How long does it take AMD to design and bring a new die variation to market?

[NTMBK]

Once die stacking takes off, this should get a lot quicker (and cheaper). Take a look at this presentation for more info: http://www.microarch.org/micro46/files/keynote1.pdf But basically, instead of building a bunch of independent parts into an SoC at the die level, you put multiple individual dies on an interposer. Maybe you have a die for the CPU, a die for the GPU, a die for the LLC, a die for the memory controller, and a die for the physical controller stuff, all on an interposer. You want a new SKU with a bigger GPU? You just redesign the GPU die, keep everything else the same, and boom, new part. You want to add your custom crypto coprocessor? Put it on the interposer. Compression acceleration? Put it on the interposer. Want HBM? Put it on the interposer. It should be a massive boon for the semicustom BU.

 

[cytg111]

I like this one

"There is research to be done for cooling irregular shapes"

I imagine so..

 

[cbn]

Another nice thing about having a hexcore die.....I'll bet the chances of yielding a dual core are extremely low.

This, in contrast, to the current quad core/dual module APUs where dual cores SKUs are prevalent.

 

[cbn]

Regarding the large amounts of cache on Vishera (6MB L2 and 8MB L3), I am assuming this is for the server market?

Because for a desktop enthusiast chip (especially one that currently struggles at various price points), I would imagine AMD could eliminate a lot of that cache and end up with a much better performance to cost ratio.

So if reducing cache down, what would be the best way to go about this?

P.S. Currently Excavator in Carrizo appear to have 1MB L2 cache per module. There is no L3 cache like other previous APUs. (Meanwhile, each core in Haswell only has 256 KB of L2 sharing a varying amount of L3 with other cores depending on the SKU level)

 

[cbn]

http://www.tomshardware.com/news/amd-carrizo-carrizo-l-notebook-apu,28345.html

AMD is debuting its high-performance Carrizo and Carrizo-L system-on-a-chip design at the 2015 Consumer Electronics Show (CES), and we're the first to offer pictures of the prototype.

While Carrizo-equipped systems won't be available until the mid-year time frame, the company has working silicon in hand. AMD has thin and light prototype systems jointly developed with Shuttle Inc. and is showing them off at CES. According to Kevin Lensing, Senior Director, Product Management, AMD, "It's a 'Carrizo'-based system sporting an upcoming mobile discrete GPU and it delivers exceptional, unprecedented gaming performance in a thinner and lighter platform than previously possible due to the energy efficiency and performance focus we have in 'Carrizo.'" The CES demonstrations are limited to private appointments and aren't available to the general public, but Tom's Hardware will be there and we look forward to seeing what Carrizo can do.

Tom's also touts the company's new unified platform strategy: "Our scalable approach, which hasn't been used in the PC industry for many years now, addresses the complete stack from top to bottom with a single platform, meaning that our OEM partners will be able to use "Carrizo" and "Carrizo-L" to serve a wide range of the market – spanning form factors and price points from all-in-ones and mainstream notebooks, to ultra-thin and light performance and gaming notebooks – without needing to design, develop, and validate multiple platforms."

Carrizo checks a lot of boxes: a unified SoC platform that the company claims can scale from mainstream to enthusiast class performance, the new Excavator CPU core, the GCN graphics architecture, Mantle and DirectX 12 API support, Project FreeSync support, Heterogeneous System Architecture (HSA) 1.0, and an integrated ARM TrustZone processor on die.

Despite the impressive feature list, it may come down to plain-and-simple ability. Is Carrizo capable of generating playable frame rates in games while using an acceptably low amount of battery power? AMD claims that "…'Carrizo' delivers HD and Ultra HD graphics horsepower, increased compute performance and next-generation mobile gaming," and our discussions with representatives lead us to believe that the company considers this an important aspect of Carrizo's arsenal. We've also been told that technology research firm Gartner has indicated that thin and light notebook gaming is on the rise.

We're inclined to agree. If Carrizo has the chops to handle PC games in HD while maintaining modern power-efficiency expectations, it would open a new market up to AMD. Imagine a capable, reasonably-priced gaming notebook equipped with a built-in FreeSync monitor. But AMD has had capable on-die graphics chipsets for some time now. The question is, can it get power usage under control?

John Byrne, AMD's president and general manager of the company's computing and graphics business unit, has gone on the record saying that "…we're bringing a brand new graphics architecture integrated into Carrizo. This will be our biggest leap ever from an energy efficiency perspective." We're not sure how the company has tweaked GCN in order to achieve this purported leap in energy savings, but we've recently noted that AMD Appears To Have Power-Saving 'Dynamic Frame Rate Control' Feature Incoming. Maybe this has something to do with AMD's claims, maybe not, or maybe it's just one piece of the puzzle. Either way, we can't wait to get our hands on a Carrizo for testing later in 2015. In the immediate future, stay tuned for our impressions of our hands-on time with the Carrizo prototype system, coming very soon.

In addition to the usual comments we normally see like "Carrizo won't be available till mid 2015" and "Common BGA socket for Carrizo and Carrizo-L" I was really surprised to find out the AMD Carrizo prototype also has a discrete GPU.

For these gaming notebooks that will be using discrete graphics, maybe that is another reason to create an additional die variation.

Hexcore Excavator with 64 GCN stream processors could work for both notebook and FM2+, while at the same time providing a cheaper and higher performance platform than the more integrated Carrizo.

The major difference would be that the Hexcore die would also need a southbridge for desktop and notebook. (Carrizo already has this integrated).

 

[Enigmoid]

The usual hype that was heard with Kaveri. Not reading anything into it for several reasons.

No performance benchmarks (they say they only got the ES chips several weeks ago) which is fine. But then how can they know what kinds of gains are possible?

When I see the benchmarks then I will believe.

 

[Arachnotronic]

The usual hype that was heard with Kaveri. Not reading anything into it for several reasons.

No performance benchmarks (they say they only got the ES chips several weeks ago) which is fine. But then how can they know what kinds of gains are possible?

When I see the benchmarks then I will believe.

Intel showed working prototypes of Skylake back in September. If Carrizo silicon just came back from the fab a "few weeks ago" then this is pretty bad.

 

[Abwx]

Intel showed working prototypes of Skylake back in September. If Carrizo silicon just came back from the fab a "few weeks ago" then this is pretty bad.

So you have info that Intel s SL was definitive silicon while current Carrizo is not a definitive die.

More seriously AMD did a demo, what about Intel s demo, what did they display..??.A cpu that manage to boot windows.?.

 

[rgallant]

[ADDENDUM] We have since had an opportunity to see Carrizo running in three systems at CES. The first was the shuttle-developed laptop pictured above. In person it looks much more impressive with a red illuminated keyboard and stylish cutouts, like something you'd see on ASUS' ROG series. We weren't allowed to take a picture, unfortunately, but it left a positive impression.
The second system was running Lego Batman. It's a nice-enough looking game, and Carrizo ran it well, but certainly not one of the most demanding titles out there. We'd rather have seen something that requires a little more horsepower.
The third system showed 4K video playback, which is accelerated by Carrizo's new HEVC hardware decoder block. It played wonderfully smoothly next to a Core i3 that was obviously stuttering. We're told that Carrizo uses about 5 to 6% CPU utilization during this test, while the Core i3 was almost maxed out.
That's the extent of hardware demos we were shown. The only other information we gleaned is that Carrizo-L will draw in the 15 to 19 Watt neighborhood, while Carrizo will have a TDP up to 35 Watts. This certainly puts AMD's power draw in the neighborhood of Haswell-U and Broadwell-U chips. We'll have to wait until we can perform tests to see where performance sits, and to find out what Carrizo design wins will be priced when they hit retail before we can assess whether it will be the game-changer AMD suggests it will. [/ADDENDUM]

 

[krumme]

[ADDENDUM] We have since had an opportunity to see Carrizo running in three systems at CES. The first was the shuttle-developed laptop pictured above. In person it looks much more impressive with a red illuminated keyboard and stylish cutouts, like something you'd see on ASUS' ROG series. We weren't allowed to take a picture, unfortunately, but it left a positive impression.
The second system was running Lego Batman. It's a nice-enough looking game, and Carrizo ran it well, but certainly not one of the most demanding titles out there. We'd rather have seen something that requires a little more horsepower.
The third system showed 4K video playback, which is accelerated by Carrizo's new HEVC hardware decoder block. It played wonderfully smoothly next to a Core i3 that was obviously stuttering. We're told that Carrizo uses about 5 to 6% CPU utilization during this test, while the Core i3 was almost maxed out.
That's the extent of hardware demos we were shown. The only other information we gleaned is that Carrizo-L will draw in the 15 to 19 Watt neighborhood, while Carrizo will have a TDP up to 35 Watts. This certainly puts AMD's power draw in the neighborhood of Haswell-U and Broadwell-U chips. We'll have to wait until we can perform tests to see where performance sits, and to find out what Carrizo design wins will be priced when they hit retail before we can assess whether it will be the game-changer AMD suggests it will. [/ADDENDUM]

Okey it seems its finally Seronx time or what?
Hbm gcn2 whatever is imo not enough to reach those levels with that power hog of a cpu arch.

 

[Abwx]

Hbm gcn2 whatever is imo not enough to reach those levels with that power hog of a cpu arch.

You have power comsumption numbers to back this claim.?.

 

[Enigmoid]

[ADDENDUM] We have since had an opportunity to see Carrizo running in three systems at CES. The first was the shuttle-developed laptop pictured above. In person it looks much more impressive with a red illuminated keyboard and stylish cutouts, like something you'd see on ASUS' ROG series. We weren't allowed to take a picture, unfortunately, but it left a positive impression.
The second system was running Lego Batman. It's a nice-enough looking game, and Carrizo ran it well, but certainly not one of the most demanding titles out there. We'd rather have seen something that requires a little more horsepower.
The third system showed 4K video playback, which is accelerated by Carrizo's new HEVC hardware decoder block. It played wonderfully smoothly next to a Core i3 that was obviously stuttering. We're told that Carrizo uses about 5 to 6% CPU utilization during this test, while the Core i3 was almost maxed out.
That's the extent of hardware demos we were shown. The only other information we gleaned is that Carrizo-L will draw in the 15 to 19 Watt neighborhood, while Carrizo will have a TDP up to 35 Watts. This certainly puts AMD's power draw in the neighborhood of Haswell-U and Broadwell-U chips. We'll have to wait until we can perform tests to see where performance sits, and to find out what Carrizo design wins will be priced when they hit retail before we can assess whether it will be the game-changer AMD suggests it will. [/ADDENDUM]

Does Haswell have a problem with 4K video?

 

[Phynaz]

Does Haswell have a problem with 4K video?

Nope. I'm positive AMD was using a software decoder on the i3 vs a hardware decoder on their chip.

 

[Abwx]

Does Haswell have a problem with 4K video?

What is not said in this addendum is that it s H265 4K.

Nope. I'm positive AMD was using a software decoder on the i3 vs a hardware decoder on their chip.

And i m positive that you re completely wrong, read above what it is about, and find us some prove that Haswell run such videos since you re so confident that Intel s CPU s are flawless...

 

[Phynaz]

What is not said in this addendum is that it s H265 4K.



And i m positive that you re completely wrong, read above what it is about, and find us some prove that Haswell run such videos since you re so confident that Intel s CPU s are flawless...

You just proved my point. Thank you.

 

[Abwx]

You just proved my point. Thank you.

Lol, denial as answer, check if Haswell can decode 4K H265, of course you wont since you know the answer, keep on living on a fairy tales world, that was your point, actualy..

HEVC (H.265) decode acceleration is absent too

HEVC was standardized pretty recently, and Haswell's GPU would have long been past the design stage by that time.

http://www.anandtech.com/show/7007/intels-haswell-an-htpc-perspective/10

 

[Phynaz]

Lol, denial as answer, check if Haswell can decode 4K H265, of course you wont since you know the answer, keep on living on a fairy tales world, that was your point, actualy..

http://www.anandtech.com/show/7007/intels-haswell-an-htpc-perspective/10

As I said, AMD was using a hardware decoder on their chip and a software decoder on the intel chip. If they were comparing chips on a level playing field they would have used a hardware decoder on Core M. That would have been a comparison against what Intel will be selling when Carrizo gets around to shipping.

You would never expect Intel to make a fair comparison yet you expect AMD to make one. To think that any company would make such a comparison is just naive. Or perhaps people that think so live in a fairytale world.

To further Intel's defense, neither NVIDIA nor AMD support these two features

You forgot to quote that part of the paragraph. An Intel CPU can be bought today with hardware h.265 decode and encode. There is not an AMD cpu that you can buy today with these features.

 

[Enigmoid]

The other fact is that Carrizio won't be ready for months and HW is nearing the end of its life. Really the comparison is Carrizio vs. BW-U.

 

[Idontcare]

We weren't allowed to take a picture, unfortunately, but it left a positive impression.

Not allowed to even take a picture?

That doesn't inspire confidence that AMD even has confidence with their own demo setup when they block any effort to document the very existence of the demo equipment itself. :|

I guess we are back to the "it just feels more fluid" subjective journalism angle again :\

The second system was running Lego Batman. It's a nice-enough looking game, and Carrizo ran it well, but certainly not one of the most demanding titles out there. We'd rather have seen something that requires a little more horsepower.

So even to the audience who can't be allowed to take pictures, the setup seems a little hoke on the face of it.

The third system showed 4K video playback, which is accelerated by Carrizo's new HEVC hardware decoder block. It played wonderfully smoothly next to a Core i3 that was obviously stuttering. We're told that Carrizo uses about 5 to 6% CPU utilization during this test, while the Core i3 was almost maxed out.

They are "told" what the CPU utilization is, but not allowed to check and verify themselves? What is the point of the demo?

Hey I have a computer that runs on zero-point energy and actually charges the battery itself while you use it! What's that? You want to take a picture? No way ma'am! This is proprietary hardware here. Oh and by the way our competition sucks, just take our word for it. Please go print that on your website, please, we kinda need anything we can get at this point, we are that desperate.

 

[Abwx]

As I said, AMD was using a hardware decoder on their chip and a software decoder on the intel chip. If they were comparing chips on a level playing field they would have used a hardware decoder on Core M. That would have been a comparison against what Intel will be selling when Carrizo gets around to shipping.

You would never expect Intel to make a fair comparison yet you expect AMD to make one. To think that any company would make such a comparison is just naive. Or perhaps people that think so live in a fairytale world.

Ok, fair point, that s not usual from you..

That said we can speculate that they settled the Intel plateform as well as they could with its current possibilities in matter of acceleration.

You forgot to quote that part of the paragraph. An Intel CPU can be bought today with hardware h.265 decode and encode. There is not an AMD cpu that you can buy today with these features.

I forgot nothing since Kaveri wasnt released at the time and it can decode H265 using its HSA features, although surely not at 5-6% CPU usage like Carrizo.

Together with Telestream, AMD has developed HEVC codec that uses HSA that's able to play 4K HEVC content on Kaveri with a very low load on the CPU.

 

[Abwx]

Not allowed to even take a picture?

That doesn't inspire confidence that AMD even has confidence with their own demo setup when they block any effort to document the very existence of the demo equipment itself. :|

According to hardware.fr review it was the laptop that couldnt be pictured, the reason is that it was a raw sample that wasnt as good looking as the one in the published pics, anyway i like it how you interpreted it the most negatively possible, now can i also use my ignorance of a given fact as an excuse to post worthless argumentations..?.

 

[Arachnotronic]

You forgot to quote that part of the paragraph. An Intel CPU can be bought today with hardware h.265 decode and encode. There is not an AMD cpu that you can buy today with these features.

Broadwell has a hybrid H.265 decode, but it does not have H.265 encode. Skylake is the part that comes with both hardware encode/decode.

With that in mind, a little unfair on AMD's part to compare fresh-out-of-the-fab Carrizo with Haswell chips that have been on the market since mid-2013. That said, in the mobile world (where Intel is the underdog) it is probably just as guilty of unfair benchmarks/comparisons.

To my knowledge, Intel demo'd Skylake back in Sept. 2014, and probably had silicon back well before then. That chip is more capable in the media encode/decode area than either Carrizo or Broadwell. I'm reasonably sure Intel could have released Skylake around, well, now, but OEMs probably wanted to keep selling current machine designs (with updated CPUs) until the big Windows 10 launch.

 

[ctsoth]

They are "told" what the CPU utilization is, but not allowed to check and verify themselves? What is the point of the demo?

Hey I have a computer that runs on zero-point energy and actually charges the battery itself while you use it! What's that? You want to take a picture? No way ma'am! This is proprietary hardware here. Oh and by the way our competition sucks, just take our word for it. Please go print that on your website, please, we kinda need anything we can get at this point, we are that desperate.

If I'm not mistaken, andI'm pretty sure I'm not, the 4k hardware decoder is a component of GCN 1.2ish (whatever is actually inside of Carrizo), so it would make complete sense to claim the CPU usage is 5-6% from a marketing perspective, because the CPU isn't actually decoding it.

The real question is, what is the GPU load during 4k playback?

As you can see in this review, GCN 1.2 found in the r9 285 supports H264, so I am confident in speculating that further 4k decode/encode support was added to the Carrizo GPU

http://www.anandtech.com/show/8460/amd-radeon-r9-285-review/4

 

[ShintaiDK]

If I'm not mistaken, andI'm pretty sure I'm not, the 4k hardware decoder is a component of GCN 1.2ish (whatever is actually inside of Carrizo), so it would make complete sense to claim the CPU usage is 5-6% from a marketing perspective, because the CPU isn't actually decoding it.

The real question is, what is the GPU load during 4k playback?

As you can see in this review, GCN 1.2 found in the r9 285 supports H264, so I am confident in speculating that further 4k decode/encode support was added to the Carrizo GPU

http://www.anandtech.com/show/8460/amd-radeon-r9-285-review/4

The GPU load should be the same as on 285 I think. Its more or less all handled by the UVD block. Encode however may be entirely different.

 

[ctsoth]

The GPU load should be the same as on 285 I think. Its more or less all handled by the UVD block. Encode however may be entirely different.

You may be right, but assuming all the work is done by the CPU and CPU usage is Truly 5-6% we can conclude that barring OS/background processes AMDs newest two module super chip is capable of encoding up to 20 H265 streams at once.

Does this strike anyone else as absurd?