AMD CPU speculation... and expert conjecture

[esrever]

Im pretty sure GCN can use memory bandwidth better which is something AMD needs in the APU.

 

[Cazalan]

Richland uses the same 32nm SOI process than Trinity does. I'm interested to see which metrics increase by 20-40%. If they've implemented such performance gains without taking a significant hit in power or or area I'd be very impressed.

The 20-40% numbers are solely based on 3DMark 11.

PCMark 7 was 10-12%.

 

[Blandge]

The 20-40% numbers are solely based on 3DMark 11.

PCMark 7 was 10-12%.

Thanks for the clarification. These numbers are actually much more reasonable.

Early on, VLIW5 vs VLIW4 saw VLIW4 gaining 7% right away.
Since that time, it has been much improved, thru drivers.

If we guess a lil process/speed/TDP improvements, even VLIW4 and a small speed increase, which can have a decent effect on igpu, then theres not too much room for dust

I agree. Somebody on the Anandtech forum said he email AMD directly and received the following response:

I understand the concern; we didn't move to 28nm and the architecture didn't change, so where is the uptick coming from? Performance gains have been achieved through a combination of increased frequency, improvements in the bi-directional power management between the CPU and GPU, as well as more precise thermal management that allows for running at higher boost speeds more often.

Translation: Trinity was rushed and there was a ton of power management they could have added (and have added in Richland). I think I can wrap my head around this. Even so I'll be impressed if they do indeed deliever on these claims. 20-40% without any significant microarchitectural change is no small feat.

 

[cgner]

5% per clock improvement, 5% higher clock. Simple!

 

[noob2222]

It appears as though it will be something like A10-5900K according to the link you provided (below). You should considering reading the articles you reference more carefully before posting about them.

These new Richland-based 5000-series APUs are already shipping to OEMs for use in notebooks with desktop variants possibly following at a later date.

Maybe I should have quoted the entire article so you wouldn't have to read it yourself. But since you pointed that out, apparently kabini is actually Kab ini.

Thing is there is conflicting information on almost every article on richland, none of them say the exact same thing, I should just spend my time listing every possible typo or conflict in possible speculation.

But I can't do that since I don't read apparently.

 

[sarinaide]

Vishera was the exact same process as Zambezi and it managed 7 percentiles to 15 percentiles in general performance yields, why is it not foreseable that Richland will not do the same notably with a stronger iGPU part? more mature process, core level changes and a stronger iGPU. considering the Trinity on a similarish iGPU to llano managed around 40% improvement why is 20-40% not possible hmmm?

They cynicism is baffling.

 

[JAYDEEJOHN]

I think some see Vishera as too power hungry, yet it replaces a lessor chip in the same power envelope, and this too can be done down the line to lessor chips using the newer power gating, and taking advantage of the better/improved process tech.
It only makes sense, as AMD has already done this much, and adding an improved tech in their igpus, plus again better clocks thru the better, more mature process.
Added together, those claims, at least in what AMD is so far showing is very doable, tho how this all transistions outside of synthetic benches is up for grabs

 

[JAYDEEJOHN]

Since we are speculating here, I will go this far also.
Going strictly to GCN2 at this point in time isnt ideal for AMD.
the transition to a SoC solution, eliminating kernels is obviously the next big step, and a slower transition is needed, so they just dont sit on the same tech till we are there.
My 2c

 

[sarinaide]

Ditto ^

 

[truegenius]

magic pixy dust.

few months ago i read that AMD's new drivers shows upto 20% performance increase for gcn based cards
so maybe that is the source of that magical dust/dragon ball wish without improving much on any thing (maybe gcn it is without hampering power envoloupe :/ )

 

[Ranth]

This is weird.... If it's true..

https://www.youtube.com/watch?v=eu8Sekdb-IE

Does somebody have a 8350 and a 3570k to test it? Cause I don't get it.

 

[noob2222]

here is the problem with richland being 20% on drivers. Drivers update trinity as well.

Media would have a ball throwing AMD under the bus if trinity increased by 20% and richland by 20% over trinity when only used with old drivers so that richland to trinity with the same drivers improved 0%.

 

[noob2222]

This is weird.... If it's true..

https://www.youtube.com/watch?v=eu8Sekdb-IE

Does somebody have a 8350 and a 3570k to test it? Cause I don't get it.

rofl at 4:18-4:25

 

[de5_Roy]

...

http://www.xbitlabs.com/cms/include...ht=308&cache=1&quality=99&aspect=0&format=jpg

I agree. Somebody on the Anandtech forum said he email AMD directly and received the following response:

I understand the concern; we didn't move to 28nm and the architecture didn't change, so where is the uptick coming from? Performance gains have been achieved through a combination of increased frequency, improvements in the bi-directional power management between the CPU and GPU, as well as more precise thermal management that allows for running at higher boost speeds more often.

Translation: Trinity was rushed and there was a ton of power management they could have added (and have added in Richland). I think I can wrap my head around this. Even so I'll be impressed if they do indeed deliever on these claims. 20-40% without any significant microarchitectural change is no small feat.

looks like richland is mainly going for ultrabook type notebooks. amd keeps talking about thermal management, seems like richland will have better turbo at least. i'm all for better igpu, it's already the best.
i don't think trinity was rushed. quite a few people credit amd for 'rushing out' architectural improvements in such a 'short notice' with such 'limited resources and budget'. these take years to develop and current base architecture is over 3 years old. s/a claimed that trinity was supposed to come out in 2011. i am inclined to belive them after seeing how bd was supposed to compete with nehalem and how glofo slooowly, [strike]fumbled[/strike] transitioned to 32nm process. may be that's why these new launches seem crowded to me. imagine the cpu/apu with longer gaps.
i remember amd claiming 30% performance (or was it perf per watt) improvement with high density cell library technology (supposed to be incorporated in excavator) without fabricating on a smaller process. if they're claiming 40% max. improvement, then they have found something that trumps or competes well with hdcl. may be cyclos' rcm is better implemented this time.

here is the problem with richland being 20% on drivers. Drivers update trinity as well.

Media would have a ball throwing AMD under the bus if trinity increased by 20% and richland by 20% over trinity when only used with old drivers so that richland to trinity with the same drivers improved 0%.

heh. may be it's a trickle down effect (firmware locking, among other trickeries) so that richland gets the majority of improvement.

you can still count on biased (there are plenty of them) to hype richland as 'the only apu one will ever need' (until amd launches the next one) or instantly drop golden awards.

edit: when i read about thermal management improvement and better turbo, i thought of rcm as the reviews earlier said that rcm was not implemented significantly. add mature process, other tweaks etc.

lazy edit:

'cons n' 'fers!
http://semiaccurate.com/2013/01/10/global-foundries-shows-off-20nm-and-14nm-wafers/
i can see the finfets in 3d!
http://semiaccurate.com/2013/01/09/amd-shows-off-kabini-and-hondo-silicon/

was there any doubt that tsmc would make the lp and ulp apus?
http://www.fudzilla.com/home/item/30102-amd-28nm-chips-are-still-tsmc

Crucial Demos DDR4 Memory Modules, Plans to Ship in Late 2013.
http://www.xbitlabs.com/news/memory/display/20130110183846_Crucial_Demos_DDR4_Memory_Modules_Plans_to_Ship_in_Late_2013.html

8 core fx just got served:
Samsung Unleashes Exynos 5 Octa: World’s First Eight-Core Mobile Chip.
http://www.xbitlabs.com/news/mobile/display/20130110144741_Samsung_Unleashes_Exynos_5_Octa_World_s_First_Eight_Core_Mobile_Chip.html
can fx8350 output to 4k? ofc not, it doesn't even have an igpu. Burn!
i see this apu outsell 8 core fc by a factor of FouR.four to one...

Process tweaks and power management define AMD's Richland APU
http://techreport.com/news/24190/process-tweaks-and-power-management-define-amd-richland-apu
New details, early benchmark data revealed for 15W Kabini APU
http://techreport.com/news/24186/new-details-early-benchmark-data-revealed-for-15w-kabini-apu

 

[JAYDEEJOHN]

Now, Kabini will likely compete against cheaper—and lower-power—processors than the i3-3217U and i5-3317U. Kabini's 15W TDP applies to the whole system-on-a-chip, including the integrated I/O, while Intel's 17W Ivy Bridge processors require a separate platform hub. The HM67 Express PCH in our ultrabook has a 4.1W TDP, according to Intel's website, so the Intel solutions' total power requirements add up to 21.1W or so.

http://techreport.com/news/24186/new-details-early-benchmark-data-revealed-for-15w-kabini-apu


So it appears Mr Taylors show at CES was spot on, and SoC for AMD is there.
Like he said, what here (at CES) requires large powerful x86 cpus?

 

[esrever]

This is weird.... If it's true..

https://www.youtube.com/watch?v=eu8Sekdb-IE

Does somebody have a 8350 and a 3570k to test it? Cause I don't get it.

Im not sure if he is a reliable source but would be somewhat interesting to test. The video is mostly show a lack of professionalism due to the wrong numbers on the screen. He does test games that don't really stress cpu and the numbers are within a small margin of error. The exception is trine 2 but I really have not seen any other benchmarks for that game.

 

[noob2222]

heh. may be it's a trickle down effect (firmware locking, among other trickeries) so that richland gets the majority of improvement.

you can still count on biased (there are plenty of them) to hype richland as 'the only apu one will ever need' (until amd launches the next one) or instantly drop golden awards.

The only problem with something like that is people are out for blood against AMD. The last thing AMD needs to do is give them a reason. Even if there was such a "lockout driver" people would find a way to make it work and post the numbers. Intel fanboys would have a heyday spreading the information like wildfires. It would even piss off AMD fanboys.

 

[viridiancrystal]

This is weird.... If it's true..

https://www.youtube.com/watch?v=eu8Sekdb-IE

Does somebody have a 8350 and a 3570k to test it? Cause I don't get it.

That was interesting to watch. I am unsure of it, because of he being only one to show the 8350 being clearly superior. Curious, though.

 

[-Fran-]

He's over blowing it a little with his "amazement", but I can get where he's coming from.

And to be fair, in all sites, the 8350 was actually good coping with the 3570k in most gaming stuff and excelled in multi tasking. That's something we've known already.

Thing is when you OC the Intel K's you get a lot more per Mhz than on the 8350.

I think we've been over this a couple of times already, but oh well.

Cheers!

 

[mayankleoboy1]

And to be fair, in all sites, the 8350 was actually good coping with the 3570k in most gaming stuff and excelled in multi tasking. That's something we've known already.

And yet when asked about build advice, you and I (and most of us) will never recommend a BD/PD setup. If budget constrained, we would rather suggest a lower i5 or high i3 , rather than a 4/6/8 core PD.

Reason ? AMD is good, but Intel does everything so much moar better. At everything. Intel does not fear Floating-Point calculations. Pus, it will always use lesser power.

Contrary to popular beliefs, single threaded is still very important for most workloads.

 

[noob2222]

and just like he said, insult anyone who recommends or uses AMD, even if it actually fits the needs of the user.

At everything.

especially the bandwagon effect.

 

[mayankleoboy1]

especially the bandwagon effect.

Sometimes the road less traveled is less traveled for a reason, conspiracy theories aside.

 

[noob2222]

Too many fanboys pointing and laughing? People have a tendecy to succomb to peer pressure, and are afraid of the alternative.

other people are too stubborn and are labeled a conspirator, hated by those who can't make them listen to their bandwagon of advertising.

 

[mayankleoboy1]

not much merit in being a hipster, buying and promoting non mainstream non performant stuff , just because it has a lesser market share.
root for the underdog, by all means, but not in the face of clear evidence. Otherwise, just a Fanboi.

 

[mayankleoboy1]

and just like he said, insult anyone who recommends or uses AMD, even if it actually fits the needs of the user.

Dont fkucing put words in my mouth. I never said that, nor said words to that effect.
But you dont read what you link......... :lol: