AMD Q414 results

[BSim500]

Note the qualifier you used, high end games. High end gaming is a niche of the gaming world. Just look at steams hardware survey, it shows the kind of hardware most gamers use...

Steam's HW Survery GPU Makeup:-
44.0% nVidia
22.0% AMD
19.8% Intel
13.8% Other
http://store.steampowered.com/hwsurvey/videocard/

Out of AMD's 22% GFX market share, Kaveri's are included under the 0.55% user base "AMD Radeon R7 200 Series" shared with desktop dGPU's like R7 240/250/260/260X, etc. Remove those dGPU's, and Kaveri users basically make up nearer 0.20-0.25% (1 in 400-500) of the gaming market. Not exactly sure how that results in dGPU's being made "fringe" for gaming or AMD APU's being "mainstream" (either for gaming OR compute)?

There seems to be a constant "split" between the assumption of "20% of Steam users use Intel iGPU's therefore AMD's APU must be selling well" or "everyone's interested primarily in APU compute performance" vs what 99.5% of the market is actually doing to the contrary in observable reality...

 

[sham63]

During my time on active duty in the Navy, most guys had laptops to game with, forgoing any sort of gaming "system." About half of the laptops at any given Navy exchange are AMD based for their APU capabilities, giving sailors and marines the ability to have a good gaming experience and portable computing. Even though many of us we're guaranteed to be on that base for 1.5-2 years, I not once ever saw a desktop computer inspecting barrack rooms. Almost everyone was playing games like Skyrim, Starcraft 2, WoW among many others, happy with the integrated graphics. I hate to say it but mobile is a big deal already and will only get much bigger.

We're talking at least 400 laptops to 0 desktops in these two barracks. This was three years ago. Now that Intel has better IGPs than before and the advancements of AMD's line, I have no doubts about what the market is shifting to.

An IGP/APU is a very adequate solution. The goalposts that wannabe "hardcore" gamers throw out are always moving and current products are never good enough.

Could some of that have been do to space? I know when I was in the USAF in the mid 80's the rooms were kind of small and you had a roommate. I could not see a room that size today having the room for a desktop or two. A laptop you can use on your bed, a desktop not so much.

 

[cbn]

It seems to me that people doesn’t understand why APUs were made and for what market segments. APUs were made for Compute, it is the only thing that Intel cares and the reason they continue to invest die space for the iGPU with each generation. If AMD APUs were not existed, nobody would talk about gaming with APUs that much. Gaming is an added feature that comes with the APUs due to the common design they share with the dGPUs but not the main focus.

It will take some time more but eventually APU (iGPU+CPU) compute performance will be the main metric for Desktop and Mobile in the near future. From Skylake onwards Intels CPUs will devote more than 50% of their die size for the iGPUs. Performance metrics will start to change towards the iGPU compute than the CPU.

Software will have to ride the APU (iGPU+CPU Compute) wagon train or else applications will start to lag badly behind in performance. Imagine a Video editing application relying only to CPU performance in an age of Skylake/Carrizo where iGPU performance will be much much faster than 4 CPU threads in the consumer world. Imagine that a 15W TDP Mobile APU running an application that can leverage the APU strength to be faster than a 4 Core 95W TDP Desktop CPU running an application only with its CPU cores. Second application will have a tremendous disadvantage in the market.

The battle in x86 consumer segment is changing, from now on it will be on iGPU compute first and not CPU. APUs will devote larger and larger die size for the iGPU with every new generation. But that doesn’t mean they will replace dGPU in the high-end. dGPUs will remain dominant in their segment for a long time.

I think even for compute the buyer of such hardware is going to be pretty niche.

So the question then becomes what is the most likely use of larger iGPUs: compute or gaming?

Unfortunately I am thinking the answer will be gaming, and probably always be that way. (One reason is that there are more younger people than older people, and those younger people are more likely to use their GPUs for gaming than any compute related tasks--things like heavy video editing, etc.).

 

[mrmt]

I think even for compute the buyer of such hardware is going to be pretty niche.

So the question then becomes what is the most likely use of larger iGPUs: compute or gaming?

Is AMD spending marketing money to show how good the igpu is to compute, or does AMD portray its chips as gaming powerhouses?

 

[Ramses]

I don't see anyone marketing compute to anybody even remotely consumer-ish.

 

[monstercameron]

I don't see anyone marketing compute to anybody even remotely consumer-ish.

intel and ms do it in alot of product ads, it is photoshop.

 

[Ramses]

is this one of those things where everyone calls it their own cutesy name?

 

[DrMrLordX]

GPGPU is being subtley pushed on the gamer public as a part of Mantle, or at least that's what I take away from it. DX12 is gonna be like that too.

All it's going to take is some benchmark where a review site shows Broadwell Iris Pro beating up on a Haswell-E using the same dGPU in some DX12 title to start up that fire, if AMD doesn't manage to do it with one of their APUs first. I sort of expect Intel to beat them to the punch . . . if they can ever launch the darn things.

 

[cbn]

GPGPU is being subtley pushed on the gamer public as a part of Mantle, or at least that's what I take away from it. DX12 is gonna be like that too.

All it's going to take is some benchmark where a review site shows Broadwell Iris Pro beating up on a Haswell-E using the same dGPU in some DX12 title to start up that fire, if AMD doesn't manage to do it with one of their APUs first. I sort of expect Intel to beat them to the punch . . . if they can ever launch the darn things.

What you write about is pretty interesting, but Iris Pro and Haswell-E are at the highest end of the market (where volume is really low).

In contrast, even most people in this enthusiast forum only use a Core i5 (of some type) that they may have been holding onto for quite some time because they haven't found the need any extra CPU performance.

 

[cbn]

Here is what i would like AMD to do for desktop:

1. Establish more affordable big core CPUs with small iGPU (fabbed at Global foundries)
2. Bring dGPUs to Global foundries.
3. As the process nodes decrease make the decision as to how much extra iGPU should be added to the cpu based on practical use and affordability. (However, since AMD doesn't have the performance crown for cpu they should resist adding iGPU and just focus on lowering cost and retail price instead.)

 

[mrmt]

GPGPU is being subtley pushed on the gamer public as a part of Mantle, or at least that's what I take away from it. DX12 is gonna be like that too.

As much as GPGPU will play a bigger role in the future, neither AMD nor Intel sell a CPU today to offer a benefit years down the road, they must offer something right now to entice people to buy their processors. It would be very foolish from AMD's part if they decided to burn half the die of their products for the sake of a nonexistent feature like compute, but it would be extremely foolish if on top of burning half of the die they didn't advertise this feature.

Nope, APUs today are graphics. This is what AMD tries to sell us, this is where AMD puts their money in.

Here is what i would like AMD to do for desktop:

Desktop market is a write off. None of the companies will develop products specifically for it again,

 

[cbn]

Desktop market is a write off. None of the companies will develop products specifically for it again,

The rumored Bristol Ridge (for Q3 2106) is specifically a desktop die.

 

[mrmt]

The rumored Bristol Ridge (for Q3 2106) is specifically a desktop die.

It will be rather curious if they launch an Excavator die by the time they should have their next wonderchip, designed by the hands of an EE God, rolling out on the market.

 

[cbn]

It will be rather curious if they launch an Excavator die by the time they should have their next wonderchip, designed by the hands of an EE God, rolling out on the market.

Bristol Ridge Deskop APU is supposed to be a quad core Excavator with 512 sp iGPU, but instead of it having an integrated southbridge ( like Carrizo) it will use the FM3 desktop southbridge. It also has a DDR4 memory controller (unlike Carrizo with DDR3).

 

[cbn]

Here is my opinion of AMD Bristol Ridge:

http://forums.anandtech.com/showpost.php?p=37135424&postcount=236

So in a nutshell, here are some reasons I would like to see AMD Bristol Ridge desktop reconsidered:

1. Bristol Ridge is only a Excavator quad core. (This will yield both quad cores and dual cores from the die. The quad should be adequate for a lot of gaming, but the dual will not. With that mentioned, an AMD quad core is still not optimal for a lot of newer games like Battlefield 4 64 player where I would like to see at least six AMD cores with the current Steamroller levels of IPC and single thread.)

2. Bristrol Ridge 512sp iGPU is cost adder that pushes die size over the sweet spot for cost. There is also the issue of how high this iGPU will be able to be clocked while remaining under 95 watt TDP for socket FM3. (IMO, in order to maximize the expensive investment of the 512sp iGPU, clocks should be at least 1000 Mhz or greater like the R7 250 (1000/1050 Mhz) , R7 250X (1000 Mhz) , R7 260 (1000 Mhz), R7 260X (1100 Mhz).

3. The dual channel memory cost issues I mentioned in the previous posts will likely add even further to a system price disadvantage compared to a value gamer using dGPU.

Overall, I would say Bristol Ridge is probably a step in the right direction but still too weak and low value compared to what AMD should be giving us in late 2016. IMHO, I would like to see something with at least the same single thread as Kaveri, but with two more cores added and a much smaller iGPU.

 

[mrmt]

Bristol Ridge Deskop APU is supposed to be a quad core Excavator with 512 sp iGPU, but instead of it having an integrated southbridge ( like Carrizo) it will use the FM3 desktop southbridge. It also has a DDR4 memory controller (unlike Carrizo with DDR3).

Either Zen was canned or flopped or its performance targets are waaaaay lower than the construction cores, and the construction cores performance are nothing to write home about. The other issue I see is with timing. If they are going to have Excavator ready in 2015, why bother with it in 2016/2017 when it will face stiffer competition?

 

[cbn]

Either Zen was canned or flopped or its performance targets are waaaaay lower than the construction cores, and the construction cores performance are nothing to write home about. The other issue I see is with timing. If they are going to have Excavator ready in 2015, why bother with it in 2016/2017 when it will face stiffer competition?

There is supposed to be a Zen octocore with L3 cache (using the Samsung licensed 14nm) also available on FM3.

But I think the fact Excavator is still being used so late (according to rumors) in the game tells me either 1.) there are still issues with the WSA demanding 28nm parts or 2.) that Zen is slower in single thread.

Zen being made on a lower power 14nm process makes me believe it will probably be slower in single thread than construction cores.

So I am thinking the idea for FM3 might be if a person wants the highest multithread they can go with Zen, and folks wanting the best AMD single thread could get an iGPU-less Athlon variant of Bristol Ridge (or an APU version of Bristol Bridge).

 

[monstercameron]

As much as GPGPU will play a bigger role in the future, neither AMD nor Intel sell a CPU today to offer a benefit years down the road, they must offer something right now to entice people to buy their processors. It would be very foolish from AMD's part if they decided to burn half the die of their products for the sake of a nonexistent feature like compute, but it would be extremely foolish if on top of burning half of the die they didn't advertise this feature.

Nope, APUs today are graphics. This is what AMD tries to sell us, this is where AMD puts their money in.



Desktop market is a write off. None of the companies will develop products specifically for it again,

How can you say that compute is nonexistent?
It is true that you don't see it everywhere but a lot of a/b editors-Adobe suite and Sony Vegas- use compute and/or gfx acceleration. As time goes on more software will use it. Also note this isn't an amd thing as Intel is going the same route.

 

[BSim500]

How can you say that compute is nonexistent? It is true that you don't see it everywhere but a lot of a/b editors-Adobe suite and Sony Vegas- use compute and/or gfx acceleration.

Personally, I don't think it's non-existent. It's just that people who shell out hundreds of dollars on Adobe suites and buy 16-32GB RAM tend not to match them up with sub-$100 CPU's... And the average Joe whose "photo editing" needs consists of basic cropping / resizing / color correcting / red-eye removing his vacation snaps / porn collection pretty much uses GIMP or Paint.NET and has little to "accelerate" on cropping single-layer images that don't require complex multi-layered blurs.

It's another case of APU's being stuck between a rock (simple baseline crop/resize stuff doesn't even max out a laptop Pentium) and a hard place (professional users quite happily buy expensive CPU's to match their expensive 32GB RAM, expensive software, expensive dedicated "scratch disk" SSD's, etc).

 

[podspi]

2.) that Zen is slower in single thread.

Zen being made on a lower power 14nm process makes me believe it will probably be slower in single thread than construction cores.

So I am thinking the idea for FM3 might be if a person wants the highest multithread they can go with Zen, and folks wanting the best AMD single thread could get an iGPU-less Athlon variant of Bristol Ridge (or an APU version of Bristol Bridge).

I won't say there is no chance this is true because - hey - I said the same thing about Bulldozer and we know what happened there...

But this makes very little sense. The construction cores are nowhere near adequate in single-thread performance to compete with Intel, and unless AMD is planning on ceding the higher end (heck, really the mid-range too) another regression in ST performance is just not in the cards.

I mean, the Bulldozer architecture has advanced since its original release, but for the most part per clock it exchanged blows with Thuban, which truth be told was probably a bit slower than the C2D per clock. At this point, even lower-power architectures are getting up to those speeds...

 

[mrmt]

Zen being made on a lower power 14nm process makes me believe it will probably be slower in single thread than construction cores.

I don't think AMD can make Zen worse than Excavator, so the only scenario this would make sense is if Zen will be severely TDP limited, to the point a subpar design like excavator would make more sense than Zen, despite the worse performance per watt excavator should have. Samsung 14nm really points out to that, being a node tailored for SoCs, and not for high performance designs, but if that's the deal AMD on the desktop market has no future, as the nodes Samsung is going to build are also tailored for SoCs.

But also begs the question, if Excavator is ready in 2015, why withhold the desktop version to 2016?

 

[VirtualLarry]

But also begs the question, if Excavator is ready in 2015, why withhold the desktop version to 2016?

Might have something to do with AMD "stuffing the channel" with parts?

 

[mrmt]

Might have something to do with AMD "stuffing the channel" with parts?

Nope. Those aren't produced parts, AMD will be manufacturing them around Q116, they would have time to sort whatever problems in the channel until there. The main question is, why launch Excavator for desktop by the time Zen would be finished? The only reason I can think of is that Zen is somehow TDP-capped to the point it cannot be a desktop chip.

 

[AtenRa]

The only reason I can think of is that Zen is somehow TDP-capped to the point it cannot be a desktop chip.

So they making an 8Core 16Thread mobile SoC ???

What the hell people, soon you will come to the conclusion that ZEN will be a 16thread Tablet SoC

 

[ShintaiDK]

Its quite obvious that Zen and K12 will be another idiotic attempt for AMD to come back to servers. Yet again dragging the rest of the company down with it.