The AMD Execution Thread [2007 - 2017]

[Davros]

Yay for frivolous lawsuits wasting taxpayer dollars.

Does not the loser have to pay the costs ?

 

[Deleted member 13524]

So all those FPU-less ARMv6 SoC models we've had in feature-phones and early smartphones for years should be called what?
Zero-core processors?

 

[Alexko]

And as Scott Wasson pointed out, what about CUDA cores?

 

[Silent_Buddha]

Does not the loser have to pay the costs ?

Some of the costs. But it doesn't come close to covering all the costs associated with taking a case to trial. AFAIK, no country in the world has a plaintiff or defendant bear all of the costs associated with a trial if they lose. Usually, it just covers the costs of the lawyers, which while significant, isn't usually the majority of the costs involved with a trial.

Regards,
SB

 

[Alexko]

And then there's just the court's time, which you can't give back.

 

[Deleted member 13524]

Raja Koduri in an interview with Forbes:

Raja also talked about how Advanced Micro Devices’ RTG will need to execute on their architectural designs and create brand new GPUs, something that Advanced Micro Devices has struggled with lately. He promised two brand new GPUs in 2016, which are hopefully going to both be 14nm/16nm FinFET from GlobalFoundries or TSMC and will help make Advanced Micro Devices more power and die size competitive.

Two, as in 2 different GPUs for the whole 2016? After releasing one (1) for the whole 2015 (Fiji), one (1) for the whole 2014 (Tonga) and one (1) for the whole 2013 (Bonaire)?
So AMD will be completely uncompetitive within at least 2 price ranges?


This is so bad...
Maybe the interviewer got it wrong and they're going to launch 4 new chips on 2 different quarters? Flagship + mid-low-end/mobile solutions in Q1, and then mid-end + high-end in Q4?

If not, then I just can't see what their R&D teams have been up to for the past 3 years. Apart from HBM and color compression, the latest 3 GPUs received small evolutionary updates, while nVidia got a whole new range (Maxwell has had 5 distinct GPUs for almost a year?).
I thought they were getting ready to make a big comeback in 2016 with the availability of 14/16FF, but.. 2 GPUs?

 

[Triskaine]

If not, then I just can't see what their R&D teams have been up to for the past 3 years

In the last couple years large parts of their R&D have ceased to exist due to lack of funding or leaving for greener pastures. As an anonymous source (half)jokingly put it, nowadays AMD is composed of loyal greybeards or overpromoted interns. Make of that what you will.

 

[Deleted member 13524]

Then why keep spending money on half-assed attempts?

Either they can financed enough to become competitive or they should just cease to compete in some markets. This isn't getting them anywhere.

 

[Alexko]

With APUs on top of it, they can probably get away with only 3 distinct GPUs, but I don't see how 2 could be enough, unless they rename half their product stack again; but those renamed SKUs would get butchered by Maxwell ones, let alone their Pascal replacements.

 

[Triskaine]

Either they can financed enough to become competitive or they should just cease to compete in some markets. This isn't getting them anywhere.

You are right, they should the shut down the company and give the money back to the shareholders. The BoD however, seems intent on riding this ship all the way to the bottom.

Then why keep spending money on half-assed attempts?

If you haven't worked in a large multinational company its hard to see the numerous constraints their internals have to operate under. Believe me the R&D group is quite aware of whats going on but that doesn't mean they are going to lie down and wait for death. As another anonymous AMD Engineer on the Bulldozer team said back in 2009(!): "Bulldozer isn't all that great, but sometimes you gotta ship what you have." The same applies to their other efforts.

 

[Alexko]

You are right, they should the shut down the company and give the money back to the shareholders. The BoD however, seems intent on riding this ship all the way to the bottom.

AMD has more liabilities than assets; there's nothing to give back to shareholders, the only way out is to turn the company around.

 

[3dilettante]

You are right, they should the shut down the company and give the money back to the shareholders. The BoD however, seems intent on riding this ship all the way to the bottom.

My cursory searching indicates creditors would come first, and AMD has a lot of debt to get through before anything gets to shareholders.

If you wanted to get something back if a wind-down is inevitable, maybe you start parceling out workable divisions for sale like graphics and ride the tail of a hopefully higher-margin but legacy space in hopes of retiring enough debt that something is salvageable. There's a tail of existing x86, and the Zen architecture is built in the last gasp of a cross-licensing agreement whose capture period has expired. Zen's instruction support documented thus far appears to capture the meat of Haswell's ISA support, minus anything forward-looking. The IP base Zen's successors could safely draw from would likely start to age out around the time that the first round of debt comes due.
That leaves a possible division of a design house that could be sold for non-x86 stuff.

 

[Deleted member 13524]

With APUs on top of it, they can probably get away with only 3 distinct GPUs, but I don't see how 2 could be enough, unless they rename half their product stack again; but those renamed SKUs would get butchered by Maxwell ones, let alone their Pascal replacements.

Exactly. Unless they manage to put HBM2 in Fiji and resell the Nano + 8GB as their next high-end part.

You are right, they should the shut down the company and give the money back to the shareholders.

Not what I said. My suggestion would be e.g. to give up on APUs altogether until they can get a Zen+GCN+HBM APU ready for the consumer and stop their pitiful Pitcairn rebranding attempts on the mobile GPU market.
What has Carrizo and Kaveri brought to AMD? A couple of design wins on bottom-of-the-barrel 15" laptops that are coupled with 1366*768 screens and mechanical hard drives?
After its utter failure in the laptop market, why bring Carrizo to the desktop? Who's going to buy that?

 

[Triskaine]

My suggestion would be e.g. to give up on APUs altogether until they can get a Zen+GCN+HBM APU ready for the consumer and stop their pitiful Pitcairn rebranding attempts on the mobile GPU market.
What has Carrizo and Kaveri brought to AMD? A couple of design wins on bottom-of-the-barrel 15" laptops that are coupled with 1366*768 screens and mechanical hard drives?
After its utter failure in the laptop market, why bring Carrizo to the desktop? Who's going to buy that?

As I already said, sometimes you have to ship what you got. Doing nothing is not an alternative. In fact, there is nothing worse than doing nothing.

 

[3dilettante]

The next year's quad-core APU looks to be a warmed-over Carrizo, whose downside has primarily been handled by the initial Carrizo. Then there's the WSA and the need to buy wafers or pay for the privilege of not buying them.

 

[Dave Baumann]

Two, as in 2 different GPUs for the whole 2016? After releasing one (1) for the whole 2015 (Fiji), one (1) for the whole 2014 (Tonga) and one (1) for the whole 2013 (Bonaire)?

I think you are forgetting Hawaii and Iceland in there as well. Two releases in a year is pretty much the norm - i.e. have NV released more than GM200 and GM206 in 2015?

 

[Deleted member 13524]

I think you are forgetting Hawaii and Iceland in there as well. Two releases in a year is pretty much the norm - i.e. have NV released more than GM200 and GM206 in 2015?

Granted, 3 GPUs in 2013. I honestly thought Oland had been released in 2012 with the rest of the initial batch of GCN GPUs, and that Iceland was just a rebadge of Oland (seeing how there's no color compression or audio DSP).

 

[Dave Baumann]

Oh, yeah, I forgot about Oland as well, so that's another. Iceland is for notebook and is basically just a 3D accelerator with no ancillaries (i.e. Display, video encode/decode or Audio DSP); it is GFXIP8 though so contains compression

 

[DuckThor Evil]

Well 2 GPUs can cover a pretty wide range, say around 400mm2 and 200-250mm2 or something like that.

 

[Deleted member 13524]

Well 2 GPUs can cover a pretty wide range, say around 400mm2 and 200-250mm2 or something like that.

At 16nm and 400mm^2, they could probably fit ~1.3x the transistors of Fiji, being a more powerful GPU in the end. I think Fiji is the range AMD will try not to update during 2016, since Pascal seems to be coming in late 2016 and the Fury X is steadily gaining ground from newer drivers against the 980 Ti. Maybe they will be able to reuse the GPU with higher clocks and HBM2 for 8GB VRAM in a new model. Furthermore, the Fury X2 hasn't even launched yet.

IMHO, a more clever approach would be to launch:

1 - ~150mm^2 GPU for mid/high-end mobile (where GM107 rules unrivaled) with a performance between Pitcairn and Tonga, plus a single stack of HBM2 (~200GB/s?) to reduce power consumption even further.

2 - ~250mm^2 GPU to replace Hawaii at similar performance but at a fraction of the cost and power consumption. Two stacks of HBM2 to make tiny cards with up to 8GB VRAM and ~400GB/s. It would be a GM204 killer, at least until Pascal comes out. This would probably render the Fury Nano useless, though.