What is going to happen to AMD?

[Erenhardt]

Can HBM be used as system memory?

Why do I ask? AMD may be involved in new next gen gaming device based on Zen and HBM. Something based on these technologies would be a lot more powerful than ps4 and xbone. 14nm could make it possible to have current high-end performance in a console apu.

Ofc, Sony and MS will not be doing this to their current consoles, but there are other players that would be more than welcome to surprise us in 2016

 

[jji7skyline]

amd will survive and zen will be great

I hope so!!

 

[VirtualLarry]

Ofc, Sony and MS will not be doing this to their current consoles, but there are other players that would be more than welcome to surprise us in 2016

Apple or Samsung, with their own gaming console?

 

[richaron]

If you mean HBM at the consumer level (like a large desktop APU or a large laptop APU), then I don't think that is true whatsoever.

HBM may have a role in some type of large server APU, but I don't see it doing much for a consumer APU anywhere in near future. This not because HBM is a bad idea, but rather than large iGPUs are a bad idea for AMD. (re: Intel is too strong in mobile and on the desktop dGPU will likely be a better value than integrating large iGPUs.)

Personally, I think AMD's best chance at making APU a success exists at the server level and at the lowest common denominator for the consumer level (ie, small iGPU APU)

Then use that lowest common denominator APU (ie, small iGPU) with freedom fabric and fight Intel's superior node advantage that way (just realize it won't work for all server applications*)

*Maybe to help out the CPU of such a small iGPU APU, AMD could develop some instruction set extensions that shift load from the CPU's FPU to the iGPU? (That would make sense considering I don't expect future Intel x86 CPUs to even need such iGPU instruction set extensions. Floating point for Intel x86 will be contained within the core itself (and be quite excellent) without needing an iGPU).

I do mean HBM in the comsumer level, especially with "small" APUs. Since memory bandwidth is the main thing separating the video performance of a "small" APU with an entry level dGPU. A single chip with OK CPU and entry level dGPU performance on a competitive node is what I ment when I said "answer to many prayers" (tho perhaps not AT enthusiasts).

I'm sure stacked memory has plenty of applications...

But it would be silly of AMD to cut down the iGPU on an APU as it is their main advantage and selling point atm. That being said I hate the wasted silicon of an iGPU on a high performance platform, but maybe future technology will utilise it better.

And yes, shifting instructions to the iGPU is the area which I meant when I stated AMD is "years ahead" of their competition. They must capitalise on that, but simply having a head start with HSA will help. And providing working platforms will allow academics and the industry to move forwards.

 

[Tuna-Fish]

They have 800 million in cash on hand... In other words, they aren't going bankrupt.

Not only that, their CFO is surprisingly good. None of their debt matures until 2019, and their interest payments are quite manageable.

I doubt they will make much profit, but unless something quite drastic happens, they are in no immediate risk of bankruptcy.

 

[AtenRa]

HBM may have a role in some type of large server APU, but I don't see it doing much for a consumer APU anywhere in near future. This not because HBM is a bad idea, but rather than large iGPUs are a bad idea for AMD. (re: Intel is too strong in mobile and on the desktop dGPU will likely be a better value than integrating large iGPUs.)

R7 260X at 28nm has a die size of 160mm2(including 128bit memory). It has 896 cores with a TDP of 115W and 2GB GDDR-5 memory.

At 14nm (lets take 2x density only) it will have a die size of ~80mm2, add 4GB of HBM Gen2.0. In 2017 add four 14nm ZEN cores and we have a very nice and reasonably die size of 110-130mm2. That is the same as current Broadwell Dual core + HT GT3 at 14nm.
That APU at 15-35W TDP will have 2-3 times the performance of the current APUs. This is a perfect Laptop product, many including Apple would want to have.
Intel will not have anything close in that performance at the time.

 

[jpiniero]

flat out misinformation. see the CFO presentation from AMD FAD 2015. As of May 6,2015 there is no term debt maturities till 2019.

It wouldn't surprise me if the data I had was wrong. At the same time it wouldn't surprise me if the $600M figure is really about covering those bonds. They can kick the can down the line, but eventually they will be stopped.

You cannot design a APU with a grounds up new CPU core without first designing it and testing it as a standalone CPU product.

Intel does it.

That changed at the analyst meeting but I was led to believe by AMD that everything was going to be released in 2016. 28 nm APUs in 2016 is going to look sooo bad.

 

[frozentundra123456]

@ Atnen Ra: Seems highly optimistic that a 115 watt gpu and a say 65 watt cpu can be squeezed into a 35 watt TDP, even with the power savings of HBM and a die shrink. That is a reduction from 180 watts to 35. I do agree though that this is where AMD needs to go, rather than Zen without an igp. I think the fact that Zen is coming out without an igp basically shows AMD did not have the resources to do it in the 2016 time frame. I always said it was very ambitious to think they could bring out a new architecture on a new process with a new memory technology all at the same time, especially with AMD's limited resources. Lets see if they can really even deliver it in 2017.

 

[Enigmoid]

Zen is not a small low power core like Bobcat designed to run at 2 Ghz frequencies. Zen is a high performance high frequency core designed to run at 3.5+ Ghz speeds. AMD has never implemented SMT before in any high performance core. That adds to the technical challenge.

http://www.realworldtech.com/jaguar/

Jaguar at 28nm was 3.1 sq mm. At 14nm Jaguar would be 1.5 sq mm. Zen is expected to be a 7-9 sq mm core in size at 14nm and much closer to Broadwell in terms of transistor count and complexity. AMD need to execute Zen perfectly and avoid any Barcelona TLB like bugs. Its clearly a do or die situation. So it makes sense to launch Zen in desktop FX and servers. By the time Zen launches the CPU core would be well validated and can easily be integrated into APUs without much risk.

The bolded is complete speculation.

R7 260X at 28nm has a die size of 160mm2(including 128bit memory). It has 896 cores with a TDP of 115W and 2GB GDDR-5 memory.

At 14nm (lets take 2x density only) it will have a die size of ~80mm2, add 4GB of HBM Gen2.0. In 2017 add four 14nm ZEN cores and we have a very nice and reasonably die size of 110-130mm2. That is the same as current Broadwell Dual core + HT GT3 at 14nm.
That APU at 15-35W TDP will have 2-3 times the performance of the current APUs. This is a perfect Laptop product, many including Apple would want to have.
Intel will not have anything close in that performance at the time.

Considering how badly 28nm 15W APUs perform compared to the 35-95W bretheren 15W 896 core APU on 14nm is never going to happen with any meaningful clocks. With a 2x efficiency gain vs. 28nm you might see the performance of desktop kaveri at the 15W level.

A 115 W Bonaire shrunk to even 1/3 TDP + 4 Zen cores is not going to fit in at 15W. This is fantasy land calculations.

 

[Silverforce11]

Imagine a beefy Zen (if 40% IPC gains are real, thats very fast CPU cores) APU with HBM for insane bandwidth & low latency.

I would absolutely love a 150-200W APU that had beastly GCN + HBM graphics. Build a tiny gaming/htpc rig out of it immediately. May even retire the giant PC box if it can handle games on high (not ultra).

Intel actually has no answer for that at all.

All this time there was no chance of making a stronger graphics be the focus in APUs due to limited system ram bandwidth crippling performance. HBM is going to be the great equalizer and allow more powerful APUs to shine.

 

[Dribble]

Lots of fantasy and dreams. In the real world they'll either try to keep being the old AMD and run out of money, or restructure down to a shadow of their former selves that makes a steady turn over doing niche products.

 

[ShintaiDK]

Can HBM be used as system memory?

Why do I ask? AMD may be involved in new next gen gaming device based on Zen and HBM. Something based on these technologies would be a lot more powerful than ps4 and xbone. 14nm could make it possible to have current high-end performance in a console apu.

Ofc, Sony and MS will not be doing this to their current consoles, but there are other players that would be more than welcome to surprise us in 2016

Absolutely. Stacked memory like HBM, HMC and Wide I/O will replace conventional memory as time goes. The biggest issue in the long run is of course with servers. But in the future you will buy a CPU with fixed amount of stacked memory without any option to expand.

However cost, density and implementaion is the issue for now.

Considering there wont be a new console from MS or Sony the next long time. Who do you propose would use this. And what config had you imagined? Even AMD wont make a Zen based APU before 2017.

 

[jji7skyline]

Absolutely. Stacked memory like HBM, HMC and Wide I/O will replace conventional memory as time goes. The biggest issue in the long run is of course with servers. But in the future you will buy a CPU with fixed amount of stacked memory without any option to expand.

Sounds perfect for SoC applications like laptops and tablets.

 

[AtenRa]

@ Atnen Ra: Seems highly optimistic that a 115 watt gpu and a say 65 watt cpu can be squeezed into a 35 watt TDP, even with the power savings of HBM and a die shrink. That is a reduction from 180 watts to 35. I do agree though that this is where AMD needs to go, rather than Zen without an igp. I think the fact that Zen is coming out without an igp basically shows AMD did not have the resources to do it in the 2016 time frame. I always said it was very ambitious to think they could bring out a new architecture on a new process with a new memory technology all at the same time, especially with AMD's limited resources. Lets see if they can really even deliver it in 2017.

What I said was the 14nm APU with HBM Gen 2.0 and 896 iGPU cores, at 15-35W TDP will have 2-3 times the performance of current APUs (15-35W Kaveri).

Considering how badly 28nm 15W APUs perform compared to the 35-95W bretheren 15W 896 core APU on 14nm is never going to happen with any meaningful clocks. With a 2x efficiency gain vs. 28nm you might see the performance of desktop kaveri at the 15W level.

A 115 W Bonaire shrunk to even 1/3 TDP + 4 Zen cores is not going to fit in at 15W. This is fantasy land calculations.

see above,

 

[ShintaiDK]

Sounds perfect for SoC applications like laptops and tablets.

It will come to desktop CPUs as well. Tablets and quite abit of laptops already uses soldered memory you cant upgrade.

Remember, desktop CPUs doesnt exist. Only mobile and server.

 

[Enigmoid]

What I said was the 14nm APU with HBM Gen 2.0 and 896 iGPU cores, at 15-35W TDP will have 2-3 times the performance of current APUs (15-35W Kaveri).

see above,

At 15W it would be absolutely insane to use 896 shaders. 15W now is badly held back by tdp, especially when the CPU is heavily loaded.

2-3x current 15W performance is on the order of 384-512 shaders at low clocks. AMD is better served by bringing two dies to the market, dual core Zen (with HT) + 384 shaders and a desktop 4 core zen + 896 shaders.

 

[frozentundra123456]

What I said was the 14nm APU with HBM Gen 2.0 and 896 iGPU cores, at 15-35W TDP will have 2-3 times the performance of current APUs (15-35W Kaveri).



see above,

I know what you said. And I said there is no way to get an 80% power reduction, which is what you are proposing would require (approx 150 watts to 30). Restating what you said will not make it realistic. And mobile is actually going away from even 30 watt TDP toward 15 watt and lower. At the *very best* one might be able to get HD7770 level or lower performance in a 35 watt TDP. Even then with 640 shaders and no memory bottleneck I think you would have to downclock it (edit: a lot) to get within 30 watts for cpu plus gpu.

 

[Qwertilot]

35w looks much more realistic if you start from a 60w 750ti and half that

Still more like 45w - 15w is plenty for a sensibly matching set of Broadwell cores of course. Quite a decent sort of product for the gaming laptop market (which does exist.).

Whether AMD can get their architectures anywhere near as efficient as the above might be another matter.

 

[Ajay]

It will come to desktop CPUs as well. Tablets and quite abit of laptops already uses soldered memory you cant upgrade.

Remember, desktop CPUs doesnt exist. Only mobile and server.

That's not 100% true. Intel still make allot of money off of desktop. It may be shrinking, but it is still a big market (especially since ASPs are much better than mobile). If Intel design targets where only mobile and server, they wouldn't need a process node that can scale up to 3.5 GHz +. So both the architecture and process are still designed to be scalable enough to meet desktop user expectations.

 

[ShintaiDK]

That's not 100% true. Intel still make allot of money off of desktop. It may be shrinking, but it is still a big market (especially since ASPs are much better than mobile). If Intel design targets where only mobile and server, they wouldn't need a process node that can scale up to 3.5 GHz +. So both the architecture and process are still designed to be scalable enough to meet desktop user expectations.

There are no CPUs designed for desktop. They come from mobile and server and follow the design rules from there.

 

[Ajay]

There are no CPUs designed for desktop. They come from mobile and server and follow the design rules from there.

That's what I used to think, but it can't be true. Why would the process node and PDK be designed to handle such high clocks - neither mobile nor server need to hit 4 GHz (or possible higher). You can't produce those kind of bins 'by accident. It takes work, time and dollars - so clearly, while not it's main emphasis anymore, Desktop CPUs matter to Intel.

 

[IlllI]

I actually hope they do file for bankruptsy. because hopefully that would allow them to get out from under that awful gloflo contract they signed.

 

[Shehriazad]

Roughly said...there are 2 options here...but neither one can be locked in as the answer for sure until Q4 2016/Q1 2017.


Option 1: Zen and RX 400 are a huge success. (Forget about 300, If it's any good I expect the market shares to maybe move by 1-2% ). Zen managed to get back anywhere from 3-10% marketshare in the Desktop area. Server market wants Zen...and the GPU market share also gained back anything between 3-10% marketshare.

AMD will finally have a chance to recover as they have returned to good profitability.




Option 2: Zen and RX 400 both fail miserably and their marketshare for anything is now nonexistent. AMD will have to choose to either sell off everything they can and try to somehow continue but be totally irrelevant for years to come...or they straight out go fully bankrupt.



Option 2.5: Only Zen or Only RX 400 are a huge success or the combined success is only "so-so". In which case it is impossible to say right now if that will be enough...but if Zen succeeds on all fronts and gains back marketshare in the mid single digit % values...there is hope.

 

[jpiniero]

That's what I used to think, but it can't be true. Why would the process node and PDK be designed to handle such high clocks - neither mobile nor server need to hit 4 GHz (or possible higher). You can't produce those kind of bins 'by accident. It takes work, time and dollars - so clearly, while not it's main emphasis anymore, Desktop CPUs matter to Intel.

Intel has 47 W Haswell mobile parts that can turbo to 4 Ghz. It's probably not for very long and only for one core, but it's there.

 

[bullzz]

Imagine a beefy Zen (if 40% IPC gains are real, thats very fast CPU cores) APU with HBM for insane bandwidth & low latency.

I would absolutely love a 150-200W APU that had beastly GCN + HBM graphics. Build a tiny gaming/htpc rig out of it immediately. May even retire the giant PC box if it can handle games on high (not ultra).

Intel actually has no answer for that at all.

All this time there was no chance of making a stronger graphics be the focus in APUs due to limited system ram bandwidth crippling performance. HBM is going to be the great equalizer and allow more powerful APUs to shine.

so you are comparing an APU that might be reality in 2017/2018 and comparing that to a current gen intel cpu?

even with 40% IPC bump i am not expecting AMD to beat skylake performance let alone whats available in 2017/18. we are already seeing EDRAM coming to lga and intel is only going to make things better