Kabini Rumors

[IntelUser2000]

In 3dmark vantage which is a better GPU test than 3dmark06,

06 is a better benchmark for these devices because the test itself would run at single digit frames for the later ones. Even with 3DMark06, with E-350 you are talking about 7-8 frames in the SM2.0 and 9-10 frames in the HDR/SM3.0 portion of the benchmark.

Games wouldn't run that low though, because its simply not playable. While in games and settings/resolutions that makes it run at single digit frames, the benchmark might be better, but its unrealistic. The Entry level 3DMark Vantage is kinda relevant because E-350 can get 15 fps.

On the high end, 3DMark Vantage and 06 is too good on the Intel chips while 3DMark 11 is too good for AMD chips. Reality is a mix between the two.

3DMark used to be rough indicator of average games performance, up until 3DMark06.

 

[MightyMalus]

KindaOffTopic: Anything about Temash? The thing I'm most interested about is in the tablet/turbodock/keyboard stuff.

And is it just me or is Temash/Kabini the same as Kaveri just with the cat cores? And by that, I mean about the HSA features. T&K are supposed to be getting new HSA features, something that big cores won't have till Kaveri, probably exactly the same additions that Kaveri will have.

Do you think this will continue as such? Small core becoming the "prototype" for AMD?

 

[lagokc]

KindaOffTopic: Anything about Temash? The thing I'm most interested about is in the tablet/turbodock/keyboard stuff.

And is it just me or is Temash/Kabini the same as Kaveri just with the cat cores? And by that, I mean about the HSA features. T&K are supposed to be getting new HSA features, something that big cores won't have till Kaveri, probably exactly the same additions that Kaveri will have.

Do you think this will continue as such? Small core becoming the "prototype" for AMD?

Probably not, it's probably something simpler like Jaguar is ready but Streamroller is not.

 

[MightyMalus]

Probably not, it's probably something simpler like Jaguar is ready but Streamroller is not.

Alright. Anyword on Temash chips and release dates?

 

[monstercameron]

Alright. Anyword on Temash chips and release dates?

they dont really have a public release day but are already shipping.

 

[sontin]

It should definitely be up to 1.4GHz in either. If it's 1GHz max when undocked then it's in serious trouble, that'll barely compete with Clover Trail in CPU perf, never mind Silvermont.

1GHz is the 3,4w Temash chip with Dualcore and a 75GFLOPs/s GPU.
Temash is very slow and i guess only there that AMD can claim that they have a "tablet" SoC. But from a performance perspective it's not even on the same level like A15 SoCs ala Tegra 4.

 

[wlee15]

1GHz is the 3,4w Temash chip with Dualcore and a 75GFLOPs/s GPU.
Temash is very slow and i guess only there that AMD can claim that they have a "tablet" SoC. But from a performance perspective it's not even on the same level like A15 SoCs ala Tegra 4.

There's also a 1 Ghz Quad core at 5.9W.

 

[mikk]

they dont really have a public release day but are already shipping.

What is your source?

 

[Exophase]

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

 

[SocketF]

they dont really have a public release day but are already shipping.

They only said "Q2" for Temash, that means they dont have to ship it yet.
Kabini is shipping since Q1.

Product mix, we have introduced new products, Kabini shipping in Q1 and Temash shipping in Q2, and obviously that would be a positive from a viewpoint of going with the gross margin.

http://seekingalpha.com/article/135...arnings-call-transcript?page=5&p=qanda&l=last

 

[mikk]

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

This is for Kabini not Temash.

 

[MightyMalus]

1GHz is the 3,4w Temash chip with Dualcore and a 75GFLOPs/s GPU. Temash is very slow and i guess only there that AMD can claim that they have a "tablet" SoC. But from a performance perspective it's not even on the same level like A15 SoCs ala Tegra 4.

Tegra 4 won't run my apps or games. And If I were to go ARM, which I might, I'd go SnapDragon 800!

IMO Tegra 3 and 4 don't seem so great. And I am anti-Apple.

 

[CTho9305]

On top of that, it takes a lot of work to get the best results from the tools; if you take an optimized hand design and compare it to a one-off automation experiment, you're likely going to conclude you should stick with hand design. If you do a quick and dirty hand design and compare it to an optimized automated design, you're going to conclude you should stick with automation. A fair comparison requires a large effort.

What's your opinion on the matter? Are synthesis tools up to hand-designed parts or is there still a way to go?

I've written quite a bit about that in the past. The short version is that I think hand design is the wrong choice in most situations. Here are some posts I've written when this topic came up before:
* Discussion with TuxDave (keep clicking the little arrows

next to our names to get back to the quoted post).
* fairly long post with my thoughts as of ~3 years ago
* Thread "Ex-AMD Engineer explains Bulldozer fiasco" about cmaier's Bulldozer explanation, in which I posted a few times and discussed with dmens. Another relatively long post from that thread.

Feel free to ask questions based on those old posts here (or start a new thread if the mods prefer) so we don't resurrect ancient threads. If I'm using terms anyone doesn't understand in some of those posts I can translate... I had written some of the posts more for other industry people rather than for the layperson.

Isn't this like the subpar material you use in an engineering project, but one that allows *much* lower costs or time to build? I can see where this is going, and I agree with you that synthesis will play a much bigger role in the lagging edge of the market.

But what about the bleeding edge? When you have Qualcomm, Apple or Intel R&D budget, why wouldn't you go to a mixed mode, meaning crafting by hand a few critical parts and leave simpler things (cache, crossbars) to synthesis tools.

Or did I get you wrong and do you think that somehow even bleeding edge players will rely on synthesis tools for their designs?

Even on the bleeding edge, I believe automation is valuable, since it lets you adapt to changes (to the architecture, or to the manufacturing process) more rapidly.

 

[itsmydamnation]

This is for Kabini not Temash.

they are the same chip.

 

[SocketF]

And as for your "IPC+freq." claim for PD and missing the same one for EX part, I can say look at the SR core note . It only states "greater parallelism" but this core will be much better than BD or PD are today.

Yes but "greater parallelism" is much more than "greater performance". At least you get a glimpse how they will improve the performance.

We will most definitely have IPC improvements on top of the improvements AMD cites in workloads that stress shared HW. They explicitly stated single core execution improvements in SR core presentation. So the lack of mentioning this on the roadmap does not mean it's not there.

Well I would expect that the improvements for "parallel"-performance would also affect the IPC a bit.

Same goes for EX which is drawn as the biggest improvement gen-to-gen if you look at the chart.

Where? I only see a steady picture here, AMD is also claiming a stable 10-15% improvement:

Given the rather small improvements from BD to Piledriver, I dont expect more from Excavator. Steamroller will already have the new FP with 3 instead 4 FP-pipes. Maybe EX will use then the high-density version ( even though I assume now that Steamroller will already have it), but in any case, which performance impact would that have?

It will save a bit power and die area, that's it.

 

[mikk]

they are the same chip.

If they speak for the tablet variant they refer to Temash not Kabini. So in this context Kabini! = Temash

AMD Apr 18 2013:

Product mix, we have introduced new products, Kabini shipping in Q1 and Temash shipping in Q2, and obviously that would be a positive from a viewpoint of going with the gross margin.

we have started volume shipments of Kabini in the first quarter and have a strong portfolio of high volume entry level design wins based on a substantial performance and battery life improvements.As part of our strategy to win new client form-factors we’re ramping production of Temash

= Temash is not yet shipped. That doesn't bode well for availability in the near future by the way. It takes a couple of months from shipment to notebook/tablet availability.

 

[inf64]

@ SocketF

Here . This is the slide from March 2013 presentation by AMD on HPC computing.

Here is the older slide from 2012:

 

[MightyMalus]

The slide that puts Steamroller as "better parallelism" puts Excavator as "better performance", with an 30% increase over SR.(If memory is correct...)

 

[inf64]

According to AMD SR with "greater parallelism" should be 30% Better than BD (in threaded workloads, don't know how much the ST IPC will improve-probably 10-15%).EX core will bring some changes to the FP unit for sure since AMD won't be able to compete with Haswell-E unless they have full 256bit SIMD (and FMA) pipeline capability. This implies they will have 2x256bit fmac pipes per one shared FlexFP unit. Whether they will use the new high-density libraries to save die area and use these savings to boost the FP execution capabilities is unknown.

 

[MightyMalus]

Yup, I was wrong. It was SR. But 30% better than BD, not PD?

 

[R0H1T]

Yup, I was wrong. It was SR. But 30% better than BD, not PD?

That would make sense otherwise it'd be faster than Haswell in "certain" tests

 

[SocketF]

@ SocketF
Here . This is the slide from March 2013 presentation by AMD on HPC computing.

Can you see the difference to my slide?
AMD got a new photoshop guy ;-)

Seriously, I would not calculate the performance according to a power point slide, that suddenly changed from linear to exponential without any reason.
Well there is one reason, the x-scale also changed from "Performance/Watt" to "Performance" only. So yes, you could assume a big performace jump, together with a big power-consumption jump. However, given the already "over the top" power consumption of the BD-chips, I would rule out that possibility, too.

IMO it is just the optical effect of Rory "Predator" Read joining AMD in between. The speeches and diagrams got colorful and more aggressive, but I doubt that the chips changed, too.

AMD won't be able to compete with Haswell-E unless they have full 256bit SIMD (and FMA) pipeline capability. This implies they will have 2x256bit fmac pipes per one shared FlexFP unit.

AMD won't be able to compete with Haswell-E unless they'll use a 20nm finfet process.
Does that mean that EX will use a 20nm finfet process? Unfortunately no

2x256 fmacs would be indeed nice, but then you'll also need a doubled L1 cache interface. It's 2x128b now, you'll need 2x256b then and the power consumption would rise even more. I don't see enough room for a doubled FPU, the module size will already increase a lot from all the "parallel" improvements.

Maybe at 20nm, but I doubt that EX will use already 20nm.

I can see your enthusiastic expectations, you cannot rule them out, yes, but think about AMD's last track records... what was expected from Bulldozer and what did we get? Lower your expectations. Then you wont encounter frustrating surprises, there is only the chance of pleasant ones ;-)

 

[Idontcare]

IMO it is just the optical effect of Rory "Predator" Read joining AMD in between. The speeches and diagrams got colorful and more aggressive, but I doubt that the chips changed, too.

Rory really did a 180° on that whole predator thing once he got a full download of just how far up the creek without a paddle AMD had been sent by Meyers and Ruiz.

 

[Vesku]

Rory really did a 180° on that whole predator thing once he got a full download of just how far up the creek without a paddle AMD had been sent by Meyers and Ruiz.

Rory: "We'll be a predator"
*Hey what's this WSA thing*
Rory: "Cuts across the board, retrench for 2014"

Will be interesting if Kaveri has no single module parts. The WSA + shrinking traditional market growth should translate into an incentive to make lots of medium to large die chips and be selective on the harvesting.

Would also expect them to cross pollinate the larger kabini quad style chips over to GF.

 

[inf64]

@ SocketF

I still think EX core will be a BIG performance jump vs SR. Similar to BD -> SR jump, if not bigger. Also we have no clue on which process the EX core will be made.