Thanks for the review, Lawrence!

Great review of the new APU! Will there be an article about how these characteristics effect silencing, the sonic characteristics of the stock HSF, that sort of thing?

Why was the memory only tested @ 1333Mhz. This platform is heavily dependent on memory bandwidth. I was hoping to see if the increased bandwidth affected the power consumption.

I'm also curious if other mobo's would offer better power consumption (I thought Gigabyte mobo's are not very efficient).

Presumably they wanted to keep the same speeds for every platform. For those who don't know, the GPU is heavily bottlenecked by memory bandwidth so gaming benchmarks scale almost perfectly with memory speed.

Edit: Would it be possible to rerun some benchmarks with 1600MHz memory? I assume that would be the most common configuration for Llano.

I agree that all systems should be tested equally when possible and makes it possible to compare against older reviews. However, limiting the memory bandwidth to the lowest common supported speed (1333Mhz for the i3 compared to 1866Mhz for the A8) is a handicap. You would not test an Intel 1366 system using dual channel memory just because the competion or other Intel systems do not support triple channel memory.

I hope you are right. However, I wouldn't be surprised to see them save a nickle and use 1333Mhz memory (or charge $150 more for 1866Mhz).

While I well understand the assumption that the GPU could be RAM speed limited, for what the few raw performance tests I've read there hasn't really been that much difference between 1333 and even 1866. You can see it in the scores but overclocking the APU a little gives you more in comparison. By all means get 1600 speed memory if you buy new since the price is almost the same today, but if you have 1333 on hand it seems to be doing just fine here. Later when they present a more complex APU core then it might be a different story. AMD has always been keenly aware of where the RAM business has had its sweet spot.

Edit: I just read the Anandtech test and there the 1333 memory lags behind significantly in some games after all.

Great article! As always, love the power figures on all of the benchmarks!

I agree with those talking about testing 1866MHz RAM. I've read several articles comparing 1333/1600/1866 MHz RAM, and the results scale nearly 1:1. I see the point in keeping the platform the same, but perhaps if you count the added cost of the faster RAM in the price analysis it is fair to test with it? Llano is definitely an exceptional case in regards to memory clock/timing.

I also agree with the suspicion that the Gigabyte board is probably not the best one to be doing power comparisons with When will Intel start making FM1 boards?

Also: I believe there is an error in the article concerning the price of H67 motherboards. $160 appears to be the maximum price on Newegg, not the average. I have not checked any other figures. Perhaps you should double check the spreadsheet you used to calculate the averages? Hope this doesn't ruin all your charts!
[edit: this is per the following price-sorted list of current Newegg H67 offerings. Perhaps my search missed some boards? But $160 seems very high even without looking at data.
http://www.newegg.com/Product/ProductLi ... =20&Page=1
]

This is the one. Considering that 1333MHz doesn't cost much less than 1600, faster memory costs enough to pay for a discrete GPU and Llano is pointless if you don't need a GPU I think 1600 is the only point that makes sense. It boosts fps by 10-15%.

to me, llano is a stepping stone. amd will be changing up their gpu architecture soon to be stronger in gpu compute, and we know that their new cpu arch is right on the horizon too... it feels like theyve just put this together to prove to themselves that it can be done. maybe to get a bit more of a jump with how drivers will work, and maybe give devs a bit of a sample of where theyre headed.

AMD also had to get something out the door. They are getting pummeled in the marketplace, and were not doing too hot even before Sandy Bridge. I just pray this takes us in the direction of the Athlon glory days

What I would like to see: image comparison between the integrated GPU's, both games AND movies... maybe that's the explanation for the highly different power consumption for the i3 and A3850.

For me in the US looking at prices I see DDR3 2000 and 1866 modules reasonably priced (mixed in with modules that are overpriced). More expensive than DDR3 1600 but not so expensive I wouldn't just go for the 2000 if I were building a system. I know that isn't the case in Europe, Australia, and other parts of the world yet. I also know that you have to be darn careful choosing your ram or you'll overpay or get ram that will only run at higher voltages. (Be sure you are looking at 1.5V DDR3 when price shopping)

Think about how this will play out long term. Articles/Reviews get read for months and people refer back to them even a year or two later as they argue the merits of the latest price changes. It takes a long time for a socket/platform to go out of production and it's only a matter of time until DDR3 1866 or even DDR3 2000 is cheap in your neck of the woods.

A review should test the options and give enough data to make new choices as the prices change. Reviews can only accurately compare speeds not prices.

If the initial review can't test the options sufficiently a follow up review with high visibility (front page placement, links from to/from the prior article, mention in the discussion thread for the prior article) is the best way to remedy it. To be really thorough you'd add edits to the conclusion page + any page that mentions the ram configuration + any page that shows test results affected by the RAM/graphics portion of the APU.

I'm surprised that AMD didn't send out ram kits to go with the APUs or at least give the reviewers a heads up so the scaling data would be highlighted in more articles.

Z68 boards were also included too because some people may buy them for the SSD caching and Virtu graphics feature. Is it likely? No, but more likely than going with P67. I don't want to throw out too many motherboards that I personally think shouldn't be paired with a given processor.

As for the memory question, DDR3-1333 was used because it's the most common speed and was used on all our other test configurations. I'll test it again with DDR3-1600.

don't suppose we can get some undervolting done too?
providing the mb supports it of course.

with the a8-3800 being only 500mhz slower and having a tdp of 65w, the logical explanation is that there's some pretty hefty undervolting going on.

The price difference between 1600 and 1866/2000 does seem to be larger here in Finland than on Newegg. I wonder how on earth that's possible - air mailing a memory stick around the world can't cost much.

Speaking of undervolting, I wonder how a Llano APU would respond, considering SPCR's sample ran at 1.40v... a lot of volts for even a 65nm CPU.

Consider that the A8-3800 (2.4/2.7GHz vs 2.9GHz for the 100W TDP 3850) has a TDP of 65W. That should give you a pretty good indication of what's going on. X-Bit Labs had a 3800 that ran at 1.2-1.25V .

ht4u did some undervolting (the only site I've seen doing this so far for Llano):
http://ht4u.net/reviews/2011/amd_llano_ ... ndex31.php
They got the 1.4V sample to work stable at 1.16V (!). As you can guess this dropped (load) cpu power usage by nearly a factor of 2 (from 96W to 54W).
The high voltage is really pretty strange, I believe the 45nm A II/Phenom II have such high voltage only well past 3Ghz. OTOH if you think the safety margin AMD applied here is too much, well maybe but it didn't OC that well at 1.4V neither (ht4u sample did 3.33Ghz at 1.4V - that's just little more than the roughly 10% you usually should have). I think the chip is more optimized for low leakage (idle power) rather than low dynamic power.

A8-3850 is a 32 nm HKMG SOI chip with a 100W TDP.
A8-3800 is practically just as fast but has a 65W TDP.

And right now they are selling every one they make (higher demand than AMD predicted but more so on the mobile side than desktops). As they ramp up production of these I expect the availability, price, and power draw will improve though only the 100W versions are on newegg at this point.

Rumors make it sound like the FM1 socket will be short lived as the Trinity chip that will replace these A8 models in 2012 is to us socket FMx per AMD slides. But I think that is just a poor interpretation as the x will be FM1+ or FM2 or maybe even FM1. I don't know but I'm guessing the FMx = something with the same pin layout as FM1 plus or minus a pin or two with the thicker pins of bulldozer (0.51mm instead of the old 0.45mm). Assuming they do that it would be possible to use the old FM1 APU in the new motherboard with socket FMx whatever x may equal just like using a AM3 processor in a AM3+ board.

Thanks for the great review. I will have liked some undervolting and maybe underclocking power consumption test.

The 32nm used on AMD APU is probably quite bad since the power consumption is higher than the old 45nm part for the same core and performance.

I really love the APU concept unfortunately the power consumption is really bad so I guess I need to wait for the 22nm from Intel to make an upgrade.

Power consumption well it's running a CPU and an embedded GPU unit as well..so it's probably not that bad at all. If you took the GPU out of it then it would be a fair bit lower.

The concept is a good one this is just the start of something that could have a lot of potential.

Hi guys, I've updated the article with DDR-1600 results. Relevant changes on this page:
http://www.silentpcreview.com/article1211-page4.html

Thanks!

Nice!

It's surprising that "Alien vs Predator" increased more than the 20% the RAM speed would imply (timings were the same, yes?). IANA graphics engine expert. Any theories?

The CPU and GPU are sharing that ram so if a program were using both close to equally contention could slow both down more than 20% each. Any time waiting on data means zero work.

Think traffic jam vs smoothly traveling cars on a highway. The number of cars doesn't change anywhere near the ratio of speed drop. The zero work state for the CPU or GPU is very similar to coming to a complete stop on a road. It takes time to get back up to full speed and get some real work going again and any further contention would make speeds sputter instead of accelerating smoothly.

It's also worth noting that the speed increase doesn't drop off at DDR3 - 1600. Take the next step to DDR3 - 1866 and you'll see similar gains (not quite as large but still noticeable/significant). It's only a matter of time until the DDR3 1866 price premium drops away and that ram becomes the obvious choice.

4GB (2 x 2GB) 1333 $35
4GB (2 x 2GB) 1600 $40
4GB (2 x 2GB) 1866 $70
4GB (2 x 2GB) 2000 $65

For whatever reason DDR3 2000 is cheaper than DDR3 1866 right now. But if you are going for the cheapest possible A8 config 4 GB of 1600 make sense.

8GB (2 x 4GB) 1333 $65
8GB (2 x 4GB) 1600 $70
8GB (2 x 4GB) 1866 $85

At 8GB the percentage price premium of 1600 over 1333 is even less as is the premium for 1866 over 1600. If you are going to put 8GB with your A8 I'd go ahead and pay for the 1866 premium.

(newegg pricing without tax or S&H, be sure to do your own price searches to see what makes sense for your location and watch for sale prices and such).

I realize contention becomes much more of an issue with bus saturation, but I have trouble conceiving of a real world scenario where the overhead becomes significant compared to what would obviously be a bandwidth problem. The traffic analogy is problematic, because (if we call this a stop light), the length of time it takes for cars to get up to speed should be insignificant compared to the length of the green light (not the case for real cars). Makes me wonder if the controller is doing a poor job, and/or the graphics engine is a complete piece of junk (forcing cache-misses, forcing data synchronization in a way that causes contention without bandwidth problems). Then again, shared CPU/GPU memory channel is probably not an engine design consideration these days...

Still all seems very strange, as I would expect overhead costs to be reduced by the 20% as well (on a per-operation basis). This all assumes the timings are the same at 1333 and 1600.

Thanks! Calling those gains "significant" would be an understatement. 28, 37 and 60% boosts in 1440x900!

5570 with DDR5 has memory bandwidth of 57.6 GB/s vs 520 GFLOPS

5570 with DDR3 has memory bandwidth of 28.8 GB/s vs 520 GFLOPS (and is held back by the slower memory)

A8 with DDR3 1333 has memory bandwidth of ~20 GB/s vs 480 GFLOPS on the GPU portion and it's sharing this memory bandwidth with the CPU so it's something less than 20GB/s if the CPU is busy.

That's a lot of stop signs / red lights. It's as if the light is red 50% of the time. The ram could be twice as fast and still not be as fast as the DDR5 used on a discrete card.

Rough numbers

DDR3 1333 20 GB/s
DDR3 1600 25 GB/s
DDR3 1866 30 GB/s
---DDR3 2000 32 GB/s
DDR3 2133 34 GB/s

Maybe contention isn't the best term to use. Maybe I should just say the GPU portion is starving due to lack of data coming through due to bandwidth limitations. Or maybe we should avoid the how it works analysis and just say it is throttling the GPU or maybe even that has auto connotations so I should just say it's holding it back.

But obviously price is whats pushing people to stick slower ram in a system that would easily take advantage of faster memory.

It'll be interesting to see the benches come Christmas buying guide time when prices have shifted and the 65W versions have shown up.