AMD CPU speculation... and expert conjecture

[$hawn]

mobile i7's igpu is a joke in front of a10's
right now, amd is aiming to undercut intel in the mid to low end laptop with the mobile apus. they're trying to do the same with low power apus against atom and haswell.

You seem to have taken my statement out of context
I was just saying how AMD needs a pure CPU offering in the mobile segment, because the A10's have comparatively much weaker CPU speed compared to Intel offerings at similar price points. The GPU has been allocated a significantly huge amount of the 35W thermal budget, and that's forcing AMD to keep the CPU clocks quite low on the flagship A10.
Since almost all OEM's bundle lappies above the $650 range with discrete graphics, why not just make a mobile CPU version of the FX4300?

 

[gamerk316]

I wonder who is feeding these unrealistic performance improvements of Richland to the press? It has a sinister intent because its raising the hopes of people, and if AMD fails to live up to that, people are going to trash AMD again.

Noob222, I sense Intels money behind this.

Please don't feed him any conspiracy theories, even in jest.

 

[de5_Roy]

You seem to have taken my statement out of context
I was just saying how AMD needs a pure CPU offering in the mobile segment, because the A10's have comparatively much weaker CPU speed compared to Intel offerings at similar price points. The GPU has been allocated a significantly huge amount of the 35W thermal budget, and that's forcing AMD to keep the CPU clocks quite low on the flagship A10.
Since almost all OEM's bundle lappies above the $650 range with discrete graphics, why not just make a mobile CPU version of the FX4300?

sorry if i did that.
afaik, intel doesn't have pure cpus in the mainstream laptops anymore.. since sandy bridge or may be arrendale cpus.
a10 apus are similar to quad core fx, they just trade l3 cache and high clocks for igpu (more powerful than a lot of low cost mobile discreet gpus), cost and battery life.
amd is mostly aiming the apus at general public who use laptops for personal use and entertainment e.g. watching movies and playing games. a strong cpu would only make a better number cruncher, it won't contribute much in mobile gaming. also, bd and pd are known to be power hungry, so lower clocks and powerful igpu is the best way for amd.
intel has more powerful cpu to offer but they have weaker igpu. otoh, amd has more powerful igpu to offer but they have weaker cpu. both are advertising their strengths and both are trying to improve their weak points. they're both moving towards a fully integrated soc design. a pure cpu doesn't fit into that.

 

[esrever]

richland is trinity with better clock gating and better turbo efficiency. Which is what I would have thought when I read the kaveri delay. I would be amazed if AMD got a >20% performance boost.

 

[Cazalan]

The wonders of TomsNews . Richland is 28nm ? It has HD8000 graphics ?
Any fukcing source ? no.

AMDs naming schemes are convoluted.

The slides show Radeon Cores 2.0, which is the same as on Trinity, which is VLIW4 (Cayman style HD 6950/HD 6970).

Maybe they added a few cores or were able to turn on more cores due to yields/power tweaks. With just a couple benchmarks it could be purely driver enhancements.

 

[griptwister]

@mayan, LOL! Be nice now...

 

[JAYDEEJOHN]

First off, why so serious?
Secondly, to hopefully answer a few earlier questions
http://forum.beyond3d.com/showpost.php?p=1692333&postcount=81

 

[sarinaide]

It doesn't cease to amaze that everyone finds problems in everything AMD does yet these intellectual thinktanks are not signing up to salvage AMD from this so called "plight".

As before Vishera/Trinity made tremedous leaps over past generations suggesting AMD are heading the right direction, that was done on the same die process as Zambezi, in most practical usage the FX 8350 thoughput is roughly 10-20% faster than the FX 8150 most notably in its IMC which is 50% faster which means somewhere around a i7 930-950 level, IPC in general usage is also discernably faster than Zambezi and Thuban, then of course there is the power usage, achieving 4Ghz clocks at 60+w less was a pretty good effort considering the mess they were in.

Right now Steamroller is purely specaltive but again metal and core level improvements will see even more drastic improvement, possibly a chip competing in the SB/IB i5/i7 legitimate performance or thereabouts. As for Kaviri vs Trinity/Richland, some suggesting iGPU performance gains over 50% on GCN technology along with SR core architecture so its still speculative.

 

[mayankleoboy1]

GCN should bring huge benefits. Specially in compute tasks, which is one of the backbones of the whole HSA thingamjig.

 

[de5_Roy]

It doesn't cease to amaze that everyone finds problems in everything AMD does yet these intellectual thinktanks are not signing up to salvage AMD from this so called "plight".

it's not amazing that misinformation spreads every time you pop up. okay.. i am calling it misinformation 'cuz what i've seen and read doesn't match claims made in your post. ofc, if i turn out to be wrong i'll change my standing.

As before Vishera/Trinity made tremedous leaps over past generations suggesting AMD are heading the right direction, that was done on the same die process as Zambezi,

single digit to low double digit (15%, benchmark dependent) performance increase ain't 'tremedous leaps'.... may be okay for amd as they are now... i guess it depends on who you ask.

in most practical usage the FX 8350 thoughput is roughly 10-20% faster than the FX 8150 most notably in its IMC which is 50% faster which means somewhere around a i7 930-950 level,

i'd like to see some evidence of this. from what i've seen, vishera's memory bandwidth with ddr3 1600 ram is slightly faster than i7 9xx with ddr3 1066 ram. with ddr3 1866 it's even faster than i7 9xx with ddr3 1066 ram.
from here:
http://www.tomshardware.com/reviews/fx-8350-vishera-review,3328-7.html
and here:
http://media.bestofmicro.com/0/O/364200/original/res_syn_sandra_memory_bandwith.png
(http://www.tomshardware.com/reviews/cpu-performance-comparison,3370-12.html)
the articles used different versions of sandra though the numbers are similar... may be i got the percentage calculations wrong...
my observations:
■llano (mobile athlon core) actually gained the largest imc improvement over phenom ii (and among amd cpus) in the 86 cpu roundup benches.
■from the fx8350 review, pd's memory bw was actually slightly slower than bd with the same ddr3 1600 ram. same thing happened in the 86 cpu roundup article, fx 8170 had the highest memory bandwidth with ddr3 1866 ram among amd cpus.
■fx has strong ability for cpu based encryption tasks.

IPC in general usage is also discernably faster than Zambezi and Thuban,
then of course there is the power usage, achieving 4Ghz clocks at 60+w less was a pretty good effort considering the mess they were in.

60w+ less? than zambezi(fx8150) (and thuban(1100t) as well)? u crazy bro?
or are you talking about the 'other' fx 8300 cpu, the one with 95w tdp, same clock as 125w fx8350? that's 30w less, not 60w. that one is overpriced though. still, one of the best cpu offering from amd right now.

pd's attractions are great launch price, modestly competitive performance to the i5 and much saner (imo virtually non-existent) marketing.

Right now Steamroller is purely specaltive but again metal and core level improvements will see even more drastic improvement, possibly a chip competing in the SB/IB i5/i7 legitimate performance or thereabouts. As for Kaviri vs Trinity/Richland, some suggesting iGPU performance gains over 50% on GCN technology along with SR core architecture so its still speculative.

i too, am eagerly looking forward to what steamroller (or whatever it ends up as) brings to the competition. imo right now, amd it right to bet on the low power apus, no offence to amd's higher perf desktop cpus (that got outsold by the apus 10 to 1 ).

 

[noob2222]

@ Noob22 : You seriously take Wccftech seriously? I thought you'd be smarter than this. Nowhere on the net is anyone saying Richland or Kaveri to be SR based. Only Wccftech claims it, based on a 'leaked roadmap'.

And you think the Kitguru comparison is worth citing, which is based on 2 unknown quantities, the performance improvements of which is based on half-rumours and leaked slides and marketing ?

Quoting the Kitguru article :

These graphs show what would happen if AMD manages 40% faster in-game performance than its A10-5800k and if Intel manages to sweep past the Ivy Bridge HD4000 graphics results by 100%. Remember, these are their claims, not ours, and we’ve taken them at face value.

But you dont read what you link to. Because its Kitgurus assumption , you say ".... A lot of conflicting results, but this points to richland having GCN 2.0 radeon 8000 cores. " Hey Noob, do you even parse Text, or just look at pretty graphs?

As usual anandtech's bs is filled with AMD hate. Richland is trinity with software update?

So its OK to speculate wildly, but its hate by AT to report what AMD itself said in its presentation, no more, no less. But of course you like wild speculation, because speculation cant be measured. Real data can be. So you conveniently disregard all data as biased.

don't mention the 20-40% boost in graphics performance.

READ the AT article. It says so. But you dont have the habit of reading what you link or 'quote'.

looks like the software update is for facial recognition and hand gestures.

READ the AT article.AT says so. But you would have to read the article to get that. Which you dont.

How about it gets properly benched first? Or are you advocating review sites do the same thing they did before BD launched, raise expectations to unrealistic levels, and hurt AMD when they miss expectations (even if the chip itself is good)?

But all benches are already super Pro-intel, because Intel has paid money to all benchmark developers, and all benchmark developers use Intel C Compiler. So all benchmarks are worthless.
Right Noob ?

good to see AMD taking smaller steps of testing instead of jumping both feet into a bulldozed mudhole and allowing marketing to run stupid campaigns.

But BD was the best CPU ever made. Its just that all these damned benchmarks and applications are soo pro intel. AMD's 32nm is so much better than Intels 22nm. And Intels 14nm is not even possible.
And i suspect that the companies that make power sockets are also bribed by Intel to use more power when it detects an AMD CPU/mobo is plugged in.

dang dude, did someone piss in your wheaties or what? ROFL is all I have to say about your ranting here.

1) Read the topic: AMD Steamroller speculation .... and expert conjecture
apparently you missed that speculation part. As for Wcct, they just parrot other sites, I didn't feel like trying to lean chineese to try and figure out where they got the info since they didn't post it. apparently some of it came from fudzilla ... 4 days prior.

http://www.fudzilla.com/home/item/29986-richland-successor-in-2014-is-kaveri

oh, hey, lookie there, another speculation that kaveri is STEAMROLLER CORES.

Nowhere on the net is anyone saying Richland or Kaveri to be SR based. Only Wccftech claims it, based on a 'leaked roadmap'.

yep, not a single site anywhere.

As for your crying about kitguru why don't you read the article. IF marketing is accurate, Intel's GT3 likely won't catch richland, contrary to popular belief that Intel will own AMD's graphics in 2013. Again, that ENTIRE ARTICLE IS BASED ON SPECULATION based on Intel and AMD's marketing.

But I guess you can only cry about it not being hard facts, but thats not what this is about is it?

As for the conflicting reports on what gpu is actually In richland...

http://www.fudzilla.com/home/item/27782-amd-next-gen-named-radeon-hd-8000
http://www.bit-tech.net/news/hardware/2013/01/08/amd-ces-2013/1
http://technewspedia.com/amd-details-emerge-of-richland-the-successor-to-trinity-apu/

. On the graphics side, the solution will be integrated into the apu architecture of next-generation Radeon HD 8000 , with support for DirectX 11 API, 128 cores for dual-core models of Richland and 384 cores to quad-core versions.

Then you have the conflicting speculation.

http://www.tomshardware.com/news/amd-richland-28nm-a10-6800k,20272.html
http://www.anandtech.com/show/6567/amd-ces-2013-press-event-live-blog

From Anandtech's AMD live blog : http://www.anandtech.com/show/6567/amd-ces-2013-press-event-live-blog

07:34PM EST - Richland looks like Trinity with a software update

07:33PM EST - Kaveri will be 2H2013 with GCN support

So Richland is prolly not GCN. So how is AMD claiming these 20-30% increase ? Clock increase, due to better process tech ?
Also, no word about the CPU core in Kaveri .

Who is going to buy Richland, if Kaveri is just round the corner ? Who will buy Richland, since Trinity landed a few months ago?
FU, AMD managemnent.

ya, so you read one line on that AT picture article and deemed it to be the same chip as trinity, and your trying to call me out?

I guess that its bad to speculate that AMD's 20-40% increase comes from GCN 2.0 radeon 8000 cores, but good just to rant about it, or maybe you thought the 20-40% were going to come from the PD core.

And sorry I didn't read much of the AT article, because I doublt all amd said in 30 minutes worth of presentation was:

07:38PM EST - And of course the 11.6-inch tablet using AMD's Z60 APU
07:37PM EST - Both systems use standard voltage A10 parts
07:37PM EST - 15.6-inch and 14-inch AMD A10 notebooks from Vizio
07:37PM EST - All of the Vizio PCs are Microsoft Signature, no bloatware, it's insane that this is a feature
07:36PM EST - 24-inch Vizio All-in-One running AMD A10
07:34PM EST - Now talking about AMD's partnership with Vizio
07:34PM EST - Richland looks like Trinity with a software update
07:33PM EST - Kaveri will be 2H2013 with GCN support
07:33PM EST - Temash demos: 1) showing better GPU performance than Clover Trail, 2) showing 1080p gaming performance
07:30PM EST - Temash: dual/quad-core, less than 5W, goes into fanless designs - full Windows 8 compatibility
07:29PM EST - CPU + GPU performance gives it an advantage over the i3, obviously non GPU accelerated workloads won't look as good
07:29PM EST - Now Temash - tablet APU
07:28PM EST - 15W Kabini TDP, running OpenCL image processing workload
07:27PM EST - Comparing Kabini and a Core i3-3217U Ultrabook
07:27PM EST - Kabini - 50%+ performance increase from Brazos 2.0, first quad-core x86 SoC, shipping 1H2013
07:26PM EST - Face login is self explanatory, screen mirror pushes your display to other devices wirelessly
07:26PM EST - Use integrated front facing camera to recognize gestures
07:25PM EST - These sound like GPU computing applications that leverage the APU hardware
07:25PM EST - New software with Richland systems that enable things like gestures, face login, screen mirror, etc...
07:24PM EST - 20% - 40% increase in performance over AMD A8/A10 Trinity systems
07:24PM EST - Talking about Richland
07:23PM EST - HP AMD Sleekbook announced today as well - 15.6" system with touch
07:21PM EST - ASUS is taking the stage to talk about their AMD platforms
07:20PM EST - Recapping 2012 successes in the APU/CPU space
07:19PM EST - We want Kabini details
07:19PM EST - Now going to talk about APUs
07:18PM EST - Increasing performance year over year in each segment by 25%
07:18PM EST - 2nd gen GCN architecture
07:18PM EST - Talking about the Radeon HD 8500M, 8600M, 8800M GPUs
07:16PM EST - "we intend to continue to lead in this market"
07:16PM EST - 3 of 4 available consoles today, all have AMD hardware - so will the ones in the future
07:15PM EST - AMD is partnering with a CiiNOW - cloud gaming platform to stream 3D gaming content to lightweight clients
07:14PM EST - The latter is pretty easy when Nintendo, Sony and MS all will use AMD GPUs next gen
07:14PM EST - AMD wants to lead in the 4 corners og gaming: content, client, cloud and console
07:14PM EST - PC gaming hardware remains a very large and relevant market
07:13PM EST - Now talking about AMD's position on the Green500 list
07:12PM EST - Holding up the "ultimate AMD gaming notebook"
07:11PM EST - Recapping 2012 GPU launches
07:10PM EST - Talking about Radeon Graphics now
07:10PM EST - First quad-core SoC: Kabini
07:10PM EST - The 2013 lineup: Solar System/Sea Islands (GPUs), Richland & Kaveria (APUs), Kabini (ultrathins), Temash (tablets)
07:09PM EST - Going to talk about AMD's execution for the rest of the event
07:09PM EST - Talking about surround computing and how to bring that to reality, a lot of this sounds like what Carrell was trying to do before he was let go - Holodeck by 2016 right?
07:08PM EST - Lisa is going to talk about Vision and Execution
07:08PM EST - Lisa Su just took the stage

Thats hardly what I call an article. Who writes an article in reverse? Maybe that guy should have brought at least a pen so he could write on his hand at least one thing AMD reps actually said.

but apparently the only important thing there is

From Anandtech's AMD live blog : http://www.anandtech.com/show/6567/amd-ces-2013-press-event-live-blog

07:34PM EST - Richland looks like Trinity with a software update

07:33PM EST - Kaveri will be 2H2013 with GCN support

based on half-slide talking about the transition to surround computing, and you calling me out on looking at pictures and dragging speculation out of thin air? That was anandtech.

http://images.anandtech.com/doci/6567/AMD-024.jpg

yep, richland is nothing but trinity according to Anandtech, based on a picture, they are both blue.

But yet Im the one who just looks at pictures huh.

 

[gamerk316]

http://www.fudzilla.com/home/item/30081-amd-dismisses-low-end-tablets

Does AMD have a clue anymore? Putting your mobile chips in the Win8 basket?

 

[de5_Roy]

AMD: Engineers Are Gradually Shifting to System-on-Chip Development Model.
AMD’s High-Ranking Executive: There Are Benefits In Not Being AMD Veteran
http://www.xbitlabs.com/news/cpu/display/20130109181216_AMD_Engineers_Are_Gradually_Shifting_to_System_on_Chip_Development_Model.html

ASUS Unveils World's First PCI-Express 3.0 Motherboard for AMD Processors
http://www.techpowerup.com/178545/ASUS-Unveils-World-s-First-PCI-Express-3.0-Motherboard-for-AMD-Processors.html

 

[sarinaide]

i too, am eagerly looking forward to what steamroller (or whatever it ends up as) brings to the competition. imo right now, amd it right to bet on the low power apus, no offence to amd's higher perf desktop cpus (that got outsold by the apus 10 to 1 ).

Trinity appeals to a greater market, and is flexible. You get a high end platform and feature connectivity seen on expensive high end z77'and X79 platforms on a lower than mid level board budget, the discrete/integrated peformance gives perfect balance. I had said on the old thread that APU's were the way forward but what I was saying about Vishera was that despite being at least 2 generations off Intel it is significantly faster where it matters over Zambezi and importantly its put the 1100T to bed to, small step but a step in the right direction. AMD tended to situate the FX 83XX parts in between i5 and i7 in some instances it is correct but overall performance is generally around a i7 930-950 territory, on exactly the same arch as Zambezi it was a pretty solid showing across the board not one isolated synthetic the FX 8350 is by and far head and shoulders the best AMD chip right now.

10:1 ratio is hardly unsurprising the Trinity platforms as above delivers quantity and quality of features with exceptional integrated and very impressive discrete performance for very low cost, when you can run up to date software at copious ease it makes you really wonder why you break the bank for more, unless you are a enthusiast. The 5800K and FX 6XXX represent the best value for money on the AMD front and general budget performance.

 

[noob2222]

http://www.fudzilla.com/home/item/30081-amd-dismisses-low-end-tablets

Does AMD have a clue anymore? Putting your mobile chips in the Win8 basket?

Granted this is a different move, but personally a tablet is the only place I can see windows 8 having merit.

I haven't used an android tablet, but using my phone the thing gets slower every time you install a new app even if its on the sd card, and the more apps that are installed, the more it crashes. From what I gathered, the android OS has issues keeping enough free memory and tries to close background services to free up space, only to have the service restart itself.

http://kschang.hubpages.com/hub/How-to-Make-Your-Slow-Android-Phone-Go-Faster-for-Free

Granted this is on my phone and its not old, Motorola Droid 3, but If I had a tablet, the last thing I want to do is have to run it in a minimalized state just so it doesn't bog down.

Still could be a similar issue with windows 8 tablets, haven't had the desire to buy one, but if it came down to it, imy decision would be based on the more stable platform, be it windows 8 or android.

 

[Blandge]

Based on AMD's slides:

Richland
Process: GF 32nm SOI - (Same a Trinity)
CPU: PD based - (Same a Trinity)
GPU: Pre-GCN VLIW4 based Cayman derrivative or something similar - (same as Trinity)

So where does the 20-40% come from? Beats me.

 

[esrever]

Based on AMD's slides:

Richland
Process: GF 32nm SOI - (Same a Trinity)
CPU: PD based - (Same a Trinity)
GPU: Pre-GCN - (same as Trinity)

So where does the 20-40% come from? Beats me.

magic pixy dust.

 

[noob2222]

Based on AMD's slides:

Richland
Process: GF 32nm SOI - (Same a Trinity)
CPU: PD based - (Same a Trinity)
GPU: Pre-GCN VLIW4 based Cayman derrivative or something similar - (same as Trinity)

So where does the 20-40% come from? Beats me.

If richland is the exact same chip as trinity, it would be an even dumber marketing move than BD's comparison to I7 980. just call it a12 5900k and be done since its the same chip.
Mass Media would eat AMD alive if they did something that stupid. Would be the first time in recent history for a new cpu architecture to have nothing changed.

Richland will be a transition product between Trinity and the upcoming Kaveri APUs. By utilizing a revised HD 8000G graphics core design, AMD claims Richland outperforms Trinity in graphics intensive workloads by 20-40%. Those are some impressive numbers but how are they derived? There seems to be some wizardry going on behind the scenes since clock speed increases would never account for such drastic improvements. It is possible that AMD is finally integrating their GCN-based architecture into APUs ahead of Kaveri but for the time being, this is just speculation.
http://www.hardwarecanucks.com/forum/hardware-canucks-reviews/59053-richland-kaveri-kabini-temash-amd-s-2013-apu-lineup-examined.html

Instead, things are pretty hushed about what is ACTUALLY in the chip. Maybe AMD is trying to catch Intel off guard

 

[de5_Roy]

who the heck will buy a 'transition product'? amd doesn't have enough pr power to bs people into buying their junk. i checked newegg yesterday and there are still llano laptops selling alongside trinity laptops. now amd is adding richland powered laptops and will be adding kabini and temash based laptops and tablets, probably all of them under $700 mark, same as current apu based laptops.

 

[jdwii]

i too, am eagerly looking forward to what steamroller (or whatever it ends up as) brings to the competition. imo right now, amd it right to bet on the low power apus, no offence to amd's higher perf desktop cpus (that got outsold by the apus 10 to 1 ).

Of course they did and i'm sure its the same thing with Intel when it comes to their I7 vs Pentium/I3 market.

If richland is the exact same chip as trinity, it would be an even dumber marketing move than BD's comparison to I7 980. just call it a12 5900k and be done since its the same chip.
Mass Media would eat AMD alive if they did something that stupid. Would be the first time in recent history for a new cpu architecture to have nothing changed.

Richland will be a transition product between Trinity and the upcoming Kaveri APUs. By utilizing a revised HD 8000G graphics core design, AMD claims Richland outperforms Trinity in graphics intensive workloads by 20-40%. Those are some impressive numbers but how are they derived? There seems to be some wizardry going on behind the scenes since clock speed increases would never account for such drastic improvements. It is possible that AMD is finally integrating their GCN-based architecture into APUs ahead of Kaveri but for the time being, this is just speculation.
http://www.hardwarecanucks.com/forum/hardware-canucks-reviews/59053-richland-kaveri-kabini-temash-amd-s-2013-apu-lineup-examined.html

Instead, things are pretty hushed about what is ACTUALLY in the chip. Maybe AMD is trying to catch Intel off guard

Richland is going to be a very slight improvement on the CPU. Looks like Amd is focusing on power consumption more then anything this time. Wait until next year to see anything big from these guys.

http://www.anandtech.com/show/6600/intel-haswell-gt3e-gpu-performance-compared-to-nvidias-geforce-gt-650m

Looks like Amd might have a serious issue. Its OK for them to be behind in one area but not 3. CPU,GPU performance as well as power consumption.

 

[esrever]

Richland will fill A8 and A10 range of performance and kabini will fill the E-A6 line.

Richland will not be anything special. It Will be just something akin to brazo 2.0 but slightly more performance. AMD might allow Richland to go to 45W to allow for the performance improvements they are promising.

Richland is almost definitely trinity silicon by the roadmaps AMD have shown. Probably improvements in power gating and the fabrication process but not a completely new design.

OEMs need AMD8000 series inside their laptops which is why Richland will allow for a rebrand and a release. My speculation is there will not be any Richland APUs for desktop for sale, Only OEM supplies. Unless AMD has something amazing covered up, it wouldn't really have enough improvements.

 

[Blandge]

If richland is the exact same chip as trinity, it would be an even dumber marketing move than BD's comparison to I7 980. just call it a12 5900k and be done since its the same chip.. Mass Media would eat AMD alive if they did something that stupid.

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.

Would be the first time in recent history for a new cpu architecture to have nothing changed.

Just because it's largely based on the same underlying microarchitectures doesn't mean that nothing has changed. Just look at Phenom 2 or PD. In both cases there were not huge changes, but both saw a noteable increase in performance. I fully agree that even a 20% increase in performance makes no sense without some pretty big changes. Maybe they'll increased die size?

Richland will be a transition product between Trinity and the upcoming Kaveri APUs. By utilizing a revised HD 8000G graphics core design, AMD claims Richland outperforms Trinity in graphics intensive workloads by 20-40%. Those are some impressive numbers but how are they derived? There seems to be some wizardry going on behind the scenes since clock speed increases would never account for such drastic improvements. It is possible that AMD is finally integrating their GCN-based architecture into APUs ahead of Kaveri but for the time being, this is just speculation.
http://www.hardwarecanucks.com/forum/hardware-canucks-reviews/59053-richland-kaveri-kabini-temash-amd-s-2013-apu-lineup-examined.html

HD 8000G graphics doesn't imply GCN. Remember that Trinity was marketed as 7000 series, but really only used a 69xx series GPU microarchitecture. My primary reason for believing Richland is not based on GCN is due to the slide below. If Richland were GCN based it would say so on this slide.

Instead, things are pretty hushed about what is ACTUALLY in the chip. Maybe AMD is trying to catch Intel off guard

What would Intel do with that information? They already have a gigantic team working on graphics. I really do believe that AMD is more concerned with selling product than hiding information from competitors.

 

[-Fran-]

Does it matter if it's VLIW4 or GCN at this point? /scratchhead

From what I remember, VLIW4 power figures were quite impressive and they actually did the trick for having good performance per watt. GCN was computing oriented and hence it draws more power overall. Point is, Richland improving the VLIW4 on Trinity should be the obvious discussion here. I really don't think putting GCN like it is now and, more importantly, with no HSA yet is a smart move... Like, at all. Even if it is GCN 2.0, there's still no HSA to actually back it up yet. If you ask me, putting VLIW4 is actually the smarter move here. You can improve around that arch and slap a few things for the DX11.1 API and decode/encode stuff in the fixed pipeline.

Also, I think you might be right, Blandge. AMD could have increased the size of the die keeping the same power envelope. It's not so hard to achieve down the road I guess. I kinda forgot, but wasn't Richland in the same 32nm node?

Well, whatever they did, it better be as good as pixie dust, hahaha. I want to upgrade my A8 with more playback thingies. Hi10p is getting old

Cheers!

EDIT: Leave n00b alone! Bullies.

 

[Blandge]

Does it matter if it's VLIW4 or GCN at this point? /scratchhead

It only matters in that GCN would almost certainly account for a 20-40% performance increase, while a Cayman derrivative might make such gains more difficult to achieve.

From what I remember, VLIW4 power figures were quite impressive and they actually did the trick for having good performance per watt. GCN was computing oriented and hence it draws more power overall. Point is, Richland improving the VLIW4 on Trinity should be the obvious discussion here. I really don't think putting GCN like it is now and, more importantly, with no HSA yet is a smart move... Like, at all. Even if it is GCN 2.0, there's still no HSA to actually back it up yet. If you ask me, putting VLIW4 is actually the smarter move here. You can improve around that arch and slap a few things for the DX11.1 API and decode/encode stuff in the fixed pipeline.

Also, I think you might be right, Blandge. AMD could have increased the size of the die keeping the same power envelope. It's not so hard to achieve down the road I guess. I kinda forgot, but wasn't Richland in the same 32nm node?

Well, whatever they did, it better be as good as pixie dust, hahaha. I want to upgrade my A8 with more playback thingies. Hi10p is getting old

Cheers!

EDIT: Leave n00b alone! Bullies.

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.

 

[JAYDEEJOHN]

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