Richland review: Even the most diehard fan boy can admit AMD’s not in the hunt against Intel’s top-end processors
That doesn’t mean
still can’t give
a hard time. While AMD can’t compete with the
or even the
, the company’s rush to merge CPU and GPU to make the
has put more pressure on Intel than Intel would probably want to admit.
APUs, for example, blew away peoples' sales and performance expectations. Since then, AMD has kept its foot on the gas peddle. While it hasn’t eclipsed Intel’s efficiency per core on the x86 side, the graphics portion of the APUs have always lead Intel’s integrated graphics. The company’s Llano parts have been even more competitive, giving up better integrated graphics and fair x86 performance. From
, AMD introduced Trinity and the latest sequel:
Richland doesn’t bring any fancy new 3D transistors or a fabrication process that’s the envy of the technical free world, but it still offers a pretty compelling message: a high-clocked quad-core chip with graphics that’s surprisingly decent.
Since Richland rolled out just as Intel’s
hit, we had a nice set of systems all ready to go for direct comparison using the latest graphics drivers and UEFIs. Obviously, we’re not making a direct comparison between the $142 A10-6800K, $122 Core Core i3-3220 and three Core i7 procs, but since we had the numbers, we’re including them to give users an idea of how much they’re missing and not missing by spending twice or three times as much on a CPU.
For our testing, we outfitted three different systems with the same
GeForce GTX 580
cards, 16GB of Corsair DDR3/1600 and 240GB
Corsair Neutron GTX SSD
was used for all of our testing.
Obviously, you need to dig into the details of our tests. But rather than having you click around looking for the conclusion, we’ll just put it right here. Our impression of Richland is generally good. The graphics performance obviously schools the Intel parts that we’ve seen. Intel’s HD5000 and HD5200 is another story, though, but somehow we doubt you’ll get those in $140 CPUs.
Richland’s big problem from the enthusiast perspective is that we don’t quite get it. The APU, when rolled into a NUC-sized or HTPC mini PC, is a pretty damned decent story. What we don’t get is why you’d ever build with a Richland in a full-sized FM2 board. Our instincts tells us to suck it up and build on AM3+ with discrete graphics for maybe $75 more. Talking to motherboard vendors, though, we’ve heard FM2 boards in full ATX are overwhelmingly favored by consumers. That tells us that most of the people buying FM2 systems have very tight budgets, and putting out another $75 for a GPU is out of the question. And that's really the magic of the FM2 platform and Llano/Richland.
For those on very tight budgets, Richland does the job and it does it very well. For those of us with a bit more scratch, we think AM3+ or even LGA1155 is the better choice for a full-sized desktop box.
Click the next page for our Richland Benchmarks.
A good GPU test shouldn’t vary much from the CPU, and here we see amazingly close scores despite the differences in clock speeds, core counts and generations of cores. We found similar results in the overall scores in 3DMark11 (incorrectly labeled 3DMark2011 in our chart) as well as for the new 3DMark, so we decided to simply skip those charts here. The results is an overall score, and higher is better.
The story changes when we flip over to physics testing. The test is a theoretical physics test and favors core and thread count. Here’s a first glimpse of the steep hill AMD’s Piledriver cores still face: the dual-core Core i3-3220 is a dual-core Hyper-threaded part without Turbo Boost. That means the A10-6800K’s 4.1GHz to 4.4GHz clock speed and quad-core/shared modules don’t give up much more than the i3. The result is an overall score and higher is better.
Echoing 3DMark11, the new 3DMark also favors the higher core count parts, and we found the A10 and Core-i3 on equal footing, despite one being a quad-core. The results is an overall score, and higher is better.
The popular and free 7-Zip has a built-in benchmark mode that gives approximations of how well a processor would perform in compression duties. Here, the quad-core A10 actually proves itself well against the Core i3-3220 and—surprise—comes amazingly close to the new Haswell chip. A10, pat yourself on the back. The result is an overall score of MIPS, and higher is better.
Cinebench 10 is the older version of Maxon’s popular 3D rendering benchmark. It features the ability to render the test on a single core. This should give an idea of the efficiency of each respective CPU microarchitecture. Sandy Bridge is slower than Ivy Bridge, which is slower than Haswell. The Core i3-3220’s weakness is likely due to its inability run at higher clock speeds than 3.30GHz. Seen in red, the A10 single-core performance could use some help, especially when you consider it’s running at 4.1+GHz. The result is an overall score, and higher is better.
Oh, but those cores in the A10 do pay off when it comes to a multi-threaded workload. The A10 outruns the Intel dual-core handily. Those longer bars? You have to pay for those, buddy. The result is an overall score, and higher is better.
Cinebench 11.5 is the newest version of Maxon’s popular benchmark. The A10 again comes out on top of the Core i3, but the more efficient Intel cores and Hyper-Threading make it closer than you’d think. For those wondering if a 6-core Core i7-3930K is even worth it, look at the hexa-cores’ numbers. If you are getting paid to render out a scene, the hexa-core will save you time and money. The result is an overall score, and higher is better.
POV-Ray is a ray-tracing engine that’s been around the block. It favors core and thread count, and the A10’s quad-cores trounces the Core i3’s dual-cores, despite the presence of Hyper-Threading. Count this as a win for AMD. The results are in seconds, and lower is better.
There is an OpenCL version of Handbrake floating around in beta. But for now, we stuck with the public version of the latest build, so you’re seeing purely x86 performance, not GPU performance here. The A10’s quad part comes out on top of the Core i3 dualie, and again, we see the hexa-core part outrun even the new Haswell chip. Did your buddy tell you hexa-core’s are worthless even when you said you intended to transcode a lot of video? He was wrong. The results are in seconds, and lower is better.
For our Premiere Pro CS6 test, we take 1080P video shot on an EOS 5D Mk II and export it as a Blu-ray formatted video. All of the encoding here is done on the CPU rather than the GeForce GTX 580, as we want to see CPU performance. Keep in mind, there is a new version of Premiere Pro coming out with OpenCL support rather than pure CUDA. Still, this test is a pure CPU x86 test, and the A10 gets trashed really, really badly. AMD fans will say that’s Intel’s hand in making compilers that gives them the upper hand. We’ll say that whatever the reason is, it’s pretty clear that if you are using Premiere Pro CS6, you need to be running Intel silicon with at least four cores aboard and Hyper-Threading. This is simply horrible performance from the A10 here. The results are in seconds, and lower is better.
TechARP’s X264 HD 5.01 is a pretty thread-heavy benchmark that tests how fast a CPU can encode using the free X264 encoder library. It makes two passes for the benchmark, and the A10 and Core i3 are pretty much tied. This test also shows that if you intend to encode video, pay more for your CPU. The results are in frames per second, and higher is better.
Pass 2 of the X264 HD 5.01 benchmark favors thread and core count more, so we now see the quad-core A10 outpace the dual-core Core i3 by a healthy clip. The results are in frames per second, and higher is better.
For our Proshow Producer 5 test, we take a couple of hundred images shot with a Canon EOS 5D MkII, create a custom slideshow, and export it to a 1080P video file. ProShow Producer 5 pretty much tops out at four-cores. The A10 is slightly faster than the Core i3 part, but nothing to really text home about. The budget parts do amazingly well on this, considering how much pricier the other parts are in this roundup, which means photoslide show production isn’t something anyone should consider a heavy duty workload anymore. Perhaps ProShow Producer 6 will step it up. The results are in seconds, and lower is better.
HDRs are so popular now, it’s a built-in feature on point-and-shoot digital cameras and phones. Those modes are just fakes, though. If you’re going to make an HDR, you need different exposures, which is what HDR Soft’s PhotoMatix does. Favored by HDR experts, PhotoMatix has a batch mode that’s unavailable in the main interface, because it grinds most machines into the ground. It favors core and thread count, and here we see the A10 outpacing the dual-core Core i3. PhotoMatix is also one of those apps that actually show spending more money on the CPU helps. The Core i7-4770K is roughly twice as much (plus some change) as the A10-6800K, and the performance you get is actually the equivalent too. The results are in seconds, and lower is better.
Bah, we don’t give a damn that your phone has a 16-megapixel sensor. The future is about Gigapixel, so we used a GigaPan EPIC Pro to shoot 263 images with a Canon EOS 7D and stitched them with Stitch.EFx 2.0 to make a 1.1 gigapixel image. Stitch.EFx2.0 is interesting because the first two thirds of it is single-threaded, with the last third exploiting all of the cores available. It’s a good mix that favors efficiency, clock speed, and core count. The A10’s high clocks and more cores hands the dual-core Core i3 -- a pretty handy defeat which, frankly, gives the even the pricier Sandy Bridge and Ivy Bridge quad chips a good scare. The results are in seconds, and lower is better.
Click the next page to see some Richland game benchmarks.
For some real-world gaming tests, we ran Hitman: Absolution at 13x7 resolution with graphics quality turned way down. We expected the performance to be pretty uniform, but that didn’t pan out. The A10 actually came in last with the Core i3 running at nearly 1GHz lower clock speed, beating it. It’s pretty clear that Hitman: Absolution likes quad-core parts, and not quad-cores that share resources the way AMD’s Piledriver cores do. These are framerates, and higher is better.
Resident Evil 6 also saw the much higher-clocked A10 lose to the Core i3. It’s not by much, but a loss nonetheless. In reality, the gameplay between the six chips would likely be indistinguishable at 1080p with graphics quality turned up. These are framerates, and higher is better.
Total War: Shogun 2 CPU benchmark shows that pricier quad-cores have a pretty heavy advantage over the budget parts. In budgetland, the AMD chip was again slightly trailing the Intel chip, despite its core and clock advantage. This reminds us of a forum post we read on another site that said AMD’s chips were faster in gaming. Our results, and most of the hardware community, would disagree. These are framerates, and higher is better.
Valve’s particle test was created around the time the Core 2 Quad was released and made to show off a pure CPU-based physics engine. Again, the Core i3-3220 running at 3.2GHz or lower owns the A10-6800K running at 4.1GHz to 4.4GHz. We've seen this particular test favor cache size and memory latency, which is why we think the fat hexa-core part beats even the Haswell chip. These are framerates, and higher is better.
Echoing the Core i3’s prowess at gaming, we ran Dirt 3 at low resolution with features turned down to take the GPU out of the equation. The result was an A10 part that is OK with discrete graphics, but it's clearly trailing a chip that’s clocked almost 1GHz lower. Let's also take a moment to admire the Haswell's thrashing of the other quad parts and the hexa-core chip. These are framerates, and higher is better.
PCMark 8 came out just as we finishing our testing, and rather than being even later with this story, we rolled with our existing PCMark 7 numbers. The test has never particularly favored core counts, and here we see it’s mostly a wash between the Core i3 and A10 CPU. These are overall scores, and higher is better.
Click the next page to check out Richland's integrated graphics.
Haswell Integrated Graphics Benchmarks
AMD fans may be ready to declare non-victory and go home, but let’s now look at where the A10-6800K is made to really live: in integrated graphics.
For that test, we ripped the GeForce GTX 580 card from the systems and compared the respective CPUs on their graphics prowess. The Core i3-3220 represented HD2500, the Core i7-3770 flew the flag for HD4000, and the new Haswell Core i7-4770K shows up with HD4600. AMD’s A10-6800K features Radeon HD 8670D graphics. All three were running dual-channel DDR3/1600 RAM. For those who don’t know, if you intend to run integrated graphics only and want the best gaming performance, pay for the fastest RAM you can afford, as DDR3/1866 or DDR3/2133 will give you nice benefits.
The A10 doesn’t disappoint. It simply destroys the HD2500 and HD4000 graphics. Intel’s new HD4600 gets closer, but there’s no cigar. Maybe HD Graphics 5200 aka Iris Pro will beat it, but we didn’t have it to test. These are overall scores, and higher is better.
This test isolates the graphics, and we again see the A10 make a monkey out of all of the Intel parts. Don’t expect this to change much, either. These are overall scores, and higher is better.
We’re actually running out of things to say, as we really aren’t seeing anything change for the Intel graphics here. Maybe it’s just time for Charlie Sheen to step in with a cliché. These are overall scores, and higher is better.
The new 3DMark actually sees the results close up a bit between the A10 and 4000-series Intel graphics, but the A10 still wins, especially against the HD2500 Core i3 CPU. These are overall scores, and higher is better.
Firestrike is the heavy-duty test in the new 3DMark and is more apt for testing discrete graphics, but it’s still interesting to see the Intel parts lose out to the A10. These are overall scores, and higher is better.
3DMark’s Ice Storm Graphics test is intended for mid-range PCs. Here we see HD4600 close up the distance with HD4000. But again, the A10 reigns supreme. These are overall scores, and higher is better.
The Cloudgate test in 3DMark is aimed at entry-level computing. For the first time, the HD4000 and HD4600 outpace the A10. The reason? The emphasis shifts from graphics performance to x86 performance, when the physics is factored into it, and the Intel chips now come out on top. Still don’t get it? If you’re running a game so old and so graphically easy as say Quake III, the Intel graphics and CPUs can actually be faster due to the better x86 performance. Well, except for HD2500 graphics, which is apparently there to insult or tease you. Hmm, maybe Intel shouldn't be allowed to use the word "graphics" with HD2500. Just as you can't call something "juice" unless there's actually fruit in it, maybe the HD2500 should be the purple-colored drink of the graphics world next to the grape juice A10. These are overall scores, and higher is better.
In the graphic’s only portion of Cloud Gate when physics and x86 is not factored in, the A10 actually pulls even with the new HD4600 graphics. That pretty much tells us that Intel’s HD5000 and HD5200 graphics will be faster—they just won’t be cheaper. These are frame rates, and higher is better.
We ran Hitman: Absolution at 13x7 with the image quality trimmed way low, and we still saw pretty sub-par frame rates. It’s pretty clear that as good as integrated graphics has gotten recently, sometimes you really will need a real discrete GPU. These are frame rates, and higher is better.
Resident Evil 6 is surprisingly difficult on integrated graphics, but we think it’s almost playable at a low resolution. Still, count this as a win for the A10. This is an overall score generated by RE6, and higher is better. For what it's worth: Resident Evil 6 benchmark, can you please stop when you're done, rather than looping over and over again?
We ran Dirt 3 at ultra-low quality and 13x7 resolution. Believe it or not, it looks surprisingly good at that resolution. The Haswell’s HD4600 also motors past the A10 a bit, which might be why Intel keeps showing Dirt 3 in its benchmarks. These are frame rates, and higher is better.
Unlike most shooters, driving games don’t really need high frame rates to be playable. We ran Dirt 3 at 1920x1080 resolution and medium quality to see how integrated parts would run. The A10 and Haswell do OK. In fact, it would likely be pretty playable to most people's eyes. The HD4000 graphics in the Core i7-3770K would be pushing it and HD2500? Forget it. These are frame rates, and higher is better.
We’re still turning our OpenCL testing, but we thought CyberLink’s MediaEspresso 6.7 would be a fair test of the chips’s encoding performance, using the graphics side rather than the x86 side. Intel’s QuickSync has always been pretty fast, and here we see even the lowly HD2500 give the A10 a good run for the money. Interestingly the Core i7-3770K and Core i7-4770K are surprisingly close. HD4600’s graphics should be improved for computing tasks such as encoding, so we suspect MediaEspresso just can’t exploit it yet. The results are in seconds, with lower being better.
Sandra has a GPGPU test. It’s purely synthetic but the results actually mirror the A10’s gaming performance, for the most part. We’ll be adding more OpenCL tests using real-world applications as we move forward. The results are expressed as MIPS, and higher is better.
Richland packs in 1.3 billion transistors and measures 246mm2 using the same 32nm process and High-K Metal Gate process. The cores are the same Piledriver modules used in Vishera FX parts.