More Llano leaks (A8 APU extensively benchmarked)

[formulav8]

Looks good!

Edit: I still think they should have made a DualCore based Llano on 45nm. They let Intel get the leg up on them to fusion first. Although even Intel's high-end igp's is junk for anything half-way intensive.

 

[Blitzvogel]

Looks good!

Edit: I still think they should have made a DualCore based Llano on 45nm. They let Intel get the leg up on them to fusion first. Although even Intel's high-end igp's is junk for anything half-way intensive.

I'm impressed that Sandy Bridge can do what it does, but it does so well only because it runs at insane clock speeds. AMD still beats both Intel and Nvidia's butt at performance/watt/transistor when it comes to graphics. If someone was to run the GPU part of Llano at 1100 MHz, I'm sure SB would only be about a quarter as powerful. Besides, transistor counts of SB's graphics I don't think include the video encode/decode hardware, unlike counts for actual GPUs, which will have those, and the added count of transistors needed for DX11 compatibility.

 

[Khato]

Oh how I'm looking forward to actual reviews. For those interested, this 'leak' is claiming that the 400 SP of the A8-3850 overclocked to 870MHz with effective 2320MHz DDR3 scores a 3dmark Vantage Performance preset GPU score of 5250. This is in comparison to a Radeon 6570 with 480 SP clocked at 650MHz and effective 4GHz GDDR5 scoring a ~6k GPU score... and the more interesting one being the only 3dmark Vantage Performance preset GPU score I could find for a DDR3 version of the 6570 - it's a 6570 with 480 SP clocked at 840MHz and effective 1880MHz DDR3 and scores 4878. aka, the 5252 score would be -very- impressive as it would imply that Llano's memory bandwidth is pretty much equivalent to that of a DDR3 based discrete card, despite the fact that it's sharing it with the CPU.

 

[Khato]

I'm impressed that Sandy Bridge can do what it does, but it does so well only because it runs at insane clock speeds. AMD still beats both Intel and Nvidia's butt at performance/watt/transistor when it comes to graphics. If someone was to run the GPU part of Llano at 1100 MHz, I'm sure SB would only be about a quarter as powerful. Besides, transistor counts of SB's graphics I don't think include the video encode/decode hardware, unlike counts for actual GPUs, which will have those, and the added count of transistors needed for DX11 compatibility.

Yeah, unfortunately Intel's shaders are more along the lines of NVIDIA's design than AMD's. As we've seen, slower, smaller, and simpler ends up with comparable performance for slightly less die size and power. That said, the ~41mm^2 die size area of SB's graphics does include the video decode/encode logic - only thing it doesn't include is the display I/O logic for the FDI links. Sure doesn't include DX11 compatibility though! But at least it's DX10.1 compatible eh? Haha

 

[nonameo]

legal mumbo jumbo

In that case, I guess they bin the chips in malaysia? Forgive me if I sound like an idiot, I'm not too knowledgeable about CPU manufacturing. Is it even possible for them to test the chip with just the core and no packaging?

Looks good!

Edit: I still think they should have made a DualCore based Llano on 45nm. They let Intel get the leg up on them to fusion first. Although even Intel's high-end igp's is junk for anything half-way intensive.

well, one line of thinking might be that they already have brazos, why have so many products for the same space?

 

[busydude]

The hope is for a thin/light notebook with good battery life at a low price.

A new video was released yesterday.. comparing the battery life of a core i3 mobile with AMD A8-3500M.

http://www.youtube.com/watch?v=LkbdyNO95hA

 

[Arkadrel]

Wow that is seriously impressive.

Llano @3.77ghz (30% OC) + iGPU@870 (45% OC)

3Dmark 11 P1591 (45% GPU OC => 38.5%result increase! )
3Dmark Vantage P6160(45% GPU OC =>40% result increase! )
3Dmark 2006 10492 (45% GPU OC => 37.1% result increase! )
note: CPU is OCed by 30%,from 2.9 to 3.77Ghz,DDR3 memory is OCed by 45%

Lmao at a 3Dmark 06 score of 10,492 3d marks.
That is impressive for a integrated GPU, that goes with your cpu, and uses your system ram.

Looks like it does atleast 2x the 2600k score's, in 3Dmarks tests.

 

[VirtualLarry]

Wow that is seriously impressive.

Llano @3.77ghz (30% OC) + iGPU@870 (45% OC)

Lmao at a 3Dmark 06 score of 10,492 3d marks.
That is impressive for a integrated GPU, that goes with your cpu, and uses your system ram.

Looks like it does atleast 2x the 2600k score's, in 3Dmarks tests.

Yeah. 10K 3dMARK06 is like a 9600GSO 96SP discrete card with GDDR3. Very nice for an IGP.

 

[nyker96]

I have to say, the low end gpu market is finished! both intel and amd integrated gpu core are quite good. you can almost game on it now.

 

[Arkadrel]

Yeah. 10K 3dMARK06 is like a 9600GSO 96SP discrete card with GDDR3. Very nice for an IGP.

source:
http://www.guru3d.com/article/geforce-9600-gso-386-mb-review-point-of-view/10

Yeah but a 9600GT 512mb scores like ~4900 3Dmark Vantage (dx10).
The Llano IGP (oc'ed) got over 6000+ score.

Obviously the Llano's IGP (OCed) is alot faster than a 9600GT, dispite the 9600GT haveing more memory bandwidth.

 

[podspi]

A new video was released yesterday.. comparing the battery life of a core i3 mobile with AMD A8-3500M.

http://www.youtube.com/watch?v=LkbdyNO95hA

I can't recall a time I have ever seen AMD marketing as aggressively as they have this year.

I still think that Unlocked. Ready. And Stable. was a little too over the top, but I approve of everything else :thumbsup:

 

[Gigantopithecus]

I can't recall a time I have ever seen AMD marketing as aggressively as they have this year.

According to the Steam hardware survey, ~20% of Steam users are using discrete GPUs that are less powerful than the A8 APU.

I think that's a very revealing statistic.

 

[CTho9305]

In that case, I guess they bin the chips in malaysia? Forgive me if I sound like an idiot, I'm not too knowledgeable about CPU manufacturing. Is it even possible for them to test the chip with just the core and no packaging?

You can connect probes to the die and use them for "sort" testing even before the wafer is sliced up.

 

[wahdangun]

source:
http://www.guru3d.com/article/geforce-9600-gso-386-mb-review-point-of-view/10

Yeah but a 9600GT 512mb scores like ~4900 3Dmark Vantage (dx10).
The Llano IGP (oc'ed) got over 6000+ score.

Obviously the Llano's IGP (OCed) is alot faster than a 9600GT, dispite the 9600GT haveing more memory bandwidth.

yeah, I feel shocked too, and to think about it, llano is amd key of success to penetrate mobile market and increasing their desktop market share rather than bulldozer ever will.

 

[Kevmanw430]

I really think AMD is on a roll here. Also, if they can somehow, with Hybrid XFire, get the 400SP IGP to get the frame buffer from the GDDR5 on a 6750M or 6770M, then you could be getting the equivalent of 6870M performance. (880SP vs 800SP, 1GB GDDR5 on both w/128 bit busses.)

 

[Idontcare]

You can connect probes to the die and use them for "sort" testing even before the wafer is sliced up.

Impossible, didn't you read about it? The innovative technology necessary to even make this happen was only just recently developed and reduced to practice at IBM.

http://www.eetimes.com/electronics-news/4216299/IBM-to-test-wafer-pruning

The new wafer pruning technique is currently being characterized by IBM Corp. for its 45-nanometer process, using on-wafer monitoring structures that can be probed during fabrication to spot bad wafers early-on.

Now some would appear to argue that this approach has existed in practice for decades, and you appear to hail from this false school of thought.





P yes this post is entirely sarcastic, sadly though the EETimes article and claims by IBM/UCLA are not...I hear rumblings that IBM and UCLA are experimenting with replacing aluminum wiring with some newfangled technology called "copper"...)

 

[CTho9305]

Impossible, didn't you read about it? The innovative technology necessary to even make this happen was only just recently developed and reduced to practice at IBM.

http://www.eetimes.com/electronics-news/4216299/IBM-to-test-wafer-pruning



Now some would appear to argue that this approach has existed in practice for decades, and you appear to hail from this false school of thought.





P yes this post is entirely sarcastic, sadly though the EETimes article and claims by IBM/UCLA are not...I hear rumblings that IBM and UCLA are experimenting with replacing aluminum wiring with some newfangled technology called "copper"...)

That sounds like they're rejecting a whole wafer based on scribe-line test structures, rather than doing die-by-die sort. Seems obvious to me, but I wasn't around for early bringup of pre-45nm nodes to know whether or not various fabs were doing it.

 

[Idontcare]

That sounds like they're rejecting a whole wafer based on scribe-line test structures, rather than doing die-by-die sort. Seems obvious to me, but I wasn't around for early bringup of pre-45nm nodes to know whether or not various fabs were doing it.

Its been industry standard practice for decades (>2 that I know of personally) rejecting not only whole wafers but also entire lots for all manner of established reliability metrics of concern based on probeable test structures contained in the scribe. (at least as far back as 1um, the gate-oxide integrity check was done with a scribe test structure post M1...if it failed then the wafer was scrapped inline asap, SOP for the whole industry as far as I was aware)

I'm fairly certain this is an issue of a non-industry academic doing his best to hype his pet university project to an otherwise non-industry tech journalist while invoking IBM's name as a means of garnering credibility by association. I can see it happening. Its just sad that it did/has because it is laughable.

It is also possible that the tech journalist just horrible misrepresented the original claims of the UCLA professor and it was never intended to be hyped as the invention it claims to be or it really does deserve the hype but the part that makes it novel got lost in translation somewhere?

 

[Gigantopithecus]

It is also possible that the tech journalist just horrible misrepresented the original claims of the UCLA professor and it was never intended to be hyped as the invention it claims to be or it really does deserve the hype but the part that makes it novel got lost in translation somewhere?

 

[Cerb]

source:
http://www.guru3d.com/article/geforce-9600-gso-386-mb-review-point-of-view/10

Yeah but a 9600GT 512mb scores like ~4900 3Dmark Vantage (dx10).
The Llano IGP (oc'ed) got over 6000+ score.

Obviously the Llano's IGP (OCed) is alot faster than a 9600GT, dispite the 9600GT haveing more memory bandwidth.

Bandwidth contention will be an issue, and there will be good reason for desktop users to get faster DDR3, if they want light gaming, but the 9600GT is nothing special. It has been meh for years, now. With stock specs just shy of a 5570, I would hope that an OCed one could beat the crap out of a 9600GT! That it gets right at [stock] 9800GT scores, OCed, is pretty impressive, and could spell the end to casual gamers and MMO-only gamers buying cards.

It is nice to see that the CPU speeds, even prior to release, are largely TDP-limited, as well, and that performance increases almost perfectly with clocks. Great news for budget BIYers in the near future.

 

[AtenRa]

I have to say, the low end gpu market is finished! both intel and amd integrated gpu core are quite good. you can almost game on it now.

Both AMD and NV will produce 28nm Low End discrete GPUs that will be smaller, faster and have lower power usage than current 40nm Low End chips.

Llano is nice and it has an adequate performance but Low End discrete cards will continue to be with us for a long time.

 

[krumme]

Perhaps we disagree what low-end is...
In my view SB is killing lowend sales now. This stuff will kill all low-end and mid-end stuff on the mobile side. Effectivesly goodbuy 64bit bus. Ask the OEM, but i think they will have to stuff the fastest gfx in the laptops to make such a difference, - that it is a difference for the average consumer.

Have fx. a look at Dells site. How much sold is in the order of say 800 shaders and up?

There is tons of 400 / 48-96 stuff on the mobile market. What is the highend above that? 0,3% of the total mobile laptop market?

At the time 28nm comes the Llano will change to a bd based, raising speed further, because it should be more simple to make the fast interface between dual modules and the gpu part instead of this old 4 core stuff.

 

[386DX]

A new video was released yesterday.. comparing the battery life of a core i3 mobile with AMD A8-3500M.

http://www.youtube.com/watch?v=LkbdyNO95hA

LOL Another AMD marketing video. First off in case you didn't notice the video is "Simulated", those aren't "real" systems as you can tell the AMD system is just a mirrored image of the Intel system (numpad, enter key on wrong side of keyboard, etc). Secondly at the end of the video where they briefly mention the test systems you can see they are testing a "Retail Intel System" vs an "AMD Press Sample System". If the Intel retail system was an HP like the simulated video would suggest you can be sure its loaded with bloatware that's obviously gonna never let the system idle compared to the AMD System that's gonna be bloat free. To further try and skew the results to AMD's favour the AMD system was also put on "Fixed Power-Saving Low Power Mode" and they specifically specified that "Discrete Graphics" was turned off. It was never mentioned what the Intel system had but since its a "retail" system it could potentially have a discreet graphics onboard that was never turned off.

If AMD wanted any credibility with there "benchmarks" they should really do it with settings people are "actually" gonna get when they buy a system and not some tweaked system for whatever benchmark they are currently doing. You know when it comes time to put out CPU/GPU benchmarks they are not gonna be using a system with "Fixed Power-Saving Low Power Mode"

 

[Nemesis 1]

Ya I seen the mirrored image yesterday when I watched the video. Lets just wait for the release see how the review sites handle this . Will they O/C the intel IGP to 1.8 ghz which is rather easy to do . Will they overclock the memory which they seldom do . Its going to be interesting But tis a cpu the Non 3D benchmarks will tell the whole story .

ANand will likely get a Driver update from intel if he already hasn't putting Cl on the cpu to work.

 

[jpiniero]

I thought mobile Llano looked interesting until I saw the speeds. The highest end part is 1.9 Ghz with 2.6 Ghz Turbo? Even at Quad Core, that's not going to cut it. I'm skeptical of Llano's appeal outside of gamers, who would be better off with Sandy and a discrete anyway.