Was AMD holding back on their fusion line?

[AtenRa]

Yes it is but how much better is it?

That's what everyone else is saying. Who cares about $20 more if it doubles the fps.

Granted the Athlon 750k will have lower CPU performance but its unlikely to bottleneck many games with a R260 and the CPU can be mildly overclocked. Power consumption will go up but power use will be so low that realistically you are going to be using the same PSU on both builds (~300W). SFF yep, thats a problem.

That said you could get the 750K + 8 GB RAM and a R250/250X.

With the Athlon 750K you compromise on all fronts in order to have more Graphics performance. That is acceptable ONLY from a gamers perspective. But i have clearly said from the start i was not talking only about gaming. And even if we would only evaluate the Gaming performance against an Intel + dGPU the A10-7850K would be faster (keeping the same price and the rest of the characteristics).

 

[frozentundra123456]

Talk about compromise. Kaveri is the ultimate compromise and an expensive one at that. It is a compromise in *both* CPU and gpu performance, ultimately being outstanding at neither. With a different CPU you can tailor the system to your specific needs. Kaveri forces you to accept both mediocre CPU performance (relative to Intel or even FX) and mediocre gpu performance (relative to a discrete card.)

If one tries hard enough, perhaps they can construe some very limited conditions that make this a good value, but for the vast majority of cases I just don't see it.

 

[AtenRa]

Talk about compromise. Kaveri is the ultimate compromise and an expensive one at that. It is a compromise in *both* CPU and gpu performance, ultimately being outstanding at neither.

Same could be said for Core i3 then, it is a compromise in CPU performance relatively to Core i5/7 and an even more compromise in iGPU performance compered to Kaveri.

You are an OEM/User, you have a specific budget to build a PC System. You can use the Core i3 + $60-70 dGPU or the A10-7850K. Are those two competitive ??? were do you compromise with the one and where do you compromise with the second ???
Now, do you people understand what is competitive and how you evaluate the two ???

 

[raghu78]

AMD is just a generation away from making the entry level dGPU irrelevant. AMD's problem today is bandwidth which will be addressed with HBM and 2.5D stacking.

http://www.microarch.org/micro46/files/keynote1.pdf

a TSMC 20nm AMD APU with 768 sp and a HBM stack running at 800 Mhz (102.Gb/s) provide GPU performance on par with Xbox One. There are slight compromises like the single thread performance. AMD is working on a new architecture. Excavator is the last of the ill conceived Bulldozer design. A new CPU core architecture is due by late 2015 or early 2016. A 2016 APU built at TSMC 16FF+ with 4 of those new cores with 768 sp and 100 Gb/s bandwidth should satisfy any mainstream gamer.

 

[Phynaz]

You are an OEM/User, you have a specific budget to build a PC System. You can use the Core i3 + $60-70 dGPU or the A10-7850K. Are those two competitive ??? were do you compromise with the one and where do you compromise with the second ???
Now, do you people understand what is competitive and how you evaluate the two ???

How are those OEM Kaveri sales doing compared to i3?

Your argument is invalid.

 

[sm625]

The fact that 512 GCN cores is so much slower in kaveri compared to a discrete gpu is total proof that fusion was a failure. Having an integrated gpu should have meant vastly reduced RAM requirements. The entire paradigm of storing massive amounts of textures within VRAM should have changed. Too many resources are wasted copying and moving data around from one processor to another when none of that should be happening inside a truly integrated APU. But alas, AMD/ATI was not smart enough or big enough to basically reinvent the entire cpu/gpu concept from the ground up.

 

[tential]

Can we PLEASE keep this thread on topic and actually address the OP's point.
"Yet they power the two big consoles of the next-gen.

Was AMD holding back all this while on their APU line, to differentiate the market? Or does this mean that performance for those next-gen will ultimately be disappointing?"

To sum up the OP's post, he states that because AMD's APUs are in the next gen consoles, they are much more capable than benchmarks/previous products have shown.

It's PAINFULLY clear though that a chip used in a console in which you have direct access and are only coding for ONE chip is vastly different than a PC interface where you're interfacing with the hardware through Windows.

This whole thread is completely based off a false pretense that because the next gen consoles "work" that AMD is holding back. The next gen consoles aren't even anything special. NO ONE in the PC world was impressed. In fact, it's widely held notion throughout the community that the next gen consoles held back PC gaming due to the underpowered hardware.

So why is is that we are now randomly entertaining that there is a VAST amount more power under the hood for APU's when we all know this to not be true?

 

[Abwx]

The fact that 512 GCN cores is so much slower in kaveri compared to a discrete gpu is total proof that fusion was a failure. Having an integrated gpu should have meant vastly reduced RAM requirements. The entire paradigm of storing massive amounts of textures within VRAM should have changed. Too many resources are wasted copying and moving data around from one processor to another when none of that should be happening inside a truly integrated APU. But alas, AMD/ATI was not smart enough or big enough to basically reinvent the entire cpu/gpu concept from the ground up.

Only someone who is unaware of what are hUMA and HSA could write such a post.

 

[Enigmoid]

Only someone who is unaware of what are hUMA and HSA could write such a post.

While HSA and hUMA could certaintly help things they don't help the bandwidth problem unless coded for very low bandwidth situations (almost certainly having an effect on performance).

Kaveri with 512 sp needs bandwidth. That or l3 cache.

AMD needs to address this.

 

[R0H1T]

While HSA and hUMA could certaintly help things they don't help the bandwidth problem unless coded for very low bandwidth situations (almost certainly having an effect on performance).

Kaveri with 512 sp needs bandwidth. That or l3 cache.

AMD needs to address this.

Maybe HBM will come to their rescue :sneaky:

AMD is just a generation away from making the entry level dGPU irrelevant. AMD's problem today is bandwidth which will be addressed with HBM and 2.5D stacking.

http://www.microarch.org/micro46/files/keynote1.pdf

a TSMC 20nm AMD APU with 768 sp and a HBM stack running at 800 Mhz (102.Gb/s) provide GPU performance on par with Xbox One. There are slight compromises like the single thread performance. AMD is working on a new architecture. Excavator is the last of the ill conceived Bulldozer design. A new CPU core architecture is due by late 2015 or early 2016. A 2016 APU built at TSMC 16FF+ with 4 of those new cores with 768 sp and 100 Gb/s bandwidth should satisfy any mainstream gamer.

 

[SViscusi]

While HSA and hUMA could certaintly help things they don't help the bandwidth problem unless coded for very low bandwidth situations (almost certainly having an effect on performance).

Kaveri with 512 sp needs bandwidth. That or l3 cache.

AMD needs to address this.

Bandwidth limitations don't equate to fusion being a failure and it also doesn't mean that there won't be improvements that lessen or eliminate those problems.

 

[SViscusi]

Can we PLEASE keep this thread on topic and actually address the OP's point.
"Yet they power the two big consoles of the next-gen.

Was AMD holding back all this while on their APU line, to differentiate the market? Or does this mean that performance for those next-gen will ultimately be disappointing?"

To sum up the OP's post, he states that because AMD's APUs are in the next gen consoles, they are much more capable than benchmarks/previous products have shown.

Good idea.

in the pc space, their fusion chips I find to be highly fascinating, but ultimately not really competitive vs a traditional intel cpu + discrete GPU.

Yet they power the two big consoles of the next-gen.

Was AMD holding back all this while on their APU line, to differentiate the market? Or does this mean that performance for those next-gen will ultimately be disappointing?

I don't think they're holding back. The environments are too different and the PC area lacks the all investment that Microsoft and Sony has for their consoles to the point where they're treated as a loss leaders. That, because of the nature of the PC distribution market, can't happen so the market for premium technology is limited. Dell or HP aren't going to pay extra for the same type of tech the Sony and MS are willing to because they don't have the customer base for it. If they want to serve that type of market there are premium solutions available, dGPU's.

Then you have limitations of what tech is available at a reasonable price and it's market availability. I'm sure they'd love to go with triple channel memory or add l3 cache, but they're available manufacturing nodes don't make it feasible for them. They also could use a faster memory tech but that's not available yet. They just don't have the resources to fix those problems by themselves and still produce a product that can hit the market at the price point they need to stay in buisness. Sure they theoretically could produce a Kaveri chip with enough onboard memory to feed it, but how many would they sell at $300? Not many, it would be a pyrrhic victory. That's just the reality of the situation.

I ultimately hope AMD's most recent rounds of products actually do lay the groundwork for a change in the approach to the modern PC. At a minimum gains by AMD would result pushing Intel. Last time that happened we got Core 2 and then Core I, all great products.

 

[bunnyfubbles]

You don't think that Steamroller's design wasn't based on compromise?

of course there's compromise for everything, Jaguar cores are simply far more compromised, which is what my stipulation was... emphasis

At least the Jaguar cores don't have shared FPUs.

which does us how much good when its that much slower? have you even bothered to look at the benchmarks? or are you just thinking, "oh hay, moar cores! and they're 'real' cores!"

Quite frankly, if they clocked a bit higher, I would prefer the better IPC + FPUs of the Jaguar cores.

IPC is worse than Pentium C2D...but hey, at least you'd have 8 'real' cores!

 

[frozentundra123456]

Only someone who is unaware of what are hUMA and HSA could write such a post.

Granted it is too soon to call fusion a "failure", but considering AMD's market share and the number of apps that currently use it I would certainly not label it a rip-roaring success so far. Part of that is due to lack of software development, but blame must rest on AMD as well, because only now are truly HSA compatible APUs coming to market, what 4 or 5 years after the ATI acquisition.

I also seems strange that they are charging such a high price for their first HSA capable APU, since one would think they would want to get as many as possible to market to pehaps drive software adoption.

 

[mrmt]

Granted it is too soon to call fusion a "failure"

When do you think it's the time to call it a failure? When they fold the CPU? AMD has lost market share in every segment except in the very bottom since it started to pursue Fusion, and has been consistently losing money in the meantime.

Sure, they had Bulldozer, which is bad enough, but they failed to monetize the GPU part of the APU, and this despite having a significant advantage in this area.

 

[pw257008]

IPC is worse than Pentium C2D...but hey, at least you'd have 8 'real' cores!

At least in CB, Jaguar's on par with AMD's big cores single threaded, and better multithreaded (within its clock range): http://forums.anandtech.com/showthread.php?t=2361017&page=3

 

[Abwx]

Granted it is too soon to call fusion a "failure", but considering AMD's market share and the number of apps that currently use it I would certainly not label it a rip-roaring success so far. Part of that is due to lack of software development, but blame must rest on AMD as well, because only now are truly HSA compatible APUs coming to market, what 4 or 5 years after the ATI acquisition.

As ATI was acquired GPUs requirements kept increasing such that it became economicaly feasible only starting from 40-32nm hence the apparent delay that was actualy primarly node restricted and dependant.

I also seems strange that they are charging such a high price for their first HSA capable APU, since one would think they would want to get as many as possible to market to pehaps drive software adoption.

I cant answer to this question other than with speculations , if price stay high it means logically that the existing production is fully bought at current pricings so AMD has no incentive or pressure to slash their revenue, not counting that OEMs get very large rebates anyway.

 

[SViscusi]

When do you think it's the time to call it a failure? When they fold the CPU? AMD has lost market share in every segment except in the very bottom since it started to pursue Fusion, and has been consistently losing money in the meantime.

And you think that's because of their APU line? Their APU's aren't responsible for their market share, if anything they've been propping that part of their company up. AMD's falling market share is because of their CPU's and their inability to produce a competitive product at anything other than mid range desktops and lower and Intel producing damn fine products that are competitive even in areas where AMD leads.

Sure, they had Bulldozer, which is bad enough, but they failed to monetize the GPU part of the APU, and this despite having a significant advantage in this area.

Do you have any facts to back up this assertion? I would think their console wins and the effect they've had on AMD's bottom line contradict you're statement completely.

 

[mrmt]

And you think that's because of their APU line? Their APU's aren't responsible for their market share, if anything they've been propping that part of their company up. AMD's falling market share is because of their CPU's and their inability to produce a competitive product at anything other than mid range desktops and lower and Intel producing damn fine products that are competitive even in areas where AMD leads.

AMD GPU occupies half the die of AMD APU and yet what determines their MRSP is not the GPU performance, but how they compare to Intel processors in CPU performance, despite the superior iGPU performance AMD APU offers. If this isn't a failure of the strategy, I don't know what it is.

Do you have any facts to back up this assertion? I would think their console wins and the effect they've had on AMD's bottom line contradict you're statement completely.

The console wins are nothing compared to the market share and revenue volumes they lost since the ATI acquisition, let alone the value they had to write down from the ATI acquisition. In fact, whatever increase in operating margins from the console sales wouldn't be enough for AMD to keep the old levels of R&D.

With APU they are selling bigger chips and can't monetize the GPU part, and given their precarious position in CPU performance the bigger GPU wrecks havoc in the cost structure of the product line. As a result, AMD CPU division is smaller than they were BEFORE the ATI acquisition in 2006 by a huge margin.

So yes, APU is a failure.

 

[VirtualLarry]

So yes, APU is a failure.

Cue HSA / OpenCL benchmarks, showing Kaveri outperforming a i7-4770K in LibreOffice spreadsheets.

 

[ShintaiDK]

Cue HSA / OpenCL benchmarks, showing Kaveri outperforming a i7-4770K in LibreOffice spreadsheets.

Thats great for about nothing. Look at the sales, APUs are a massive failure. The customers didnt want it. And AMDs marketshare is still in free fall.

 

[norseamd]

I would love to see a consumer version of the PS4 APU, complete with GDDR5 used for system RAM. I think it would kick ass, relatively speaking. But Sony probably wouldn't let that happen.

It would indeed be interesting, if AMD made an 8-core small-core APU for the desktop space, and added enough GPU CUs to make it comparable to console APUs for gaming purposes.

8 excavator cores with a ps4 sized gpu and gddr5 would kick ass

not sure how you would go about buying gddr5 so you might need it built in with 32 gb since programs are likely to start using up to 16 gb soon

as for graphics you would go with a descrete card even with that large gpu built into the apu. you can use the gpu built into the apu for open cl compute tasks

 

[norseamd]

Thats great for about nothing. Look at the sales, APUs are a massive failure. The customers didnt want it. And AMDs marketshare is still in free fall.

no

the apus have been the most profitable product of their business

 

[ShintaiDK]

no

the apus have been the most profitable product of their business

Documentation please.

Because their marketshare says otherwise. Going from 20% to 15% and still dropping is not a success. And please look at the financials of the CPU division before you call it "most profitable".

http://ir.amd.com/phoenix.zhtml?c=74093&p=irol-newsArticle&ID=1892414&highlight=

The "Computing solutions" aka PC group is a loss with a continual drop in revenue. Thats anything but a success.

 

[norseamd]

Documentation please.

Because their marketshare says otherwise. Going from 20% to 15% and still dropping is not a success. And please look at the financials of the CPU division before you call it "most profitable".

http://ir.amd.com/phoenix.zhtml?c=74093&p=irol-newsArticle&ID=1892414&highlight=

The "Computing solutions" aka PC group is a loss with a continual drop in revenue. Thats anything but a success.

q4 was a profitable quarter no? they sold a facility to make that profit but non the less they still made a profit

the predictions for 2014 seem to say profitable all the way

they have the console business to provide a stabilized and significant source of revenue for several years

the apus in low power applications like notebooks have been more sucessful than desktop cpus or hpc

rory read seems to be working out just fine for amd

they also hired a lot of old talent

so i think the prospects look good for amd