It’s not every day we look at a ‘mainstream’ part. AMD’s soon-to-be-released Lynx platform (Llano architecture for desktop systems) is as mainstream as it gets. Targeted for a total system cost of $500-$600, this platform is geared toward the biggest market segment for PCs today.
While many of our readers wouldn’t go for this platform as their main PC, many of us have other uses for Lynx – HTPCs for one. Let’s see what AMD has cooked up for us with their latest accelerated processing unit.
The Llano Architecture
AMD’s desktop version of their Llano architecture (code-named Lynx) comes to light today in their A- series APUs (Accelerated Processing Units). Lynx contains an on-die CPU and GPU, much like Intel’s Sandy Bridge CPUs released earlier this year.
While Intel focused more on the CPU strength (let’s face it, Sandy Bridge’s processing power is killer), with Lynx AMD has gone the other route. They’ve slightly revamped their Stars core (which is found in existing Athlon II and Phenom II CPUs) and given the memory controller a makeover, giving support for up to DDR3-1866. I can attest to the fact that it takes no tweaking at all to run that kind of speed with timings of 8-9-8-24.
These CPUs share the same architecture as the laptop APUs released in mid June. Make no mistake though, they are targeting the broadest spectrum of PC purchasers hard with this APU.
As you can see, the APUs are available with two different clock speeds and two different graphics cores. The higher-clocked CPUs come complete with a higher TDP, but operate at higher frequencies all the time. The A8-3850 we’re looking at today has the highest clock speed at 2.9 GHz. It also has the higher-specced IGP – the 6550D.
All of the SKU’s ship with the strong DDR3-1866 capable memory controller. It’s a good thing too, because the IGP uses system memory and it needs the speed/timings a strong IMC is capable of. Everyone considering one of these systems should bear in mind, timings are very important according to AMD. Fast memory with tight timings are ok, but you’ll lose out if you go for memory MHz and disregard timings.
There are two IGPs available on Lynx. The more powerful of the two is the HD 6550D, clocking in at 600 MHz with 400 shader cores. Its little brother is the HD 6530D and specs out at 443 MHz and 320 shader cores.
The Surro graphics core is a revamped Redwood core, which was from the previous generation of ATI cards and was housed in the 5670 released in early 2010 (see Anandtech’s review). It’s no small feat to fit actual discrete-level graphics power on a CPU die. While we can’t expect miracles, AMD put a lot into this APU; I was very excited to see how it performed.
All of that fits in this little package. It’s amazing when you think about how far they’ve come being able to fit such a powerful GPU on-die. I remember not too long ago when GPUs this strong weren’t even available on a discrete card!
The top-side looks like any current AM3 CPU, but the pin array is very different. Lynx is esigned to fit in the new FM1 socket and work in conjunction with the A75 FCH.
The A75 FCH – Fusion Controller Hub
The A75 FCH has a host of features for a very reasonable price. With native support for both SATA 6GB/s (six devices) and USB 3.0 (four ports), it can definitely handle all the accessories you can throw at it.
AMD’s APU vs. Discrete GPU and Sandy Bridge
Unfortunately we were not able to present a direct comparison with Sandy Bridge. With the A8-3850, AMD is targeting both lower-end discrete graphics cards and the i3 2100 and 2105. According to their presentation at least, it performs very admirably against the discrete competition.
Features wise, Lynx clearly out-shines Sandy Bridge. While it won’t compete in direct CPU computing, it does stand a chance when using GPU-accelerated applications. AMD is pushing OpenCL hard with this release, stating there are over 50 programs that can take advantage of GPU acceleration.
Graphics wise, AMD definitely thinks it has a winner on its hands. In this chart, they have compared even an i5 2500K, which comes up short in the gaming department. Before you point it out, yes, the i5 2500K is by far a superior CPU. That’s not the target for this APU and AMD hasn’t claimed that.
Judging from AMD’s results, they have a very healthy GPU advantage. That chart isn’t even comparing the top-dog, it’s using the A6-3650, which has a weaker IMC. Of course, being company-submitted results they should be taken with a grain of salt, but AMD is definitely not in the habit of being untruthful about their testing results.
The A8-3850 from AMD came complete in-socket in a Gigabyte A75M-UD2H. It’s a stout mATX board with a pretty solid feature set. For an APU platform it should do nicely. Gigabyte recently launched their A75-chipset boards on their website and you can check out the full specifications and features there.
Gigabyte does a good job of making their boxes look nice, with pretty graphics plastered all over the place. It does do its job of protecting the motherboard and accessories on their way to you, so we can’t fault them for a little marketing.
There aren’t a lot of accessories, but there aren’t a lot needed. They give you four SATA cables, a back-plate, driver DVD and the manual. Oh, and for those using one of these setups for their HTPC (a worthy goal), they give you a Dolby home theater emblem for your HTPC case.
The board itself is a nice looking combination of blue and white. The baby blue has an electric-blue look to it. Combined with the graphite gray heatsinks, Gigabyte serves up a nice looking socket FM1 motherboard.
Here are a couple more shots of the front and rear for your viewing pleasure:
Socket FM1′s cooler mounting holes are in the same position of those with AM3, AM2+ and AM2 before it. AMD is nothing if not consistent, a fact for which we thank them. If you have a decent cooler already, there is no reason to force an upgrade between sockets. Intel didn’t get that memo (see the socket 775, 1366 and 1156 mess…though to give credit where it is due, they didn’t force a change from 1156 to 1155).
With 905 pins, the socket itself definitely can’t be mistaken for any other AMD sockets; if nothing else the keying will prevent you from putting the wrong CPU in there.
The upper left houses the power section, for which you have a 4-pin CPU power connector. That’s atypical compared to a lot of enthusaist boards but is plenty for this purpose. On the right are the RAM slots as well as the 24-pin power connector, two fan headers (one at the top, one at the right), an LPT connector (interesting) and TPM connector.
You can also see above those connectors the dual BIOS chips, which is a nice touch for any overclocker.
Moving down, we have two PCIe x16 slots (one runs at x16, the other at x4) and one slot each of PCIe x1 and PCI. There is interconnectivity galore, with two headers USB 2.0 and one each for USB 3.0, firewire, front panel audio and even a COM port.
While the controller can handle six SATA 6 Gb/s devices, there are only five internal ports. The other one is an eSATA port on the rear I/O panel.
The PCIe x16 slot can be used for AMD Dual Graphics if you add one of the qualifying (inexpensive) discrete GPUs for up to a 50% boost over the discrete card alone.
Speaking of the rear I/O panel, there are tons of connectors back there. Gigabyte has conveniently labeled them all for us.
As overclockers are inclined to do, we promptly pulled the heatsinks off to see what sort of contact they made.
The MOSFET heatsink is held on with spring-loaded push-pins. While screws are preferable, it seemed to do a good job and had excellent contact throughout.
The NB heatsink was screwed on and had some of the nastiest thermal ‘paste’ I’ve had the displeasure of removing. I thought ASUS had it bad with their flaky yellow stuff. This pink gunk took even longer to remove. I’d rather have hard-to-remove solid than hard-to-remove sticky any day. Not that many people will need to remove the heatsink; it had good contact and it’s not necessary to replace the TIM anyway.
There are fully 18 MOSFETs on this board to power the APU. Gigabyte uses “Lower RDS(on) MOSFETs”, which they say run at lower temperatures with high efficiency. The heatsink definitely didn’t get hot to the touch during testing, so there may well be something to that.
The BIOS is well laid out and contains pretty close to everything you’ll need to tweak the entire platform in the MB Intelligent Tweaker (M.I.T.) menu. Conveniently, the IGX configuration (graphics) sub-menu exists both in the M.I.T. menu and the Advanced BIOS features menu.
The rest is close to common throughout most BIOSes on the market.
All in all, a solid board in a small package. It does the new platform well.
Author’s Note – The original article incorrectly mentioned that there were no HDMI or DisplayPort outputs. It should not have made it through and was an error I meant to remove prior to publishing, as they are obviously there! That error has now been corrected.
Our test system today consists of today’s GA-A75M-UD2H and A8-3850 APU along with some nice G.Skill RAM. The kit is that which we reviewed earlier this year, but is operating at speed and timing of the RipjawsX F3-14900CL8D-4GBXM 4GB kit that will work nicely with Lynx’s strong IMC. The rest of the system consists of a Corsair VX450 PSU, a Patriot Inferno 100G SSD and a Thermaltake Big Typhoon CPU cooler.
The Big Typhoon is older but seems to do a respectable job cooling the APU. There is one important note on the PSU you choose for one of these systems – there are potentially issues with using large PSUs with a Lynx setup. The PSU to go with should be quality and sensitive to lower loads and not tuned for, say, 800-1,200W.
Lynx does overclock relatively well and there are videos out there to prove it. Unfortunately the few hours I had to try and overclock I didn’t get very far. My sample just would not cooperate no matter what I tried; even settings from a person inside AMD didn’t work, but I have not given up. Assuming I’m able to get this thing to overclock (a hopefully reasonable and safe assumption), I will post a follow up article with benchmark results. For now, though, we leave you with a video from VR Zone. They have the system running Linpack and Crysis 2 at the same time at a base clock of 150 MHz, CPU core speed of 3.38 GHz and GPU core speed of 900 MHz.
You should also know there are results in the wild that aren’t going to jive with reality. There is an issue with the BIOS on this board (and presumably others, but I’m not sure) where you can raise the multiplier as though it were unlocked. You can also set the IGP core speed up manually. While the BIOS will set the values and the software will oblige by reading what the BIOS is telling it, performance will not change. You can set the multiplier all the way up to 47x (for a speed of 4.7 GHz), but it does not affect performance of the CPU since it’s still operating at its multiplier-locked 2.9 GHz.
AUTHOR’S NOTE – See, all it needed was a little more time. This setup has been overclocked and is ready to bench. I’ll run the benchmarks and write a second article on the Llano overclocking experience and results. Stay tuned!
Performance in Benchmarks
To show performance in CPU benchmarks we’re going with the method you should be used to by now – relative performance. When reading the 2D benchmark graphs, the A8-3850′s time or score always equals 100%, with results from the competing CPUs expressed as a percentage of how this CPU performs. The actual benchmark times or scores are reflected below their percentage in the graph.
SuperPi. For AMD chips all the way up to the Thuban, it only gets better than this. They just can’t compete with SuperPi and don’t pretend to. For an AMD chip though, Lynx does a decent job and shows its performance is roughly that of a similarly clocked Phenom II.
WPrime is much better for AMD. They always emphasize multi-threaded performance and for good reason. The A8-3850 beats out an Intel i5 655K and is rather close to the Athlon II 640′s performance numbers.
Cinebench rendering tests are more real-world than SuperPi or WPrime and give a good indication of how the chip will perform in every day tasks that take advantage of multiple cores. Lynx doesn’t disappoint her, only losing to the 500 MHz-higher-clocked Phenom II x4 965 BE.
We don’t have any similar-performing systems to run PCMark 7 on, but the score is very respectable, with performance numbers in line with what you’d expect from the platform.
We were graced with a little competition from forum members Shelnutt2 and Evilsizer for the 3D portion of this review, with the most important competition being in the gaming benchmarks.
Regrettably, Shelnutt2′s results didn’t make it in. We were attempting to get some IGP competition going and his poor Intel Atom processor couldn’t run DirectX11 and would have barely been blips on the graphs. Nonetheless, we wholeheartedly thank him for the effort he put in!
Evilsizer’s comparison was more fair, if a little over-compensating on the CPU. He was running an Intel i7 920 at 3.4 GHz and an NVIDIA-based Galaxy GT430, right at the discrete level that Lynx can compete with.
Indeed, in the first comparison – HWBot Heaven Benchmark their scores are rather close to each other. Heaven does take some advantage of the stronger CPU and that is reflected in the results.
Because the 3DMarks could be heavily influenced by CPU power, we didn’t include 3DMark results from his system. You can see the results below; if you don’t want to hunt around on the screenshots, the scores are in the picture captions. What you’ll find is a strongly performing integrated graphics solution.
Gaming Performance – Where the Rubber Meets the Road
This is where AMD puts their money where their mouth is. Their documentation and press conversations heavily emphasized the IGP side of this APU. They understand there are stronger CPUs, but the claim is there is no better IGP on the market. Indeed, with Sandy Bridge’s inability to run DirectX 11, that puts the platform dead in the water for many modern games’ best graphics.
To test Lynx’s gaming acumen, we ran several gaming tests. There are three data points for almost every configuration (all except one iteration of Dirt 2) – the A8-3850 running 1080P resolution, the Galaxy GT430 running 1080P and finally the A8-3850 running at 720P.
I ran them that way because let’s face it – 1080P is a tough pill to swallow for an integrated graphics solution. 720P is still playable with great looking graphics up to at least a 21″ monitor, at least in my opinion based on watching these benches while they ran.
First up, we’ll look at Stalker: Call of Pripyat in three different configurations. They were all run in DirectX 11 mode with three graphics detail options. MSAA and tessellation were disabled for these tests. Again, it’s an IGP after all.
Color me impressed from the first bench. This integrated solution has some serious power. It kept up with the discrete solution and even beat it at low detail. When you drop resolution down to 720P it runs playable frame rates even at high quality.
Aliens vs. Predator is a bear for less powerful graphics solutions due to its tessellation and MSAA settings in its default configuration. The medium detail setting cut those settings in half as well as dropped detail levels and the low detail setting turned them off completely while dropping the detail settings as low as they could go.
So it’s not a mega powerhouse. It doesn’t deal well with 8x MSAA and tessellation, that’s for sure. But when you turn those things off it does make a comeback. When you move to the more reasonable resolution it’s playable even with some of the computing-intensive options enabled. AMD beat out the discrete NVIDIA card in every iteration of AvP.
HAWX2 and Dirt 2 are both a bit easier on GPUs. These were all run with 4x AA turned on. HAWX 2 did so well with every detail setting on high, I didn’t bother with another detail setting. Dirt 2 did well at high and better at medium.
The two competing graphics solutions traded blows here, with AMD winning HAWX 2 and dropping behind in Dirt 2. HAWX 2 is playable at high detail with 4x AA on. There is some seriously efficient graphics programming going on in that benchmark.
Dirt 2 is kind of a strange beast. I watched the runs of these benches all the way through to get a feel of how it would behave when gaming. With Stalker, framerates in the low 20′s were barely playable, with obvious lag spikes. Dirt 2 had no such spikes. It was as though 20 FPS equaled 35 FPS. There was no jumpiness at all. So, despite the low- and mid-20′s FPS, the game looked surprisingly playable throughout the benchmark period. That said, it never hurts to get an additional 10-15 FPS and at 720P, the game looked just as good.
Final Thoughts and Conclusion
AMD has definitely got a winner on their hands with their A-series APUs. The CPU itself acts like a similarly-clocked Phenom II x4 CPU and the IGP…what an IGP it is. Not one to overestimate generally weak integrated graphics, this APU surprised me and is for sure no weakling.
The A8-3850 is expected to retail at the $135 price point and should arrive on the streets July 3rd. A corresponding board should run you around there or even a little less. The Gigabyte A75M-UD2H used in this review will retail for $109.99 and is available now at Newegg.
For less than $250 (the price of a lot of motherboards we review) you can get an APU and board on which to build a stout system. The combination would be perfect for an HTPC. Place them in your favorite mATX HTPC case and enjoy not only HD programming but some decent gaming. Get yourself a wireless game controller and you can have your own home-built console.
While I wasn’t able to overclock it (likely my own failing due to lack of time, for which I apologize and do plan to remedy), even at stock the Lynx platform makes for a great base to an affordable budget system. AMD’s Lynx definitely deserves to be Overclockers Approved.
- Jeremy Vaughan (hokiealumnus)
Huge thanks once again to Shelnutt2 and Evilsizer for their assistance with competing benchmarks!