Kabini Rumors

[mrmt]

Just found this about Kabini.

http://www.cpu-world.com/news_2013/2013041501_Rumored_details_of_embedded_Kabini_APUs.html

In January, AMD revealed plans to release three new processor cores, called "Kabini", "Richland", and "Temash". Mobile "Richland" CPUs were introduced in March. Desktop "Richland" processors, together with "Kabini" and "Temash" APUs, will be available in the second quarter. "Temash" and "Kabini" will be aimed at tablet and mobile markets, but there will be also embedded "Kabini" parts. We found some details on three embedded SKUs in specifications of upcoming Mini-ITX motherboard. It is becoming extremely difficult to find any information on unreleased AMD products, therefore we would prefer not to disclose the make and model of this board. We could not confirm these specifications from any other source, therefore please treat them as a rumor.
Specifications for unnamed motherboard include three Kabini "G-Series" microprocessors. On the low-end, it lists a dual-core APU with 1 GHz frequency and 9 Watt TDP. Another dual-core chip will operate at 1.65 GHz, and it will have 15 Watt TDP. On the high-end, there will be a low voltage part with 4 cores, 2 GHz clock frequency and 25 Watt TDP. All processors will support up to 16 GB of DDR3-1600 and DDR3-1866 memory. You may notice that the features of quad-core APU are identical to A10-4655M, the fastest at the moment low-power Trinity chip. Also, we suspect that the characteristics of embedded products will be similar to mobile "Kabini" APUs.

If this is true, I wound't be very upbeat about Kabini. It won't be able to compete with Haswell, let alone Broadwell in the bottom market, except on price. Frequencies are too low to match Trinity. Temash won't be able to match Silvermont too.

I don't see in Kabini all the attributes needed for tablets or notebooks, but it's the perfect chip for embedded. It is efficient, not top efficiency, but good enough efficiency. It is faster, but only good enough. Pricing should be very interesting too. Add the reusable IP part blocks that Rory Read talks all the time and we have a winner on the embedded market, much more than Brazos.

It seems that Kabini optimization for the embedded market weren't an afterthought for AMD, but that the chip was thought to be cheaper from its conception. I wonder how much "newness" there is in Rory Read idea of Embedded market, as it seems that there were people at AMD thinking on these terms well into Dirk's term.

 

[tweakboy]

The rumors should be good for team red. gl

 

[Abwx]

Models for mini itx have higher GPU frequency/perfs
than the tablet dedicated models wich will be sub 5W
for the dual core..

The most interesting part is the 25W 2Ghz quad that will
find its way in notebooks and surely bring down further
this segment prices.

 

[erunion]

Just found this about Kabini.

If this is true, I wound't be very upbeat about Kabini. It won't be able to compete with Haswell, let alone Broadwell in the bottom market, except on price. Frequencies are too low to match Trinity. Temash won't be able to match Silvermont too.

Kabini is being put in the same position as bulldozer. It's being hyped as AMD's come back chip, when we know next to nothing about it.

Sure it will be an improvement over Brazos 2.0. But small bumps in x86 and gpu performance aren't going to make it a home run. AMD hasn't shown whats changed with Kabini that will allow it to penetrate the market more than Brazos has. AMD has said its a SOC, but there's also rumors about chipsets. I'm thinking it might support an optional chipset to allow for more I/O on laptops, but on tablets it could go without it. But like I said, we don't know.

People are leaning on its GPU performance increase as its differentiator, but they must not understand just how puny its predecessors' performance was. E-350 is not even in the same league as a mere HD 4000.

 

[frozentundra123456]

Kabini is being put in the same position as bulldozer. It's being hyped as AMD's come back chip, when we know next to nothing about it.

Sure it will be an improvement over Brazos 2.0. But small bumps in x86 and gpu performance aren't going to make it a home run. AMD hasn't shown whats changed with Kabini that will allow it to penetrate the market more than Brazos has. AMD has said its a SOC, but there's also rumors about chipsets. I'm thinking it might support an optional chipset to allow for more I/O on laptops, but on tablets it could go without it. But like I said, we don't know.

People are leaning on its GPU performance increase as its differentiator, but they must not understand just how puny its predecessors' performance was. E-350 is not even in the same league as a mere HD 4000.

We have to see what the performance will be as well as the price. Problem is they are being squeezed by intel on the high end/high performance segment and by ARM on the low end/low performance. Perhaps they can find a middle ground that allows them to make a profit and also finds a good market. If the performance is at all decent, personally I would choose a low end x86 tablet over anything ARM can put out, because I really still do not like Android.

 

[pablo87]

Just found this about Kabini.



If this is true, I wound't be very upbeat about Kabini. It won't be able to compete with Haswell, let alone Broadwell in the bottom market, except on price. Frequencies are too low to match Trinity. Temash won't be able to match Silvermont too.

I don't see in Kabini all the attributes needed for tablets or notebooks, but it's the perfect chip for embedded. It is efficient, not top efficiency, but good enough efficiency. It is faster, but only good enough. Pricing should be very interesting too. Add the reusable IP part blocks that Rory Read talks all the time and we have a winner on the embedded market, much more than Brazos.

It seems that Kabini optimization for the embedded market weren't an afterthought for AMD, but that the chip was thought to be cheaper from its conception. I wonder how much "newness" there is in Rory Read idea of Embedded market, as it seems that there were people at AMD thinking on these terms well into Dirk's term.

I dunno, unless I'm missing something, Intel's current mobile quad cores, their highest end parts, made on the most sophisticated (and expensive) fabs ever are a little over 2ghz with poorer graphics, I/o that is NOT integrated (so that 17W = 23W) and start at $700 or 2x kabini's likely OPP. Haswell might be more formidable but its also going to be $$$.

While intel is no competition for Kabini, GF thru the WSA, most certainly is.

 

[beginner99]

I dunno, unless I'm missing something, Intel's current mobile quad cores, their highest end parts, made on the most sophisticated (and expensive) fabs ever are a little over 2ghz with poorer graphics, I/o that is NOT integrated (so that 17W = 23W) and start at $700 or 2x kabini's likely OPP. Haswell might be more formidable but its also going to be $$$.

While intel is no competition for Kabini, GF thru the WSA, most certainly is.

As you can see in above chart brazos graphics pretty mich sucked and was way worse than HD 4000. And Haswell mobile graphics will be a lot faster than HD 4000. So no Kabini will lose also in that area and will be demolished in CPU performance.

 

[monstercameron]

Kabini is being put in the same position as bulldozer. It's being hyped as AMD's come back chip, when we know next to nothing about it.

Sure it will be an improvement over Brazos 2.0. But small bumps in x86 and gpu performance aren't going to make it a home run. AMD hasn't shown whats changed with Kabini that will allow it to penetrate the market more than Brazos has. AMD has said its a SOC, but there's also rumors about chipsets. I'm thinking it might support an optional chipset to allow for more I/O on laptops, but on tablets it could go without it. But like I said, we don't know.

People are leaning on its GPU performance increase as its differentiator, but they must not understand just how puny its predecessors' performance was. E-350 is not even in the same league as a mere HD 4000.

I wonder how the slightly faster e2-1800 would fare compared to the e-350?

 

[quest55720]

Anyone have idea on when the mini-ITX motherboards be on market? Want to test a few at work to replace a lot of aging single core celeron's from 6-7 years ago. Right now using atom's for some of them right now to cut costs VS I3 and the fanless design.

 

[Arachnotronic]

Kabini's problem will be Bay Trail-M ("Silvermont" based SoC).

 

[AtenRa]

It won't be able to compete with Haswell, let alone Broadwell in the bottom market, except on price.

It was never indented to compete against Haswell or Broadwell, that is the job for Kaveri.

Temash won't be able to match Silvermont too.

Even a quad core bobcat at 1.7GHz would be able to compete against a dual core + HT Silvermont, Jaguar will have 10+% higher frequency + 15% IPC over Bobcat.

Not to speak about the iGPU performance, where Silvermont will stay father behind.

Another dual-core chip will operate at 1.65 GHz, and it will have 15 Watt TDP.

Dual core 1.75GHz Bobcat 2.0(E-2000) is at 18W, im expecting a Quad Core 1.7-1.9GHz Jaquar at 15W.

 

[monstercameron]

kabini x & e-series
http://www.fudzilla.com/home/item/30451-four-kabini-apus-planned-for-june-2013

 

[insertcarehere]

I find it rather hard to believe that the adreno 320 in the nexus 4 could beat the graphics in the e-350, perhaps there are some issues with drivers or something?

 

[AtenRa]

when we know next to nothing about it.

Sure it will be an improvement over Brazos 2.0. But small bumps in x86 and gpu performance aren't going to make it a home run. AMD hasn't shown whats changed with Kabini that will allow it to penetrate the market more than Brazos has.

http://www.youtube.com/watch?feature=player_embedded&v=_GXA38vFPXA#t=4048s

iGPU performance will double over Bobcat, that will make Silvermont lagging even further behind.

 

[monstercameron]

I find it rather hard to believe that the adreno 320 in the nexus 4 could beat the graphics in the e-350, perhaps there are some issues with drivers or something?

maybe the benchmark is to blame?

 

[ShintaiDK]

It was never indented to compete against Haswell or Broadwell, that is the job for Kaveri.


Even a quad core bobcat at 1.7GHz would be able to compete against a dual core + HT Silvermont, Jaguar will have 10+% higher frequency + 15% IPC over Bobcat.

Not to speak about the iGPU performance, where Silvermont will stay father behind.


Dual core 1.75GHz Bobcat 2.0(E-2000) is at 18W, im expecting a Quad Core 1.7-1.9GHz Jaquar at 15W.

Both Kabini and Kalveri will fight Haswell and Broadwell uarchs.

The Silvermont Atoms are out of AMDs reach.

 

[monstercameron]

Both Kabini and Kalveri will fight Haswell and Broadwell uarchs.

The Silvermont Atoms are out of AMDs reach.

how much info is there on the next gen atoms? and how can you draw that conclusion?

 

[ShintaiDK]

how much info is there on the next gen atoms? and how can you draw that conclusion?

There is quite a lot of information out there. And Kabini cant reach the low TDPs as Silvermont.

 

[NTMBK]

Kabini is built from the ground up to be a cheap chip. Its target isn't the i5 or i3, it's the Pentium and the Celeron- and it should do very nicely against them. Intel releases its Pentiums long after its i7s, so I doubt we will see Haswell Celerons before next year.

EDIT: But I do agree with ShintaiDK, the new Atoms are what AMD should worry about. It's a chip properly designed for the low power segment, not a cut down version of a server chip, and its been a long time coming. Hopefully Intel won't screw it up.

 

[insertcarehere]

Both Kabini and Kalveri will fight Haswell and Broadwell uarchs.

The Silvermont Atoms are out of AMDs reach.

I think the more likely scenario is that there will be both phone and tablet versions of silvermont atoms, the latter of which will compete with kabini. As I don't think Intel will like to sell their "big cores" at the lower ASPs to compete with Kabini, so it will be something like dual core silvermont on phones, quad core silvermont vs dual/quad core kabini in tablets. This is just my speculation, feel free to disagree.

 

[ALIVE]

the 4 core will be 4 real jaguar cores
or 2 modules 4 thread processor??? like bulldozer

 

[NTMBK]

the 4 core will be 4 real jaguar cores
or 2 modules 4 thread processor??? like bulldozer

4 Jaguar cores.

 

[Arachnotronic]

I love how AMD loyalists like to think that AMD, on a 28nm process, will be able to pack in as much performance in a given power envelope as Intel will on a 22nm process for a low power design.

When you have the better, smaller, and faster transistors in a power constrained environment, you win, unless your architecture is garbage.

 

[Gideon]

Yeah, hoping for Kabini to challenge low-power Haswell i5/i7s was unrealistic from the start. The do quite well though compared to lower end Ivy Bridge i3's though. There were benchmarks that showed that at 1.4 Ghz 4 core Kabini was a tad faster in Cinebench than 4 thread 2 core i3, therefore they COULD compete there, on price.

The most disappointing thing about Jaguar is Temash IMO. According to AMDs own slides, even a 2 core Temash offers no battery life improvement over Z-60 at the same frequency. And that's simply unacceptable if one looks at Z-60 power draw.

 

[zlatan]

I find it rather hard to believe that the adreno 320 in the nexus 4 could beat the graphics in the e-350, perhaps there are some issues with drivers or something?

No it's actually right in the Ice Storm test. This benchmark uses very simple shaders, so the Adreno architecture is ideal for it, because these iGPUs have very fast setup and vec4+scalar design. This is ideal for the mobile world. The HD Graphics 4000 is also win in this case, with vec4+vec4 design, and of course in the 3DMark the iGPU will always boost to 1+ GHz, because the benchmark only use the CPU for graphics commands. In a performance demanding game the HD Graphics 4000 won't run in 1+ GHz, just only 350 MHz.
Any AMD (TeraScale-GCN) and NV (G80-Kepler) architectures are different. It's logical that these are designed for PC games and not for mobile games. The E-350 is a VLIW5 system designed for heavy workloads. Take another exaple. The Mali-T600 is also designed for heavy compute, and you can see that it's not really fast in the Ice Storm benchmark.
Tha Kabini iGPU is also designed for heavy compute. It won't win in the Ice Storm test, but the performance is good enogh for any mobile games. In the other hand this will be the first ultramobile graphics that can run a compute heavy DX11 games like Dirt Showdown or GRID 2. Dirt Showdown is simply amazing in my Kabini sample. I can also turn on advanced lighting (a real GPU killer effect) and it's playable.