against GP104? Seen some of the interesting news with the new architecture (like the Primitive Discard Accelerator unit), could give AMD the performance/mm advantage.

Sure, they could release a HBM1 product with polaris, but that wouldn't make sense. Then AMD would have to support 3 different memory technologies (GDDR5, HBM1 and HBM2). You are not going to see HBM2 this early, but that is due to many other reasons (much higher cost per GB, low volume of both interposers and HBM2 memory).

The problem is that such a product, you can't use the defect for lower-end SKUs.

 

The real problem is getting the volume out through the OEM channels. Most people don't go out buying a GPU, but go out and by a new system.

 

High-end GPU's aren't where the money is at. Mainstream is. High-end is just a halo products.

 

--

 

So, there are no chances for Nvidias GP104 to launch even close to AMD polaris? I'm thinking Nvidia is doing both simultaneosuly, else they can't get the product out in time.

They don't need a whole lot of volume for the GP100, just need to get it starting (as the verification stage is longer for enterprise components) in time.

 

10 hours ago, Prysin said:

given that Nvidia is pushing their high end stuff first and AMD is holding back VEGA (their next Fury replacement), i'd say AMD is simply trying to grab the "cash cow" of the market before trying to seduce ethusiasts which get pissy and salty over meaningless numbers.

Remember Nvidia said availability in Q1 2017 for OEM server-channels. Vega is supposed to land as the same time (for consumers).

 

 

--------------------

 

Gee guys, Intel isn't going to enter the dGPU market. Way to late of entry for a dying market.

[Notional]

@Tomsen Agreed on the market perspective of your post. I don't get why people think AMD won't perform on the high end market, when VEGA is officially a thing, and NVidia puts their GP100 at the same launch time as VEGA. Whatever yield NVidia is getting on P100, they will mostly go to the much higher profit market of servers than "cheap" gaming GPU's.

 

One thing though: Interposers are super simple silicon based on 65nm node (I believe). They are cheap as dirt and the yield is insanely high due to the simplicity and node used. I doubt interposers are a problem for anyone. HBM2 is the bottleneck for both companies.

[Prysin]

12 minutes ago, patrickjp93 said:

EVGA has no exclusivity binding to Nvidia, and even if it did, then Intel could talk to Sapphire.

 

Intel is as established as you get. It's a different segment of the same IC industry. No one batted an eye when Intel said it would make accelerators. Why should we bat an eye at the idea it could make dGPUs when licensing AMD graphics IP?

 

No, consumers wouldn't be confused at all. Using IP is not the same as using the competition's architecture. And Intel is a partner in a ton of open standards. If Intel backs GPUOpen, there's no confusion and no turmoil. And Intel just has to make the biggest enthusiast dies. It doesn't have to mess with AMD's mainstream, medium-die goals.

 

That's up to the shareholders and board of directors to decide, not Huang. If Huang has an utter failure and Intel offers a buyout, I'm sure the board would leap at it. Intel has low volatility and lots of market growth potential with the IP injection.

the board/shareholders will only jump if THEY make more money doing so then staying. 

Remember, shareholders are people, most of them greedy. Some of them just care about sitting in that chair in the board room long over teh company itself.

 

Intel is established as a CPU manufacturer and laptop graphics. They are not known, or even spoken of in the "great gaming rig" segment when it comes to graphics.

Your rose tinted view of Intels power are way out of balance. Intel cannot just break into a market that is dominated by brand names simply by "going in head first".

Even if Intel was the ones to make the GPU Die only in their fabs, they still would have to sell it under the Radeon brand for a LOOONG time before people would TRUST intels GPUs. It all comes down to trust, Intel has zero trust in the high end GPU market.

Their current drivers are having plenty of issues with basic Windows 10 stuff, although shit is improving rapidly. So unfortunatly, Intel doesnt have a track record with high end graphics.

 

The top end segment is ruled by Enthusiasts, we KNOW what we buy into, but we also know what NOT to buy into. Even with Intels size, i wouldnt trust them to deliver great drivers, great features and great dGPUs for a few generations. Simply because i am not going to spend hundred of dollars based of CPU merit.

 

 

Its like we see with Catepillar and their smartphones. They are known to make fucking excellent heavy machinery, but i know a couple of people who bought their so-called "rugged phones", and they broke, was slow or simply couldnt even survive light moisture before kicking the bucket.

 

Just because you are famous for making rugged stuff, doesnt mean you will succeed when you make other rugged stuff. Similarily, Intel may be famous for their CPUs, but not a single fucking CPU they have made is remotely close to the density of a GPU. The experience of their engineers when it comes to GPUs are LIMITED. Even their Iris Pro iGPU is not even comparably dense to AMD and Nvidias GPUs.

 

Look at this: GT3e vs R7 Radeon (A10 7850k iGPU)

http://www.techpowerup.com/gpudb/2532/radeon-r7-graphics

http://www.techpowerup.com/gpudb/2790/iris-pro-graphics-6200

 

R7 = 2410 million transistors, 249mm2 (TSMC 28nm)

GT3e = 189 Million transistors, 49mm2 (Intel 14nm)

Statistics wise, the GT3e isnt a match for the R7 in any way. HOWEVER, Intels IMC is eons better then any AMD IMC will be for a while (lets just say, FM2+ cannot even achieve 75% of theoretical memory bandwidth of a 2400MHz kit of RAM, The IMC in Haswell hits closer to 85-90% of theoretical bandwidth).

 

Despite a much much much smaller node, Intels iGPU isnt especially dense, per mm2 the GT3e iGPU has 3,85million transistors, while the R7 has 9.67 million transistors.

As we can CLEARLY see with kindergarden math, INTEL ISNT FAMILIAR WITH HIGH DENSITY GPUS WHICH IS WHAT YOU NEED FOR HIGH END PRODUCTS.

 

Intel beats AMDs iGPUs by simply reducing the memory bandwidth bottleneck thanks to their eDRAM and extremely efficient IMC. 

 

But please do come up with a good counter argument.

[Prysin]

33 minutes ago, Tomsen said:

First of; Why can't polaris 10 compete against GP104? Seen some of the interesting news with the new architecture (like the Primitive Discard Accelerator unit), could give AMD the performance/mm advantage.

Sure, they could release a HBM1 product with polaris, but that wouldn't make sense. Then AMD would have to support 3 different memory technologies (GDDR5, HBM1 and HBM2). You are not going to see HBM2 this early, but that is due to many other reasons (much higher cost per GB, low volume of both interposers and HBM2 memory).

The problem is that such a product, you can't use the defect for lower-end SKUs.

 

The real problem is getting the volume out through the OEM channels. Most people don't go out buying a GPU, but go out and by a new system.

 

High-end GPU's aren't where the money is at. Mainstream is. High-end is just a halo products.

 

--

 

So, there are no chances for Nvidias GP104 to launch even close to AMD polaris? I'm thinking Nvidia is doing both simultaneosuly, else they can't get the product out in time.

They don't need a whole lot of volume for the GP100, just need to get it starting (as the verification stage is longer for enterprise components) in time.

 

Remember Nvidia said availability in Q1 2017 for OEM server-channels. Vega is supposed to land as the same time (for consumers).

 

 

--------------------

 

Gee guys, Intel isn't going to enter the dGPU market. Way to late of entry for a dying market.

Sure GP104 may launch close to polaris, but i am not sure it will be as big a hit as Maxwell. I just got a feeling Nvidia will be off to a slow start this time. But time will show.

[FirstArmada]

1 minute ago, Prysin said:

the board/shareholders will only jump if THEY make more money doing so then staying. 

Remember, shareholders are people, most of them greedy. Some of them just care about sitting in that chair in the board room long over teh company itself.

 

Intel is established as a CPU manufacturer and laptop graphics. They are not known, or even spoken of in the "great gaming rig" segment when it comes to graphics.

Your rose tinted view of Intels power are way out of balance. Intel cannot just break into a market that is dominated by brand names simply by "going in head first".

Even if Intel was the ones to make the GPU Die only in their fabs, they still would have to sell it under the Radeon brand for a LOOONG time before people would TRUST intels GPUs. It all comes down to trust, Intel has zero trust in the high end GPU market.

Their current drivers are having plenty of issues with basic Windows 10 stuff, although shit is improving rapidly. So unfortunatly, Intel doesnt have a track record with high end graphics.

 

The top end segment is ruled by Enthusiasts, we KNOW what we buy into, but we also know what NOT to buy into. Even with Intels size, i wouldnt trust them to deliver great drivers, great features and great dGPUs for a few generations. Simply because i am not going to spend hundred of dollars based of CPU merit.

 

 

Its like we see with Catepillar and their smartphones. They are known to make fucking excellent heavy machinery, but i know a couple of people who bought their so-called "rugged phones", and they broke, was slow or simply couldnt even survive light moisture before kicking the bucket.

 

Just because you are famous for making rugged stuff, doesnt mean you will succeed when you make other rugged stuff. Similarily, Intel may be famous for their CPUs, but not a single fucking CPU they have made is remotely close to the density of a GPU. The experience of their engineers when it comes to GPUs are LIMITED. Even their Iris Pro iGPU is not even comparably dense to AMD and Nvidias GPUs.

 

Look at this: GT3e vs R7 Radeon (A10 7850k iGPU)

http://www.techpowerup.com/gpudb/2532/radeon-r7-graphics

http://www.techpowerup.com/gpudb/2790/iris-pro-graphics-6200

 

R7 = 2410 million transistors, 249mm2 (TSMC 28nm)

GT3e = 189 Million transistors, 49mm2 (Intel 14nm)

Statistics wise, the GT3e isnt a match for the R7 in any way. HOWEVER, Intels IMC is eons better then any AMD IMC will be for a while (lets just say, FM2+ cannot even achieve 75% of theoretical memory bandwidth of a 2400MHz kit of RAM, The IMC in Haswell hits closer to 85-90% of theoretical bandwidth).

 

Despite a much much much smaller node, Intels iGPU isnt especially dense, per mm2 the GT3e iGPU has 3,85million transistors, while the R7 has 9.67 million transistors.

As we can CLEARLY see with kindergarden math, INTEL ISNT FAMILIAR WITH HIGH DENSITY GPUS WHICH IS WHAT YOU NEED FOR HIGH END PRODUCTS.

 

Intel beats AMDs iGPUs by simply reducing the memory bandwidth bottleneck thanks to their eDRAM and extremely efficient IMC. 

 

But please do come up with a good counter argument.

But isn't this what AMD essentially did with ATI?

[Prysin]

2 minutes ago, FirstArmada said:

But isn't this what AMD essentially did with ATI?

AMD bought ATI, they didnt do dGPUs before. AMD is not selling Radeon to Intel, that is but a wet dream @patrickjp93 is fantasizing about. AMD and Intel is in licensing talks, because apparently, Intels licensing deals with NVIDIA, which is what allows them to make iGPUs in their CPUs, for now, is about to expire.

[FirstArmada]

Just now, Prysin said:

AMD bought ATI, they didnt do dGPUs before. AMD is not selling Radeon to Intel, that is but a wet dream @patrickjp93 is fantasizing about. AMD and Intel is in licensing talks, because apparently, Intels licensing deals with NVIDIA, which is what allows them to make iGPUs in their CPUs, for now, is about to expire.

Ahh i see , well if it did end up happening it would be nice to see another player in the market maybe Nvidia will stop its shenanigans and stop being dicks

[Tomsen]

11 minutes ago, Notional said:

@Tomsen Agreed on the market perspective of your post. I don't get why people think AMD won't perform on the high end market, when VEGA is officially a thing, and NVidia puts their GP100 at the same launch time as VEGA. Whatever yield NVidia is getting on P100, they will mostly go to the much higher profit market of servers than "cheap" gaming GPU's.

 

One thing though: Interposers are super simple silicon based on 65nm node (I believe). They are cheap as dirt and the yield is insanely high due to the simplicity and node used. I doubt interposers are a problem for anyone. HBM2 is the bottleneck for both companies.

Perhaps because we allow to post peoples personal questimate as "news"? Knowing how messages gets out, details often get lost in the transaction.

I'm thinking the selling-volume for GP100 isn't going to be anything major until they reach the OEM server-channels, instead of direct sales (which are VERY limited from what I'm hearing). Maybe they will have a few GP100 dies leftover..

 

Interposers for commercial consumer products is still a new thing. They are super simple, but often super large too (fiji interposer goes above 1000mm).

Volume is low currently, but as volume increase with HBM2 products becoming more the norm, prices will fall.

 

HBM2 sure is the big bottleneck (cost goes above GDDR5 up to multiple times per GB), hence why only enthusiast products will feature it.

[Tomsen]

15 minutes ago, Prysin said:

Sure GP104 may launch close to polaris, but i am not sure it will be as big a hit as Maxwell. I just got a feeling Nvidia will be off to a slow start this time. But time will show.

What makes people think this, I'm wondering? All we have heard from Nvidia is GP100 going to be available in Q1 2017, nothing about GP104/GP106, yet we expect them to hit within months?

I'm thinking pascal is going to be more brute-force, but that is always an alternative as long as performance holds.

Doubt performance is going to suck, they do use a much bigger die.

[Notional]

7 minutes ago, Tomsen said:

Perhaps because we allow to post peoples personal questimate as "news"? Knowing how messages gets out, details often get lost in the transaction.

I'm thinking the selling-volume for GP100 isn't going to be anything major until they reach the OEM server-channels, instead of direct sales (which are VERY limited from what I'm hearing). Maybe they will have a few GP100 dies leftover..

 

Interposers for commercial consumer products is still a new thing. They are super simple, but often super large too (fiji interposer goes above 1000mm).

Volume is low currently, but as volume increase with HBM2 products becoming more the norm, prices will fall.

 

HBM2 sure is the big bottleneck (cost goes above GDDR5 up to multiple times per GB), hence why only enthusiast products will feature it.

Yeah, it's just odd as GP100 and Vega has been official public knowledge for months now. Especially VEGA, which we know is the high end part from AMD is completely ignored, which makes no sense.

 

Sure you have scale of economics lowering prices on interposers. But they are already pretty cheap. Even though it increases the price on the card by adding an extra silicon layer, it also makes the entire product simpler.

http://semiaccurate.com/2015/06/22/amd-talks-fiji-fiji-x-odd-bits-tech/

Quote

Looking at the die and related silicon bits the first thing that stands out is the interposer. It is a 1011mm^2 part made on a 65nm UMC process. Since it is passive there are no transistors on it and the metal layer count is two or three, AMD would again not be specific. Since they are meant for carrying signals a long way across the die plus have balls attached, they are likely the wider, higher layer number rules, so again cheap to make. As a side note the assembly is likely done by Amkor but there are others like ASE who have the capabilities too now.

I wasn't really thinking of HBM2 pricing, although that of course will be high, but more the availability of them. Manufacturing has only been going for a month or two and only for the small stacks, so there simply isn't any supply to saturate an entire line of cards. And then there's the yield issue on top.

 

But I do look forward to seeing both Vega and GP100, and how they both do with HBM2. Either way, we will get some kick ass cards within the next 3-7 months.

[Sakkura]

6 hours ago, Prysin said:

lol nope. Not happening.

Nvidia has the cash reserve to survive for a decade or two with negative revenue before needing to file for chapter 11

Negative revenue? ?

[awesomeness10120]

1 minute ago, Sakkura said:

Negative revenue? ?

It's a hypothetical situation.

[Tomsen]

36 minutes ago, Notional said:

Yeah, it's just odd as GP100 and Vega has been official public knowledge for months now. Especially VEGA, which we know is the high end part from AMD is completely ignored, which makes no sense.

 

Sure you have scale of economics lowering prices on interposers. But they are already pretty cheap. Even though it increases the price on the card by adding an extra silicon layer, it also makes the entire product simpler.

http://semiaccurate.com/2015/06/22/amd-talks-fiji-fiji-x-odd-bits-tech/

 

I wasn't really thinking of HBM2 pricing, although that of course will be high, but more the availability of them. Manufacturing has only been going for a month or two and only for the small stacks, so there simply isn't any supply to saturate an entire line of cards. And then there's the yield issue on top.

 

But I do look forward to seeing both Vega and GP100, and how they both do with HBM2. Either way, we will get some kick ass cards within the next 3-7 months.

Public information doesn't matter, when we have people keep echoing the same stupid nonsense throughout different forums. Perhaps also because AMD have focused their marketing solely on polaris 10/11 at this point.

 

"Cheap" depends on the way it is used. It is an expense to an already existing product. Basically a additional fee, then small values quickly become expensive with commercial products. Sure, it is cheap compared to a logic/memory die. Also, require additional verification of both the memory and logic die before you put them on an interposer.

 

Memory manufactures don't start out fabbing memory if there isn't a need for it. If AMD/Nvidia wanted more, I don't doubt they could fab more. As they said "volume increase with customer demands".

I'm thinking the prices are the bigger reason why it is still limited to enthusiast cards. Also, there is the necessary bandwidth needed, and if GDDR5 (the cheaper memory solution), can provide that, then there are no financial reason to start with HBM2 for the entire productrange.

 

It is going to be interesting. We are still missing a whole lot of information of both GP100 and mostly vega. 

 

[Humbug]

There is a disconnect between the video content on that YouTube channel and this sensationalized forum discussion. Basically AMD will enter their next gen architecture and 14nm product stack with Polaris which will be fast but not the highest end product. Their high end product will come about 6 months later in the form of Vega and with that their single GPU lineup should be GCN 3 or newer. Nvidia on the other hand will probably start from the high end and then introduce the others. But overall neither of them are going to exit high end or maintstream markets. You can't separate them that easily. High end tech trickles down to other segments. Over the years AMD's high end products are obviously as fast as Nvidia, and Nvidia in the past has also had some great value mainstream products. Neither company is going to limit themselves. Both will try to compete across multiple price ranges.

 

The fact that high end Vega comes a few months later doesn't mean much. Nvidia launched maxwell with the 750ti, and for a good time that was the only maxwell card. Others were older kepler designs. That doesn't mean that Nvidia was giving up the high end market. It was just the first step in the transition from kepler to maxwell.

[Jahramika]

On 4/28/2016 at 4:08 AM, ShadowCaptain said:

AMD don;t want to compete wit Nvidia at the high end

 

But Im sure Nvidia want to compete with AMD at the low end

there is no way they cannot compete 

Well unless Nvidia can put a X86 Apu in a console they already lost the low ending ago. If Amd Zen  Pc APU is good Nvidia will lose more low end which pushes Nvidia into mid or high end only depending on how good Amd Apu is.