So what did Rory Read do wrong?

[ShintaiDK]

Mantle development was probably never ahead of DX12, it's just that they released it early. It's much easier to release something that's only required to work for a few specific chips (a few GCN chips) and only in a few specific ways (a few frostbite games) and then fix the bugs for those use cases. MS had to make DX12 fully functional across many different architectures and for many different purposes. That takes a lot more effort to get it right.

Exactly.

With the release of both Skylake and Maxwellv2 as FL12.1. Its clear AMD is the one behind.

 

[mrmt]

Given that AMD were the GPU vendor, they will have worked closely with Microsoft on developing an API (and more importantly, toolchain) to exploit their hardware. DICE worked on D3D 11.X games, AMD at the very least worked closely with Microsoft on implementing D3D 11.X, so Microsoft's latest thinking (11.X) most likely helped inform Mantle's design, which then in turn helped inform the design of DX12. It's not like it's a one way street.

Directx 12 is embedded into W10, the driver architecture is built around it. That doesn't happen in 18 months. But building a dx12 lite for a few games does.

And if AMD was so instrumental for dx12, how come that it support less features than the others two vendors?

 

[NTMBK]

Mantle development was probably never ahead of DX12, it's just that they released it early. It's much easier to release something that's only required to work for a few specific chips (a few GCN chips) and only in a few specific ways (a few frostbite games) and then fix the bugs for those use cases. MS had to make DX12 fully functional across many different architectures and for many different purposes. That takes a lot more effort to get it right.

If DX12 were at the same stage as Mantle development, they would have put it on the XBox One over a year ago.

 

[NTMBK]

And if AMD was so instrumental for dx12, how come that it support less features than the others two vendors?

Because DX12 obviously adds features on top of Mantle, and AMD haven't significantly upgraded their GPU arch in years.

 

[ShintaiDK]

Because DX12 obviously adds features on top of Mantle, and AMD haven't significantly upgraded their GPU arch in years.

Do you really believe this? Intel and nVidia is clear examples that Mantle was a kneejerk reaction to what was coming. Something AMD couldnt foot the R&D bill for.

 

[NTMBK]

Do you really believe this? Intel and nVidia is clear examples that Mantle was a kneejerk reaction to what was coming. Something AMD couldnt foot the R&D bill for.

The only "kneejerk reaction" I am seeing around here is the instinct to trash something purely because it is associated with AMD.

The "12.1" features add specific hardware based features, support for which could have been added to DirectX 11. (And were- feature level 11.3, which you know about.) Mantle was more focused on a total redesign of the software layer, something completely orthogonal.

EDIT: Actually stuff it, I'm done arguing in circles with you about this. I'm not going to continue to derail this thread.

 

[mrmt]

Because DX12 obviously adds features on top of Mantle, and AMD haven't significantly upgraded their GPU arch in years.

So you believe that because of Mantle Microsoft threw away whatever DirectX development they had and built the current DX12 in just 18 months, all that because of AMD influence?

 

[NTMBK]

So you believe that because of Mantle Microsoft threw away whatever DirectX development they had and built the current DX12 in just 18 months, all that because of AMD influence?

Threw away? No. Refocused and made significant changes? Yes. And no doubt Microsoft knew what AMD were up to long before Mantle went public.

 

[AtenRa]

So you believe that because of Mantle Microsoft threw away whatever DirectX development they had and built the current DX12 in just 18 months, all that because of AMD influence?

No, actually Intel asked MS to do something because their iGPUs sucks and AMD took advantage of it. [/sarcasm]

It is very easy to understand what happened, AMD won the consoles and MS had to make DX-12 for the AMD GCN hardware inside the XBone. Simple as that.
On the other hand, AMD was able to make Mantle sooner because they already new about the GCN architecture even before 2011.

Now just imagine what will happen when first Samsung/Qualcomm etc HSA APUs start to be produced and released in the retail.

 

[mrmt]

Threw away? No. Refocused and made significant changes? Yes. And no doubt Microsoft knew what AMD were up to long before Mantle went public.

AMD ended up covering less DX12 features than the other two IHVs, no DX12 feature is AMD exclusive and on top of that no feature is Mantle exclusive. That wouldn't happen if DX12 would have been built around AMD GCN and Mantle.

 

[DrMrLordX]

This is the kind of strategic mistake I was talking about. Instead of pursuing desktop sales that could play into Bulldozer strengths, he tried to chase the mobile market with an inadequate product... and ended up losing the two markets.

For what it is, Carrizo is an excellent processor. It's just not an excellent processor at clockspeeds above 2.5 GHz or so. So, it's a fine mobile CPU . . . until notebook OEMs screw it up.

Mantle is and will be a big win for AMD's GPU division. Or, at least, it is/will be a big win if they can find a way to parlay that success into sales. They've set the tone for years of graphics development in games, and they've lined up their own hardware to support early DX12 titles better than Nvidia's. Can they cash in on that? We'll find out.

Now just imagine what will happen when first Samsung/Qualcomm etc HSA APUs start to be produced and released in the retail.

Uh, got links? Thus far, all the HSA "partners" appear to have done jack squat to support development of the HSA ecosystem. Which really stunk for AMD. They had all these big names signing on to support the initiative, only for them to (apparently) sit idly by and do nothing.

 

[mrmt]

For what it is, Carrizo is an excellent processor. It's just not an excellent processor at clockspeeds above 2.5 GHz or so. So, it's a fine mobile CPU . . . until notebook OEMs screw it up.

Is just OEM fault? Carrizo might be an "ok" performer but it is still a huge die, which certainly impacts pricing to OEMs. Carrizo is a failure, but not only for performance reasons.

 

[Roland00Address]

Rory Reed was hired for two reasons.

1) He knew finances so they assumed he could cut costs in an effective but not counterproductive manner. The right cost cutting will allow AMD to grow via reducing debts and costs but not destroy your potential to create products vs no cost cutting due to you cut too deep.

2) He was a very successful career prior to AMD especially with the growth of Lenovo. Rory Read's career is like this

+++IBM worked his way up and became well known. Ended up as a Vice President who was in charge of IBM Services and E-Business strategy...aka helping making IBM the company they are now today instead of the commoditized computer hardware that HP and Dell are in.

+++Lenovo Chief Operating Officer and President (remember Lenovo acquired much of IBM). During this time Lenovo went from 4th place in PC market share from 7.0% to 12.5% which was 2nd place, a year and a half later after he left Lenovo was #1 in PC market share at 16.9%, and currently they are still #1 at 18.8%. Growing a business and acquiring market share in a highly competitive marketplace while still keeping margins (for Lenovo was profitable and had 7 quarters of great growth) is a very big deal. Now Rory was not the only one who made this happened but as COO and President he was part of this big tech story.

+++AMD where he was CEO for a little more than 3 years (3 years 3 months) but he also stayed on the board and helped transistion the new CEO for 4 months after that, so lets say 3.5 years. During this time he did great jobs with cutting costs and diversifying AMD's business. Her I am quoting wikipedia

At AMD, Read inherited a company that had approximately 95% of its revenue driven by the PC market.[6][7] Read diversified the portfolio to produce revenue of 50% from five new high growth markets, building over US$2 billion in new businesses.[8][9] Under Read, AMD lowered costs by over 30% while restructuring AMD debt strengthening the balance sheet and returning to non-GAAP profitability.[2][10] He was responsible for implementing an ambidextrous X86/ARM architecture and the clean sweep removal of new game consoles during his tenure.[11][12]
https://en.wikipedia.org/wiki/Rory_Read

Since leaving AMD he is now doing the exact same job title he used to have at Lenovo but now with Dell.

------

Now the amazing things Rory Reed did was clean up AMD's balance sheet and actually starting to get revenue from embedded and custom chips. Now these custom chips do include the consoles but hopefully they will be more in the future.

The problem now with AMD is they need an architecture that people actually want, which requires having enough talent to design it, as well as making the right design decisions. Furthermore just because your architecture is great you need performance effective fabs that are cutting edge enough and competitive and you do not control these and a cost effective pricing on fabs they do not control.

Plus no delays for delays are much worse for a person who is playing catch up vs a person who has all the market share and customers that have no place else to go and would need to create relationships to switch in case of a sudden shortage.

So effectively all things that Rory may be good for but probably does not have the skills for and they hope this new CEO who specializes in the nuts and bolts can pull out a miracle.

Another downside of Rory is that they actually did not gain market share in the PC space, the growth was in places not the pc space. Now some of this is Rory's fault, but some of this is just AMD keep on fumbling the ball or its factor's AMD can not control like no 20nm.

 

[mrmt]

Now the amazing things Rory Reed did was clean up AMD's balance sheet and actually starting to get revenue from embedded and custom chips. Now these custom chips do include the consoles but hopefully they will be more in the future.

Rory didn't start the console contracts and Rory failed to get new contracts beyond the two consoles AMD already had, and on top of that the console revenue yields less money for R&D than comparable CPU/GPU revenue of the time Rory assumed AMD. This kind of diversification, getting lower margins in a highly concentrated business isn't good for AMD in the long run.

 

[DrMrLordX]

Is just OEM fault? Carrizo might be an "ok" performer but it is still a huge die, which certainly impacts pricing to OEMs. Carrizo is a failure, but not only for performance reasons.

Yes, it is just the OEM's fault, when they choose to limit the CPU to a 15W TDP instead of allowing user-selectable TDP or adhering to AMD's recommended TDP for the part (25W for 8700P, 35W for 8800P). Freed from those restrictions, it is a great step forward over mobile Kaveri in many respects.

 

[myocardia]

Yes, it is just the OEM's fault, when they choose to limit the CPU to a 15W TDP instead of allowing user-selectable TDP or adhering to AMD's recommended TDP for the part (25W for 8700P, 35W for 8800P). Freed from those restrictions, it is a great step forward over mobile Kaveri in many respects.

You forgot to add the caveat of "for the 1/10th of one percent who post regularly at the Anandtech forums", and have a 5,000 foot long extension cord, making battery life mean nothing.

 

[NTMBK]

Is just OEM fault? Carrizo might be an "ok" performer but it is still a huge die, which certainly impacts pricing to OEMs. Carrizo is a failure, but not only for performance reasons.

It is huge, but they finally integrated the southbridge on-chip, much like Intel integrated it on-package a couple of years back. This drives down motherboard complexity and area, which should help at least a little with the total system costs.

 

[ShintaiDK]

Mantle is and will be a big win for AMD's GPU division. Or, at least, it is/will be a big win if they can find a way to parlay that success into sales. They've set the tone for years of graphics development in games, and they've lined up their own hardware to support early DX12 titles better than Nvidia's. Can they cash in on that? We'll find out.

You must have missed the Fable DX12 bench.

 

[ShintaiDK]

Yes, it is just the OEM's fault, when they choose to limit the CPU to a 15W TDP instead of allowing user-selectable TDP or adhering to AMD's recommended TDP for the part (25W for 8700P, 35W for 8800P). Freed from those restrictions, it is a great step forward over mobile Kaveri in many respects.

Why is it always someone else fault with AMD?

 

[mrmt]

Yes, it is just the OEM's fault, when they choose to limit the CPU to a 15W TDP instead of allowing user-selectable TDP or adhering to AMD's recommended TDP for the part (25W for 8700P, 35W for 8800P). Freed from those restrictions, it is a great step forward over mobile Kaveri in many respects.

You mean it's OEM's fault for not using even more expensive components for Carrozo laptops, which are outrageously expensive for the CPU performance one gets, right?

 

[mrmt]

It is huge, but they finally integrated the southbridge on-chip, much like Intel integrated it on-package a couple of years back. This drives down motherboard complexity and area, which should help at least a little with the total system costs.

There's no but. It's huge, period. End of history. Done deal.

 

[NTMBK]

There's no but. It's huge, period. End of history. Done deal.

That sort of "I'm right, you're wrong, and that's that" mentality is exactly why you are so infuriating. Can we not try to actually have a discussion? :\

 

[mrmt]

That sort of "I'm right, you're wrong, and that's that" mentality is exactly why you are so infuriating. Can we not try to actually have a discussion? :\

Sorry for not properly detailing my pov in the previous post.

Carrizo certainly is an improvement over Kaveri in terms of cost structure, but look around at the competition. It was no surprise that Intel was going to be very aggressive in terms of die scaling with 14nm. Even with 22nm 200mm^2 and above dies were either server/workstation products or niche. That they thought that they could succeed with such a huge die (even with pch included) after watching Trinity and Kaveri baffles me to this today.

A product has many dimensions (cost, power consumption, performance, platform features, resale price, etc) to many players in the chain. AMD seems to think that if they focus on one the others will be solved by themselves. Carrizo is an improvement in a few dimensions but it doesn't do much in others, so all in all we end up with a bad product, regardless on how some niche users feel about it.

 

[Phynaz]

Mantle development was probably never ahead of DX12, it's just that they released it early. It's much easier to release something that's only required to work for a few specific chips (a few GCN chips) and only in a few specific ways (a few frostbite games) and then fix the bugs for those use cases. MS had to make DX12 fully functional across many different architectures and for many different purposes. That takes a lot more effort to get it right.

Mantle never made it to release, it never came out of beta.

 

[DrMrLordX]

You must have missed the Fable DX12 bench.

I didn't miss the Ashes of Singularity benchmark(s):

http://www.pcper.com/reviews/Graphi...ted-Ashes-Singularity-Benchmark/Results-Avera

http://www.pcper.com/reviews/Graphi...ted-Ashes-Singularity-Benchmark/Results-Heavy

Considering that the 390x is just a suped-up Hawaii, that's awesome.

As for Fable:

http://www.pcper.com/reviews/Graphi...formance-Testing-Continues/Results-1080p-Ultr
http://www.pcper.com/reviews/Graphi...erformance-Testing-Continues/Results-4K-Ultra

Again, we see the 390X beating the 980, which is . . . awesome.

Why is it always someone else fault with AMD?

It isn't.

You mean it's OEM's fault for not using even more expensive components for Carrozo laptops, which are outrageously expensive for the CPU performance one gets, right?

More expensive? What are you going on about? Notebook OEMs have been providing adequate cooling solutions for chips in that TDP range for years at less-than-astounding prices. And since when was Carrizo outrageously expensive?