Kabini Rumors

[Arkadrel]

This is hardly what you expect from a node shrink + Architectural enhancements (unless we're talking Intel).

Lets wait for reviews.
One thing is listed TPDs,... but avg power loads during tasks is much more important.
Id love to see them side by side, a C-60 and GX-210HA in a review.

I wish AMD would just go massively into cores. 16 core Jaguar 100W-125W, anyone?

16 cores at 2ghz.... probably closer to 65watts, than 100watts.
Assumeing it doesnt have 4 times as much iGPU + FCH as the 25w version.

I dont think such a product would be anything special.
AMD should just let the server side die and focus on makeing chips for normal consumers.

What I really want, is a Jaguar quad core clocked like ~3ghz and much much faster IGP.
So lets say you devote 30watts TPD to the CPU, and then spend 35watts on the IGP,
That way you end up with a 65watt chip.

Where the IGP is around ~7750 level of performance.

Why AMD hasnt made a product like this is what I find silly.
If it takes 4x memory channels (Quad channels) to reach bandwidth, or it takes GDDR5 system memory, so be it.

I suspect a product like that, for laptops and desktops, would kill alot of cards.

 

[Vesku]

I'd expect integrated kabini+motherboard for ~$100, would be shocking pricing for just the APU. Based on what it's replacing and that it's going to be up against OoO Atom after all.

 

[insertcarehere]

Lets wait for reviews.
One thing is listed TPDs,... but avg power loads during tasks is much more important.
Id love to see them side by side, a C-60 and GX-210HA in a review.


16 cores at 2ghz.... probably closer to 65watts, than 100watts.
Assumeing it doesnt have 4 times as much iGPU + FCH as the 25w version.

I dont think such a product would be anything special.
AMD should just let the server side die and focus on makeing chips for normal consumers.

What I really want, is a Jaguar quad core clock like 3ghz and much much faster IGP.
So lets say you devote 30watts TPD to the CPU, and then spend 35watts on the IGP,
That way you end up with a 65watt chip.

Where the IGP is around ~7750 level of performance.

Why AMD hasnt made a product like this is what I find silly.

There's the apu inside the ps4...

 

[Pilum]

The opposite extreme is full single-threaded where the IPC gain needs to be 43% coming from 1.5GHz (pushing past high end of expectations) to 71% for 2.1GHz (way past realistic for me).

They'll have to reach at least +50% IPC if they really expect Silvermont and following generations to be successful in the market - I'm not thinking AMD here, but the upcoming ARM architectures. And Intel knows full well their current business model depends on Silvermonts success. If they reach less than 50% improvement, I'd regard that as their biggest failure since Netburst.

 

[Arkadrel]

There's the apu inside the ps4...

When can I buy a PC solution like it? will there even be one?

 

[Idontcare]

Edit : Overall TSMC s bulk looks quite average
not to say crappy , there s no doubt that
the much trashed GF has better process.

If they are going to take an extra 1-2yrs to develop it and get it into production (versus the competition), it sure as heck ought to be better.

Project management 101.

 

[NTMBK]

When can I buy a PC solution like it? will there even be one?

One can only hope...

 

[Arkadrel]

They'll have to reach at least +50% IPC if they really expect Silvermont and following generations to be successful in the market.

No & No.

IPC wont increase 50%, and not its not needed.
Why? because they ll have silvermont running at higher speeds than Cedar-Trail does, also they ll have Quad cores.

And Intel knows full well their current business model depends on Silvermonts success.

Meh... everyone is trying to get into Tablets/Laptops.
But Im pretty sure that more than 1/3 rd of Intels bussiness is from servers (revenue), and that their HDDs and normal computers also make up a decent portion.

Their current bussiness model doesnt depend on their new Atoms.
There will still be servers, and normal PCs sold, and HDDs by Intel ect.

I wonder how much Intel makes off of tablets / laptops (cpu inside only) atm?
I dont expect that to be even 1/3rd of their revenue.

If they reach less than 50% improvement, I'd regard that as their biggest failure since Netburst.

High expectations.

 

[beginner99]

They'll have to reach at least +50% IPC if they really expect Silvermont and following generations to be successful in the market - I'm not thinking AMD here, but the upcoming ARM architectures. And Intel knows full well their current business model depends on Silvermonts success. If they reach less than 50% improvement, I'd regard that as their biggest failure since Netburst.

IMHO current 32 nm medfield in RAZR I is pretty competitive. The great thing about that phone is long talk time and standby time, much better compared to the ARM based RAZR M.
Were it sucks is in anything that uses the GPU probably due to it being old uArch and crappy intel drivers.
Most reviewers claim it is snappier than ARM offerings at same price.

So if IPC does increase by 50% at same power level as medfield, they have a big winner in performance/watt, way better than Cortex A15 and especially with a lot better single-threaded performance.

 

[frozentundra123456]

When can I buy a PC solution like it? will there even be one?

I agree it would be great to have an APU with the graphical performance of the PS4. However, that is a purpose built unit with the bulk of the power and thermal budget devoted to graphics as it should be. For desktop, I dont really see the purpose of 8 slow cores with a strong APU.
Considering the overhead of windows and the need to do more CPU intensive apps, I am not sure a PS4 type APU would be a great solution.

Now if they could put the gpu power of the PS4 with 4 full power cpu cores, that would be an interesting chip. Dont know if they can do this because of size, power and thermal budget though.

Edit: I am not sure it would even make good business sense. If they sell you a cpu for 100 dollars (say FX6300) and a gpu (say 7850) for 200.00, why would they want to sell you an apu with the same capabilities unless they can get 300.00 for it, which I doubt is possible. I am not saying that AMD is purposely holding back APUs, I am just saying that in a way they are competing with their own GPU division if APUs become too powerful.

 

[Arkadrel]

edited.

Now if they could put the gpu power of the PS4 with 4 full power cpu cores, that would be an interesting chip. Dont know if they can do this because of size, power and thermal budget though.

http://www.xbitlabs.com/news/cpu/di..._Kaveri_APU_Supports_GDDR5_Memory_Report.html

With the next-generation high-performance APU code-named Kaveri, AMD will finally unleash its Radeon HD’s potential thanks to a secret weapon, 128-bit GDDR5 memory controller.

MD Fusion A-series APU code-named Kaveri is projected to feature up to four Steamroller high-performance x86 general-purpose cores as well as GCN architecture-based AMD Radeon HD 7000 graphics engine. However, the most unexpected feature of Kaveri will be its 128-bit memory controller, which will support both DDR3 as well as GDDR5 memory.

By contrast, an inexpensive 128-bit GDDR5-based memory sub-system with 3400MHz effective clock-speed enables 54.4GB/s of memory bandwidth, or 82% higher compared to DDR3-based subs-system.

Its gonna happend at some point.
Its just taken AMD too long to make this product.

 

[lagokc]

When can I buy a PC solution like it? will there even be one?

It honestly would surprise me if it takes more than a week for someone to figure out how to install Linux on the PS4... maybe 2 weeks for Windows.

Considering the 7750/7770 have 72GB/s memory bandwidth and aren't really starved, Kaveri looks like it's going to be the first APU that's actually good enough to play games without frustration.

 

[Arkadrel]

It honestly would surprise me if it takes more than a week for someone to figure out how to install Linux on the PS4... maybe 2 weeks for Windows.

Lmao.... a PS4 that could dual boot into linux, and then boot as a PS4 for gameing.

That might appeal to alot of people....
I wonder if Sony would support such a project? It would be bound to help sales of the PS4.

 

[nehalem256]

Lmao.... a PS4 that could dual boot into linux, and then boot as a PS4 for gameing.

That might appeal to alot of people....
I wonder if Sony would support such a project? It would be bound to help sales of the PS4.

I think a lot would depend on if they are losing money selling the hardware.

 

[NTMBK]

Lmao.... a PS4 that could dual boot into linux, and then boot as a PS4 for gameing.

That might appeal to alot of people....
I wonder if Sony would support such a project? It would be bound to help sales of the PS4.

Boot as a PS4 for gaming? Nah, just fire up Steam.

 

[Arkadrel]

Boot as a PS4 for gaming? Nah, just fire up Steam.

^ Probably a reason against allowing dual booting on a ps4.
You dont want to add competition in gameing, if thats your bread and butter.

Still.... Linux ability on a PS4 would be "neat".
It could attract a larger crowd than just gamers.

Once I can buy a Kaveri with GDDR5 system memory, I think i ll be tempted to get one.
I wanna see what IGPs can do at that point, when system memory bandwidth isnt holding them back as much.

 

[lagokc]

^ Probably a reason against allowing dual booting on a ps4.
You dont want to add competition in gameing, if thats your bread and butter.

Still.... Linux ability on a PS4 would be "neat".
It could attract a larger crowd than just gamers.

If it's Steam for Linux I don't think there's much competition to worry about :biggrin:

Yes, I realize there's a chance several years from now I will be wrong.

 

[Pilum]

No & No.

IPC wont increase 50%, and not its not needed.
Why? because they ll have silvermont running at higher speeds than Cedar-Trail does, also they ll have Quad cores.

Higher speed is true, but I still think that aiming for less than Bobcat IPC is not overly optimistic (Bobcat has ~70% higher IPC in integer than the first-gen Atoms). Even with the higher speeds, +50% IPC would mean rough performance parity with Jaguar@2.0. I don't think Intel is aiming at being only as good as the competition... but we'll see.

Meh... everyone is trying to get into Tablets/Laptops.
But Im pretty sure that more than 1/3 rd of Intels bussiness is from servers (revenue), and that their HDDs and normal computers also make up a decent portion.

In 1Q13 revenue was 22.4% in servers, 69.0% client, 8.6% other.

Their current bussiness model doesnt depend on their new Atoms.
There will still be servers, and normal PCs sold, and HDDs by Intel ect.

I'm not thinking about x86 as such, but their position as leading silicon manufacturer and IDM. In order to afford their leading-edge process development, they have to push wafers. If they loose wafer volume, they loose the economic foundation of being the leading-edge silicon manufacturer. Then TSMC will catch up. After that, performance parity vs. the rest of design houses.

Intel needs volumes. Traditional x86 client volumes are shrinking; while the PC market won't go away, it will end up at a lower volume. Intel needs volume in other markets. Low-power mobile is one important segment; Xeon Phi is another approach to fill up wafers. They really have to succeed with at least one of those or their traditional model is in danger. They seem to have a plan B, thus the beginnings of foundry work; but I'm pretty sure Intel would like to prevent the cruel fate of becoming a foundry (even if they'd be the best foundry in the world).

I wonder how much Intel makes off of tablets / laptops (cpu inside only) atm?
I dont expect that to be even 1/3rd of their revenue.

Mobile is 52.5% from overall x86 marketshare currently (Mercury Research), so given 69% Intel client revenue share it's roughly 36% revenue.

High expectations.

Well, yes. But probably more realistic than the people expecting Steamroller +20% IPC single-threaded.

 

[frozentundra123456]

edited.



http://www.xbitlabs.com/news/cpu/di..._Kaveri_APU_Supports_GDDR5_Memory_Report.html








Its gonna happend at some point.
Its just taken AMD too long to make this product.

Yes, but Kaveri has only 512 sp, less than the PS4, and it will not be optimized as well as the PS4. In fact that is the same number of SP as the 7750, so even assuming no thermal or bandwidth limitations, it will be equivalent to low end discrete at best. Still, progress I grant, but a limited solution considering there hopefully will be more powerful low/midrange discrete cards by then also.

 

[Pilum]

So if IPC does increase by 50% at same power level as medfield, they have a big winner in performance/watt, way better than Cortex A15 and especially with a lot better single-threaded performance.

Yes, but that is a minimum requirement if they want to become a premium mobile CPU supplier. I expect next-gen ARM @TSMC 20nm to hit the market before the 14nm Atoms come out, and that is the actual performance target to beat, or at least be competitive with. Intel has to show that they can continously deliver the best CPU performance in mobile space if they really want to make inroads into the already well-established ARM ecosystem. If the performance just switches between Intel and ARM each generation, their growth opportunities will be very limited IMO.

 

[Pilum]

^ Probably a reason against allowing dual booting on a ps4.
You dont want to add competition in gameing, if thats your bread and butter.

Especially if you make a fixed profit on every console game sold, which is how the console market works. That is the mechanism by which console manufactuers can afford to sell hardware at a loss; they expect to make back the money as people keep buying games for the console.

So, no Steam. Never, ever.

 

[mrmt]

Its gonna happend at some point.
Its just taken AMD too long to make this product.

This.

You can't really be late in the MPU business. If you are late, your product that was suppose to splash the competition will only be equivalent.

Just look on how Phenom II would be a good processor if it were take on Conroe not Westmere. And look how Llano would suddenly be an interesting choice if it had to compete against Westmere, not Sandy Bridge.

As a matter of fact, Kaveri will spend most of its useful life competing against Broadwell, not Haswell, and with Broadwell comes yet another buff in GPU capacity, caches, CPU performance, power consumption, etc etc etc.

 

[Enigmoid]

I like amd's idea with system GDDR5 but the inherent problem with it is that GDDR5 is expensive.

Its going to get harder and harder to sell a system that only has 4GB RAM with no chance of upgrade in the future (by the time the system becomes available 4GB is really gonna push it). Personally I would not buy such a system.

8GB is probably going to cost too much to be cost effective vs a low end discrete card with comparable performance.

 

[MightyMalus]

Can't GDDR5 be stacked with the ones in the dGPU? Making a "Dual-G" setup much better? Just a thought, I'm clueless on that.

 

[Vesku]

This.

You can't really be late in the MPU business. If you are late, your product that was suppose to splash the competition will only be equivalent.

Just look on how Phenom II would be a good processor if it were take on Conroe not Westmere. And look how Llano would suddenly be an interesting choice if it had to compete against Westmere, not Sandy Bridge.

As a matter of fact, Kaveri will spend most of its useful life competing against Broadwell, not Haswell, and with Broadwell comes yet another buff in GPU capacity, caches, CPU performance, power consumption, etc etc etc.

Yes, if they had taken the WSA hit right away and just committed to 28nm TSMC for Brazos successor they would have had a year to themselves in that segment of x86 rather than ~3 months. But they didn't and at least anyone buying this winter will be comparing them to whatever Intel manages to get out this year. I think you are off on Kaveri though, unless you think Haswell will have a 6-9 month span as main segment.