AMD Trinity news (looks, its near to launch?)

[Manicdan]

From your link.
Ivy
http://www.computerbase.de/bildstrecke/40725/46/
Llano
http://www.computerbase.de/bildstrecke/40725/49/
GF110
http://www.computerbase.de/bildstrecke/40725/52/

lol it turned HL2 into HL1, sounds more like a mod rather than a bug.

Dualcore SB parts are also classified as 35W, and beat the 25W amd part... I think by now it would be clear how amd and intel classified there TDP, that is determined by SKU and not actual consumption.
Also I like how everyone blames ivybrdige for the bad power cosumption, when its actually the platform (the asus N56V). Someone tested actual power consumption without monitor:
Attachment 126818
http://www.notebookcheck.com/Trinity...U.74604.0.html (its german so use a translator if needed)

While under worse case gaming scenario (furmark) the HD4000 consumes more then the APU on the A10, in actual games they are somewhat comparable and not much worse then the HD3000. The APU still takes the crown in fps delivered, but no one disputed or was even surprised by that.

Also why are we again focusing only on gaming with battery? I don't see whats the point of that, even the A10 can barely get more then 1:30 out under load... What really counts is reading/surfing/light office work/dvd/streaming. And trinity did a good job there catching up with the intel based books and even beating them depending on what configuration they run.

the chart shows some very odd info
the A10 has WORSE power consumption in fifa than it does with furmark. although all results come out to right around a similar cluster between 40 and 50w
and the ivy has a spread from 40 to 70w. that looks really bad to me. basically they have a really poorly optimized gpu, and for their cpu to do any real work it over draws the platform by an insane margin. i wouldnt call the n56v the problem when the cpu can go way past its specs.

[Hornet331]

nothing strange with that since ati and nvidia uses a driver profile for furmark to artificial limit the power consumption, this dates back to when people managed to blow out the VRMs of the HD4870 with furmark, nowadays not even renaming works anymore to bypass this thanks to the dedicated hardware on the gpus... if you want you can check google for that.

Intel on the other hand runs it without any profile (would be surprised if you will find one in upcoming versions of the driver) and furmark runs full speed there.

[tbone8ty]

Hybrid graphics now support dx9. Awesome!

[Lightman]

nothing strange with that since ati and nvidia uses a driver profile for furmark to artificial limit the power consumption, this dates back to when people managed to blow out the VRMs of the HD4870 with furmark, nowadays not even renaming works anymore to bypass this thanks to the dedicated hardware on the gpus... if you want you can check google for that.

Intel on the other hand runs it without any profile (would be surprised if you will find one in upcoming versions of the driver) and furmark runs full speed there.

Small correction, AMD used driver profile for Furmark up to HD69xx where they introduced PowerTune. Now it's back to hardware power controller in AMD camp for HD69xx HD77xx+ and Trinity.

[ajaidev]

Unigine Heaven 2.1 19% faster to HD4000
Cinebench R11.5 - OpenGL 64Bit 19% faster to HD4000
Diablo III med 34% faster than HD4000
Risen 2: Dark Waters low 20% faster than HD4000
Alan Wake low 45% faster to HD4000
Anno 2070 med 22% faster to HD4000
The Elder Scrolls V: Skyrim med 36% faster than HD4000
CoD: Modern Warfare 3 high 48 faster to HD4000
Battlefield 3 low 20% faster to HD4000


More here http://www.notebookcheck.net/AMD-Rad...G.69830.0.html

Overall its much faster than the HD4000 and it certainly has much more potential but there might be bandwidth problems and optimization problems. The Intel iGPU is fairly less complicated than what AMD has done. Dont have time to go into the specifics but AMD's design is meant for the future evolution while Intel will need a substantial change that is coming with Haswell.

Also like me if anyone was interested in Haswell vs Trinity GPU matchup its gonna be pretty gruesome, i am planning to buy a A10 5800K asap and compare it to the Haswell ES while the ES is clocked at a modest speed the GPU is unlocked and i can compare them at different speeds.

[informal]

I think the A10 on desktop will be fully unlocked,ie. both CPU and GPU can be OCed by multiplier alone.

[ajaidev]

That is the reason why i choose the A10 5800k i can compare both Trinity and Haswell at Max iGPU speeds but there is a bit of a confusion about which iGPU the Haswell is carrying since the card says its a GT2 but the performance is some tests is almost double of the HD4000.

Non the less will test them out when the AMD HW is available here. With Haswell we are having major problems with RC6 among other things but overclocking seems to be a strong point for the processor.

[CrimInalA]

Because AMD's Trinity test system didn't ship with a discrete graphics chip.

Yes but i'm asking myself , why ?
Before the launch there was so much talk about this feature .

[SKYMTL]

Yes but i'm asking myself , why ?
Before the launch there was so much talk about this feature .

In notebooks and particularly the area Trinity is focused upon, power consumption is key. Adding another chip to the Trinity "genericbooks" sent to certain sites would have lowered battery life, resulting in a poor comparison to Intel's chips.

Also don't forget that Dual Graphics technology has been around since Llano.

[informal]

That is the reason why i choose the A10 5800k i can compare both Trinity and Haswell at Max iGPU speeds but there is a bit of a confusion about which iGPU the Haswell is carrying since the card says its a GT2 but the performance is some tests is almost double of the HD4000.

Non the less will test them out when the AMD HW is available here. With Haswell we are having major problems with RC6 among other things but overclocking seems to be a strong point for the processor.

Don't forget that by the time Haswell launches AMD will be having SR core out(~1 year from now,roughly). So comparison will be valid for present time but not for future time . SR will have 512GCN cores running at comparable(or higher clocks) than Trinity today. This coupled with increased efficiency of GCN vs VLIW4 should amount to ~50% better GPU performance. Now memory BW constrictions may apply and it may not reach ~50% figure but this is pretty probable result.

[G.Foyle]

In light of recent leaks about Haswell and "memory BW restrictions", I would say we know nothing for sure yet. AMD might be very well doing something big for Kaveri, just like Intel is doing for Haswell.

[informal]

That is indeed true.We must remember they are preparing their products for broader GPGPU support in coming years and in order to tap into this potential, memory BW is very important.

[FlanK3r]

how is working exactly Turbo behind Cinebench R10 with A10-4600M single and multi? Did someone tested it (clocks jumping on stay at some value (2400MHz in multi and most time example at 2800 MHz in single?)?)

[BeepBeep2]

From what I've seen, SB uses about 20% more, here, here and here.

Edit: Under load.

I see a 15% difference in the madshrimps review, techreport review, about the same in the anandtech review. Remind me to never look at http://media.bestofmicro.com/B/A/334...ge%20power.png ever again though

[Mats]

I see a 15% difference in the madshrimps review, techreport review, about the same in the anandtech review

Maybe we're not looking at the same figures. I'm talking about the 3770K.

Madshrimps: 22%
Techreport: only 17%, but that's the slower 2600K
Anandtech: 21%, also 2600K

[BeepBeep2]

Maybe we're not looking at the same figures. I'm talking about the 3770K.

Madshrimps: 22%
Techreport: only 17%, but that's the slower 2600K
Anandtech: 21%, also 2600K

Yeah, I was looking at overclocked figures. (D'oh!)
Anyway, thanks for clarifying things.

[Mats]

NP
IB surely has potential. I'd like to see solder + new revision, but by that time Haswell is just around the corner..

(yeahyeah, there's always something new around the corner..)

[The Stilt]

Virgo (DT Trinity) will be a monster, no doubt

Sumo GPU inside Llano was a monstrous overclocker (most GPUs in 3870K do 50%+ overclocks 24/7), but the performance was always held back by a rather slow NCLK domain and memory bandwidth.

In Piledriver the frequency of NCLK domain is tied to DRAM frequency with ratio of 1.25:1 (minimum).
This means the NCLK in FM2 Trinity chips must be atleast 1333MHz to match the highest officially supported DRAM frequency (DDR-2133).
The maximum officially supported NCLK frequency on Llano was 900MHz, yet this was rarely used since the memory compability was rather poor at high NCLK frequencies.

Piledriver unofficially supports DDR-2400 DRAM frequency (up to motherboard vendors if it is available or not), just like Bulldozer.
It is highly unlikely the GPU performance is held back by DRAM or NCLK on Trinity any longer.

The Devastator (excellent choice of a name btw ) GPU inside Trinity will most likely have a couple extra MHz to spare too.
If the Devastator GPU has even a part of the overclocking capabilities of Sumo GPU we will see 1GHz+ GPU (SCLK) clocks available for daily use on Trinity.

If Virgo platform can reach ~4.5GHz core frequencies, IPC is even marginally higher compared to Gen1 BD and the leakage (power draw) is slightly in better shape... AMD has unleashed the hell.

Already pre-ordered mine!

[kuroikenshi]

^^ Every time you say Devastator I think back of the Blood Ravens Space Marines.

[Drwho?]

That is the reason why i choose the A10 5800k i can compare both Trinity and Haswell at Max iGPU speeds but there is a bit of a confusion about which iGPU the Haswell is carrying since the card says its a GT2 but the performance is some tests is almost double of the HD4000.

Non the less will test them out when the AMD HW is available here. With Haswell we are having major problems with RC6 among other things but overclocking seems to be a strong point for the processor.

How are you?

Francois Piednoel

[blindbox]

That is the reason why i choose the A10 5800k i can compare both Trinity and Haswell at Max iGPU speeds but there is a bit of a confusion about which iGPU the Haswell is carrying since the card says its a GT2 but the performance is some tests is almost double of the HD4000.

Non the less will test them out when the AMD HW is available here. With Haswell we are having major problems with RC6 among other things but overclocking seems to be a strong point for the processor.

Wha, you can say that out loud?

Still, it's 1 year away anyway.

[FlanK3r]

Virgo (DT Trinity) will be a monster, no doubt

Sumo GPU inside Llano was a monstrous overclocker (most GPUs in 3870K do 50%+ overclocks 24/7), but the performance was always held back by a rather slow NCLK domain and memory bandwidth.

In Piledriver the frequency of NCLK domain is tied to DRAM frequency with ratio of 1.25:1 (minimum).
This means the NCLK in FM2 Trinity chips must be atleast 1333MHz to match the highest officially supported DRAM frequency (DDR-2133).
The maximum officially supported NCLK frequency on Llano was 900MHz, yet this was rarely used since the memory compability was rather poor at high NCLK frequencies.

Piledriver unofficially supports DDR-2400 DRAM frequency (up to motherboard vendors if it is available or not), just like Bulldozer.
It is highly unlikely the GPU performance is held back by DRAM or NCLK on Trinity any longer.

The Devastator (excellent choice of a name btw ) GPU inside Trinity will most likely have a couple extra MHz to spare too.
If the Devastator GPU has even a part of the overclocking capabilities of Sumo GPU we will see 1GHz+ GPU (SCLK) clocks available for daily use on Trinity.

If Virgo platform can reach ~4.5GHz core frequencies, IPC is even marginally higher compared to Gen1 BD and the leakage (power draw) is slightly in better shape... AMD has unleashed the hell.

Already pre-ordered mine!

What do u think about Vishera? I think, this CPU could have better IMC than Zambezi, so maybe 2600 MHz memory support. Im really interesting for pure performance of this refresh (5% clock to clock?)

Francois: ...one Trinity here do not destroy your market share . But in my eyes is Trinity really nice product for lowend and mainstream notebooks. Finally good partial competition product.

[The Stilt]

What do u think about Vishera? I think, this CPU could have better IMC than Zambezi, so maybe 2600 MHz memory support. Im really interesting for pure performance of this refresh (5% clock to clock?)

Piledriver supports DDR-2400 max.
On Vishera NCLK & MEMCLK are irrelevant, Zambezi already offers more bandwidth it can spend.

Improved manufacturing process (less leakage -> higher clocks) and slight improvement in IPC would do wonders.
Also I would not be suprised if we would see more than four CUs in a single node.

[kl0012]

Review of a "real" notebook (not AMD prototype) with A8-4500 & dual graphics:
http://translate.google.ca/translate...%3Ft%3D1061293

[CrimInalA]

Review of a "real" notebook (not AMD prototype) with A8-4500 & dual graphics:
http://translate.google.ca/translate...%3Ft%3D1061293

Thanks for the link !
Now at least we know WHY amd has shipped it's prototype laptop without dual graphics . It's a major fail