Richland & Kabini rumours

[Centauri]

Five Kabini APUs planned for June 2013

Setting aside my fanboydom; I'm very impressed by how aggressive AMD is becoming with their APU aims. Last year, saying that APUs might be AMD's salvation would have been said fairly tongue in cheek, but it's certainly starting to look very likely.

 

[bgt]

This board also apparently needs more voltage for the same frequency as other boards. Some reported that they couldn't overclock at all on the Asrock board, saw BSODs or funky behaviour at stock speeds or had to apply ridiculous voltages to get to 4.4 Ghz while other boards can hit 4.6 Ghz without much trouble.So yea, that board can either be weighted against AMD (8350 out of spec) or against MSI (shoddy board). Or perhaps against both.

My CPU can do 4.6Ghz (with CPU overdrive+no change in bios)on my MB so thats not the problem and it is a very stable/ultra smooth CPU+MB combo.

 

[bgt]

Setting aside my fanboydom; I'm very impressed by how aggressive AMD is becoming with their APU aims. Last year, saying that APUs might be AMD's salvation would have been said fairly tongue in cheek, but it's certainly starting to look very likely.

Really nice and fanboydom because you have AMD and its about AMD.......I have both brands and I am always very pleased with AMD stuff. Their graphics are always awsome be it IGP or GPU. If nobody would defend AMD in any way...........well we have Intel monopoly. Nobody wants that.

 

[Puppies04]

Setting aside my fanboydom; I'm very impressed by how aggressive AMD......

This made me laugh.

 

[Vesku]

It is telling that AMD doesn't give guidance on max safe temperature nor really anything informative about maximums with the FX series. However, MSI has had more VRM issues than most other enthusiast brands when it comes to AMD motherboards. Their AM3+ lineup is a great improvement over most of the AM3 versions but in terms of the 970A-G46 it seems like it was just a bit underestimating of the demands of the full 4 module chips compared to other vendors boards in the same category. Overclock.net has a pretty nice thread regarding AMD motherboards and their VRMs.

http://www.overclock.net/t/946407/amd-motherboards-vrm-info-database

IMO the issues with the FX series show why Kabini is important. FX power draw would be more in check if they weren't pushing the MHz envelope on their inferior to Intel process node so that they aren't completely embarrassed by their preceding Phenom II line. Similar to ARM, AMD is less disadvantaged when operating more on cost and features rather than cutting edge performance.

 

[bgt]

For the record , the reflected power is consumed by the electricity source and wont add in your electricity bill , that s why electricity providers are eager to impose minimum PF since they are the ones
that actualy pay for thoses reflected powers.

Just got a new YouLess LS110 system to measure power consumption.
Have to wait till all my power consumers are off....in the middle of the night so I can measure with steady values.
http://www.domoticaforum.eu/viewtopic.php?t=7217&p=54319

 

[Centauri]

This made me laugh.

Incomplete sentences make you laugh?

 

[Centauri]

So what is the verdict exactly; does Richland pack GCN or not?

 

[AtenRa]

The latest AMD RoadMap shows only kaveri with GCN graphics.

 

[Centauri]

Thanks.

And do we know any more about AMD's intentions for FM2? Is there still the expectation that Kaveri will be FM2?

 

[SocketF]

And do we know any more about AMD's intentions for FM2? Is there still the expectation that Kaveri will be FM2?

There were some FM3 rumors based on some Linux-BIOS files recently.

 

[Olikan]

found some really nice slides about jaguar...

with nice i mean, some really high level stuffs...

http://pt.scribd.com/doc/126362330/ISSCC2013-Final-v5-pdf

IDC, please translate

edit:

Posts: 1,337

 

[Phynaz]

found some really nice slides about jaguar...

Awesome, bookmarked under "Cool Stuff" to read later.

Thanks!

 

[AtenRa]

nice thx

 

[Idontcare]

found some really nice slides about jaguar...

with nice i mean, some really high level stuffs...

http://pt.scribd.com/doc/126362330/ISSCC2013-Final-v5-pdf

IDC, please translate

edit:

Posts: 1,337

Congrats on the LEET poscount

Looking through the slides, are my eyes deceiving me or do they still not have actual physical Jaguar in hand to benchmark yet?

•Estimated typical IPC improvement over “Bobcat”: >15%*

* Estimates based on internal AMD modeling using benchmark simulations.This information is preliminary and subject to change without notice.

Estimates based on nothing better than simulated benchmarks?

• Achieved silicon frequency >1.85Ghz*

* Estimates based on internal AMD modeling using benchmark simulations. This information is preliminary and subject to change without notice.

Claim to have achieved silicon clockspeeds only to then put it in the fine-print that it is just an estimate based on internal AMD modeling?

From what I read in the presentation I don't see anything new being divulged that hasn't already been made public.

What is personally disconcerting is that there also doesn't seem to be anything new, everything that relates to performance, power, etc are all caveated as being based on modeling and simulations.

If this thing ever moves out of powerpoint and into the fab then maybe we'll find out how good AMD's modeling and simulations are. For now it looks like all we get to go on are the powerpoint benchmarks.

 

[CTho9305]

found some really nice slides about jaguar...

with nice i mean, some really high level stuffs...

http://pt.scribd.com/doc/126362330/ISSCC2013-Final-v5-pdf

IDC, please translate

edit:

Posts: 1,337

Awesome, thanks... there's a lot of interesting stuff in there.

If this thing ever moves out of powerpoint and into the fab then maybe we'll find out how good AMD's modeling and simulations are. For now it looks like all we get to go on are the powerpoint benchmarks.

In at least some cases, it looks like they're just using the same extra-broad disclaimer for the "preliminary and subject to change without voltage" part (for example, the "measured silicon shows...leakage <10mW" statement).

 

[Idontcare]

In at least some cases, it looks like they're just using the same extra-broad disclaimer for the "preliminary and subject to change without voltage" part (for example, the "measured silicon shows...leakage <10mW" statement).

That's what makes it all the more fishy. The measured silicon either came in at <10mW or it didn't. There is no "we measured it, but the results are subject to change"...what is that? That isn't a measurement, that is a guess. Guesses are subject to change.

 

[CTho9305]

That's what makes it all the more fishy. The measured silicon either came in at <10mW or it didn't. There is no "we measured it, but the results are subject to change"...what is that? That isn't a measurement, that is a guess. Guesses are subject to change.

My reading was, "some of the silicon we have in hand today leaks <10mW in sleep mode; for any of a large number of reasons, the end product may not behave that way". For example, before volume production they might choose to skew the manufacturing process towards faster (leakier) transistors to improve performance, or they might encounter a bug that keeps the part from sleeping as desired. It's also possible that the number is more of a raw-engineering kind of number ("hey, can you go measure leakage on a few parts and send me the numbers") rather than a marketing-approved number ("get a statistically valid sample and measure it under xyz operating conditions, and give me mean + std deviation so we can make a commitment to customers") and they don't want a shareholder lawsuit if the final number doesn't end up exactly matching the quick measurements. I guess my take is that they're probably giving a sincere assessment of what engineering thinks the silicon does, but they have to hedge.

 

[AtenRa]

The product is under NDA and they CANNOT disclose any info/data, so they use what was publicly announced last summer.

Temash silicon(Jaguar) has been already manufactured.

AMD to Showcase Fusion “Temash”-Based Tablets, Convertibles with Turbo Dock Technology.

Advanced Micro Devices on Wednesday said it would demonstrate tablets and convertibles based on its highly-integrated Fusion-series code-named Temash accelerated processing units at Mobile World Congress 2013 next week.

 

[Phynaz]

Looking through the slides, are my eyes deceiving me or do they still not have actual physical Jaguar in hand to benchmark yet?

Considering that Sony just announced the PlayStation 4 is using Jaguar cores and is shipping later this year, I can't imagine AMD doesn't have functioning silicon yet.

Or Sony just made very, very big mistake.

 

[ShintaiDK]

Considering that Sony just announced the PlayStation 4 is using Jaguar cores and is shipping later this year, I can't imagine AMD doesn't have functioning silicon yet.

Or Sony just made very, very big mistake.

Sony couldnt even show the console itself. Or anything running on it.

 

[inf64]

Of course they have Jaguar running. They have it running for 1+ year since that is roughly the time from tape out to launch(if no problems occurred). But since it's not officially launched AMD has to format the claims in that particular way,they have done it before.

 

[Asterox]

Here's a little comparison from "sweclockers.com" or one test, Temash APU A6-1450/6W TDP in Cinebech 11.5 Multithread CPU test i just added the CPU frequency for this other processors.

Obviously this site has information on the number of GCN graphics cores used on Quad Core Temash APU A6-1450. When everything is taken into account, 128 GCN Cores looks very realistic and expected.

It should be noted that the Core i3 2367M (Sandy Bridge) and Core i3 3217U (Ivy Bridge) have a 17W TDP.All three of these Intel processors are 2 Cores /4 Threads CPU.

Atom Z2760 has a low 2W TDP but there is nothing strange, his crappy PowerVR DX10 graphics core is very weak and outdated.

Speaking of Trinity APU Mobile A10-4600M has 35W TDP.


It would be very interesting to see, A6-1450 at 1.8 GHz CPU freguency vs Ivy Bridge i3 3217U/1.8ghz in Cinebench 11.5 Multithread CPU test. :biggrin:

http://translate.google.com/translate?hl=en&sl=sv&tl=en&u=http%3A%2F%2Fwww.sweclockers.com%2Fnyhet%2F16597-amd-temash-specifikationer-och-prestanda

 

[VirtualLarry]

Sony couldnt even show the console itself. Or anything running on it.

They showed at least one running game on the PS4 conference broadcast.

 

[inf64]

In C11.5 JG core has 31% higher ST score versus Bobcat. That's pretty amazing perf. jump. Also it has ~2x the core count and 1.4x higher clock when docked. So maximum difference when it is docked ,in ST performance is 1.15*1.4=1.61. I picked 15% for IPC since that is what AMD cites although the difference as demonstrated by C11.5 can be quite larger(>30% and in FP/SSE dominated code even more). MT performance when not docked should be at minimum 2.3x higher and at maximum(very heavy SSE stuff,unrealistic usage model for tablet) around 4x. When docked MT performance can reach up to 5.6x (unrealistic FP heavy scenario).

GPU side received a massive boost too. 128/80=1.6 or 60% more SPs, each being around 20% faster than VLIW5 ones in Ontario. Also clock has been bumped by 10% when not docked and ~80% when docked. So when it is not docked its GPU is around ~2.1x faster than Ontario's and when it is docked the difference climbs to massive 3.7x. Pretty amazing performance jumps in every aspect of the SOC,especially given the TDP limits they are facing.