AMD Carrizo Pre-release thread

[turtile]

Like I said before, Rory was likely pushed aside for crushing AMD in high margin areas. He took his experience in PCs which has low margins and high volume which AMD can't do with it's R&D budget.

Now they have Excavator that can only fit into low budget sector until Zen comes a long... AMD doesn't have anything in the mid range/high end or server market at this point. And for ARM, they won't have a custom core out until 2016...

 

[hungtran]

Channel stuffing didn't help Bryne either. http://www.channelregister.co.uk/2015/02/12/amd_stops_chip_shipments/

 

[mrmt]

I don't see this lack of focus especially in comparison to nvidia. It seems all amd wants to do is offer apus in every market...success is another matter.

Server apus - HSA enabled Java .
Embedded apus - crypto acceleration and HSA.
Hedt apus - mantle/dx12 and HSA
Mobile apus - mantle/dx12 and high integration
Semi custom apus - high integration and low cost

This is *exactly* the definition of lack of focus, trying to chase multiple markets with the same line of products. AMD products are becoming jack-of-all-trades-master-of-none because of that.

 

[monstercameron]

This is *exactly* the definition of lack of focus, trying to chase multiple markets with the same line of products. AMD products are becoming jack-of-all-trades-master-of-none because of that.

The aren't quite the same product, they do reuse up blocks and reconfigure them as needed. Also note my comment was to compare AMDs focus compared to nvidias.

 

[mrmt]

The aren't quite the same product, they do reuse up blocks and reconfigure them as needed.

Doesn't matter, as the problem with reutilizing IP blocks in multiple scenarios is that the final products end up bearing the design constraints of all scenarios. That's why companies prefer to develop a specific product for a specific market, so that while the product portfolio gets bigger the final product only gets the constraints of the target market. By burdening their products with multiple requisites of multiple markets AMD is making sure that their products won't be the best in any market they are targeting.

 

[krumme]

The aren't quite the same product, they do reuse up blocks and reconfigure them as needed. Also note my comment was to compare AMDs focus compared to nvidias.

Partly agree; you can be a small car manufacturer and easily have many cars. It depends on eg platform strategy.
And you can be toyota and learn the hard way 25 different airbags is perhaps not the most brilliant approach.
Amd can go a long way using their new blocks and platforms and keep cost down. So its not nessesarily about slimming portfolio to 5%. Thats a mistake.
But the product they make from the blocks needs to have a customer. Slapping a a57 in a server, with a lot of technical benefit, doesnt matter if the customer is just not there. Its foremost the customer who decides what is important and what the value is.

 

[NTMBK]

This is *exactly* the definition of lack of focus, trying to chase multiple markets with the same line of products. AMD products are becoming jack-of-all-trades-master-of-none because of that.

Nvidia does the same. The same GPU is targeting gamers, professional OpenGL apps, and HPC CUDA apps. Targeting multiple markets isn't necessarily bad- if you find sufficient overlap.

 

[mrmt]

Nvidia does the same. The same GPU is targeting gamers, professional OpenGL apps, and HPC CUDA apps. Targeting multiple markets isn't necessarily bad- if you find sufficient overlap.

No, they doesn't. All their strategy is centered around their gpu line and software support, everything else is secondary. AMD's product portfolio is larger than Nvidia's, and they have less money to develop it.

 

[Arachnotronic]

No, they doesn't. All their strategy is centered around their gpu line and software support, everything else is secondary. AMD's product portfolio is larger than Nvidia's, and they have less money to develop it.

That is certainly true today.

What I would do if I were running AMD, is I would slowly but surely exit the consumer processor market altogether (keep the revenue stream from current products in the pipeline flowing for as long as possible) and dedicate all CPU/SoC efforts to embedded, semi-custom, and potentially dense server.

Then, AMD could build a custom "big" ARM core (if they're so inclined) + use off-the-shelf ARM designed Cortex cores depending on the segment and the job that they're tasked to do.

AMD could also double down on developing their GPU IP while it's still not too late so that they can stay reasonably competitive with NVIDIA in consumer and maybe professional GPUs.

 

[Shehriazad]

That is certainly true today.

What I would do if I were running AMD, is I would slowly but surely exit the consumer processor market altogether (keep the revenue stream from current products in the pipeline flowing for as long as possible) and dedicate all CPU/SoC efforts to embedded, semi-custom, and potentially dense server.

Then, AMD could build a custom "big" ARM core (if they're so inclined) + use off-the-shelf ARM designed Cortex cores depending on the segment and the job that they're tasked to do.

AMD could also double down on developing their GPU IP while it's still not too late so that they can stay reasonably competitive with NVIDIA in consumer and maybe professional GPUs.

Or or Oooooooor they actually try to use whatever is left of their fanbase and brand recognition to try the CPU market a last time...namely Summit Ridge and Bristol Ridge.

People also seem to overestimate the value of R&D funding. There are still hundreds of millions going in there...and AMDs architectures are actually very strong(Let's ignore the bulldozer family fiasko for a sec)...they're generally only behind on the nm nodes since they had the brilliant idea of getting rid of their own fabs.
Just "pulling out" would mean willingly just getting rid of something they have already spent hundreds of millions on. (You can be pretty sure that codename Zen has been worked on for like 5 years prior to release)

And their releases are generally server first, anyway. Trying for consumer at least one more time is definitely the right choice.
If AMD truly manages to bring out a multithreading design that is not working with the terrible modules but rather with real cores on a 14/16nm node with DDR4 support...then there is no reason why their CPUs can't be competetive from low to high.

Sure, they will likely never again build a CPU that totally trashes intel...but if AMD can service the high power market with a CPU that only costs 20-40% of Intels high end solution while only having like 10-15% or so less fps in games...then people are going to buy that. If Zen is only HALF as good as promised...then AMD has a chance of actually breaking even again for once...

RX 300 and codename Zen are literally the last 2 cards they have to play...and you're asking them to throw 1 card away when the other is already about to be played, leaving AMD with no options should RX 300 also fail. (Because Semicustom is not going to give them any "real" revenue anytime soon...and the server market...well...they need a new architecture for that first)

 

[frozentundra123456]

Intel's fastest cpu for games is probably 4790k. The price for that is 340.00 on new egg. Do you seriously think Zen will even come close to 90% of the performance of that chip for 135.00 (that *is* 40% of the price)?

 

[erunion]

http://wccftech.com/amd-carrizo-apu...amroller-die-consists-31-billion-transistors/

Carrizo slides are out.

This is must be why there is no desktop Carrizo, above 25W the Kaveri part is clocking higher.

 

[night]

I assume he meant 20-40% less. Still a lot.
And a 4790k can be had for 280, or a 4690 for 200. Not that that helps AMD's position.

 

[Shehriazad]

Euhm...I don't compare that to Intels 4K series...no.

I'm pretty sure Intel will have released some new CPUs in/by 2016...it's obviously going up against that...and if the pricing of Intels ridiculous 5K series is an indicator...then yes, I do mean 20-40% of the price.


You should be comparing DDR4 systems to DDR4 systems. Quite obviously. I mean come on...by 2016 Intel will have tried their best to make the 4K series obsolete...unlike AMD they have billions to throw around. Not to mention DDR4 will by then have approached acceptable pricing and more visible performance gains. (Just like DDR2 to DDR3 was not really rewarding at first and needed a few years to get going)

 

[ShintaiDK]

Or or Oooooooor they actually try to use whatever is left of their fanbase and brand recognition to try the CPU market a last time...namely Summit Ridge and Bristol Ridge.

Are these even officially existing? Everything turns back to some random sweclockers article with no edvidence other than their words.

 

[Shehriazad]

Are these even officially existing? Everything turns back to some random sweclockers article with no edvidence other than their words.

Summit Ridge is just Zen for FM3 and Bristol Ridge being an APU adaptation for FM3.

As long as I don't hear anything else...I'll have to assume that Zen/Summit is still happening ^^;

EVERYTHING is rumors until we get something official...I mean this entire Carrizo discussion was nothing but rumors until a few days ago...that didn't stop people from making ridiculous claims, either.

As far as I'm concerned Summit aka Zen is still happening until AMD gives even the slightest hint that it's not. (Just how they said in Carrizos case that it would not be coming to FM2+ a few weeks ago and until then it was all just rumor based discussion)

 

[ShintaiDK]

Summit Ridge is just Zen for FM3 and Bristol Ridge being an APU adaptation for FM3.

As long as I don't hear anything else...I'll have to assume that Zen/Summit is still happening ^^;

EVERYTHING is rumors until we get something official...I mean this entire Carrizo discussion was nothing but rumors until a few days ago...that didn't stop people from making ridiculous claims, either.

As far as I'm concerned Summit aka Zen is still happening until AMD gives even the slightest hint that it's not. (Just how they said in Carrizos case that it would not be coming to FM2+ a few weeks ago and until then it was all just rumor based discussion)

That wasnt the question. And Carrizo is a product existing on roadmaps and done so since 2013 at least.

Summit Ridge and Bristol Ridge only seem to exist due to a sweclockers undocumented article. So thats why I ask if it actually exist officially.

 

[AtenRa]

(Just how they said in Carrizos case that it would not be coming to FM2+ a few weeks ago and until then it was all just rumor based discussion)

AMD havent said Carrizo will not come to FM2+.

They have only said that BGA Carrizo will be in some AIOs and mini PCs.

 

[richaron]

AMD havent said Carrizo will not come to FM2+.

They have only said that BGA Carrizo will be in some AIOs and mini PCs.

Yes, but it's looking more and more likely that the low power chip with integrated southbridge will be kept for notebooks/AIOs(/minis?).

I can see the current Carrizo team keep working on an appropriate desktop (FM3?) chip with full PCIe 3.0 x16, modified southbridge (off/on die), and optimized for higher clocks. They will be taking over from the current FX chips in project: "keep using old GloFo nodes with old designs".

For the record I have an FM2+ board and hope I'm wrong.

 

[NTMBK]

No, they doesn't. All their strategy is centered around their gpu line and software support, everything else is secondary. AMD's product portfolio is larger than Nvidia's, and they have less money to develop it.

You were complaining specifically about "trying to chase multiple markets with the same line of products", which is indeed what NVidia does. (They inarguably do it much better of course.)

 

[richaron]

You were complaining specifically about "trying to chase multiple markets with the same line of products", which is indeed what NVidia does. (They inarguably do it much better of course.)

Lol, the AMD bashing brigade will try to use anything as ammunition.

That particular point could easily be regarded as "maximising return from investment"; business 101.

 

[mrmt]

Nvidia does the same. The same GPU is targeting gamers, professional OpenGL apps, and HPC CUDA apps. Targeting multiple markets isn't necessarily bad- if you find sufficient overlap.

Nvidia in terms of hardware has basically one GPU architecture focused in graphics, another focused in GPGPU processing and a SoC line that has both a custom and vanilla ARM cores but uses the graphics part of their GPU-focused chips. The custom core part is being scaled back. That's much more focused in my books.

Nvidia is developing different products to different markets. The high end GPGPU products aren't being constrained by the cost requirements for the mainstream PC market, while the mainstream PC parts aren't getting bloated by the compute requirements. That's completely different from GCN, which tries to be everything to everyone.

Same with their CPU line, Nvidia shifted from tablets and phones to embedded, but they were not developing both markets at the same time and their products will have the constraints of only one market. AMD future chips OTOH will be on embedded products, Microservers, mobile market and also on the consumer market, and that means both the software support teams and the IP blocks will have to bear the requirements of all these markets. Oh, and don't forget that they are developing TWO processors that will compete for the same market.

Lol, the AMD bashing brigade will try to use anything as ammunition.

That particular point could easily be regarded as "maximising return from investment"; business 101.

Maybe you could explain us how to burden a product with the requisites of multiple, diverse markets when you have less R&D money than every single one of your competitors is "maximizing returns from investment". With this strategy the only thing AMD is accomplishing is making sure that their products will be always second to something, and have to compete on price.

Ed: I think it's quite clear already that there isn't a single market bracket where AMD is the defining player of it. Intel is the defining player on the entire x86 market, while Nvidia has the marketing leadership on the GPU market. AMD is the follower (budgtet) in every single market they are in and they do not have the scale to beat the leaders on cost, as Samsung does with Apple. There's only two strategies that could work on this scenario, cut down the company to the point it will have adequated scale to compete against the smallest leader (Nvidia), as IDC pointed out, or try to find a market bracket where AMD can be the leader player and phase out the other markets along the way. Both strategies need AMD to get out of the consumer x86 market in the medium term and both strategies need AMD to have a much smaller span than they have now.

 

[frozentundra123456]

Euhm...I don't compare that to Intels 4K series...no.

I'm pretty sure Intel will have released some new CPUs in/by 2016...it's obviously going up against that...and if the pricing of Intels ridiculous 5K series is an indicator...then yes, I do mean 20-40% of the price.


You should be comparing DDR4 systems to DDR4 systems. Quite obviously. I mean come on...by 2016 Intel will have tried their best to make the 4K series obsolete...unlike AMD they have billions to throw around. Not to mention DDR4 will by then have approached acceptable pricing and more visible performance gains. (Just like DDR2 to DDR3 was not really rewarding at first and needed a few years to get going)

Actually for Haswell E, intel both brought out a cheaper hex core, and gave more cores for the same price for the extreme edition. Not sure How that is Ridiculous.

So anyway, what you are saying is Zen will give the performance of a hex core intel for less than 160.00. OK........ Anything is possible I guess. But if Zen manages to reach that level of performance, which is highly debatable, AMD would be fools to sell it for 160.00.

 

[Abwx]

So anyway, what you are saying is Zen will give the performance of a hex core intel for less than 160.00. OK........ Anything is possible I guess. But if Zen manages to reach that level of performance, which is highly debatable, AMD would be fools to sell it for 160.00.

By the time it s released current CPUs prices will be lower, not counting that it is about 100% sure that Intel would be forced to substancialy lower their i7 prices if the thing perform well enough, so perhaps not as low as 160 but probably not as high as 280-300.

 

[inf64]

Actually for Haswell E, intel both brought out a cheaper hex core, and gave more cores for the same price for the extreme edition. Not sure How that is Ridiculous.

So anyway, what you are saying is Zen will give the performance of a hex core intel for less than 160.00. OK........ Anything is possible I guess. But if Zen manages to reach that level of performance, which is highly debatable, AMD would be fools to sell it for 160.00.

AMD will price its products more or less in accordance to its performance. I think Zen will be competitive with Broadwell and more or less with Skylake, depending on workloads. Pricing might be lower in order to recapture the market share, if they chose such an approach.