AMD’s Trinity desktop processor is on the way for enthusiasts and desktop users looking for better performance over their Llano APUs and it’s on track for a launch later this year with new socket FM2 boards. If you’ve been living under a rock, FM2 is pin-incompatible with socket FM1, necessitating a complete upgrade for those of you dissatisfied with current performance on their rig. But I have to warn those of you in that predicament right now – Trinity won’t help your woes.
In fact, Trinity represents over a year of re-engineering Bulldozer modules into Piledriver, the next in the series of the revamped processor lineup that AMD has been pushing since they discontinued Athlon and Phenom chips. Its only a 15% performance increase but its an increase in the right direction. For now though, OEMs get to play and fiddle with the new chips while drivers, BIOSes and software gets sorted out. Hit the jump to find out more.
I’m not going to delve into the great mysteries of Trinity or where all the improvements are for now – I’ll leave that up to my proper analysis when the family actually launches later this year for both mainstream and OEM consumer channels. For the Tom’s Hardware team, they were a lot more interested in actual benchmarks and delved very little into the chip side of things. So for now, its only the benchmarks that I’ll be looking at.
Picking things up a bit from the quad-core Llano A8-3850, the quad-core A8-5600K Trinity equivalent comes in with a slightly lower 3DMark score and that’s a good thing for enthusiasts to note – it will retail for the same price as the A8-3850, but serve up more performance thanks to the unlocked multiplier. It looks like AMD will use overclocking as the main value-add in addition to improved performance on both the graphical and processing side of things. The higher scores are thanks to chip design enhancements and a more efficient graphics core. But don’t take these as the final results – Catalyst 12.7 improves performance in AMD’s GPUs in many ways and more driver improvements and software adjustments made to Trinity may take place in the months leading up to its Q3 2012 release to the public.
What’s really going to bake your noodle is how Trinity behaves in dual-core mode. Currently there are only two quad-cores and one dual-core in Tom’s labs but it seems that disabling or cutting out the second module completely halves performance. On the one hand, it shows that Trinity’s scaling is far more pronounced than it was in Llano – with Piledriver cores, AMD’s planned 10-core, five-module processor will certainly have enough chops to run anything you throw at it but there will be a bandwidth limitation somewhere along the line. The performance jump is as expected moving from a dual-core to a quad-core, but I’d hazard against anyone getting a dual core if you’re even expecting to play some games on the rig. Rather stick your money to a Intel Pentium based on Ivy Bridge or at least move up to a triple core Trinity A6. You’ll at least get most of the performance back and a unlocked multiplier would help a lot in the performance department as well.
You’ll also notice the 2.9Ghz A8-3850 Llano chip performing quite well in the rest of the benchmarks – that’s no mistake, the difference is really quite fine in the productivity benchmarks, where muscle outranks IPC improvements and better thermals. Yes, somewhere things got messed up and an older design is more efficient clock-per-clock than the new kid on the block. Its debatable that any desktop user would aim for that 3.8Ghz quad-core because right now it looks more like a marketing throw, a high number to draw buyers in. In single-thread benchmarks performance should be up by quite a bit, but I’ll wait for the retail chips to be reviewed before drawing any conclusions. The rest of the productivity benchmarks aren’t worth showing because the results are very close to what you see above. In general, the Llano APU trails the two new quad-cores, but performs 50% better than the new A6 in multi-threaded benchmarks.
I’ve always loathed APU reviews because performance with a discrete GPU is never, ever shown. I”m not likely to be interested in gaming performance on the built-in GPU; that’s likely what it was designed for, but being a fan of good graphics performance I’d at least like my games to have some visual impact. That said, the gaming benchmarks do point out something rather interesting – performance with the integrated GD7660D on the highest-SKU, the A10-5800K, is rather good. Pair that with a Radeon HD6670 in, say, a console-like box and you’d have very playable 1080p performance at medium quality settings. Is this what we could expect from consoles in the future? I know many ITX builds that could benefit from good 1080p performance in a small package. Hell, a cheaper Trinity-based Alienware X51 desktop would easily fetch less than the $799 asking price the entry-level Core i3 and GTX540-packing SKU has inside.
In the move from 720p to 1080p, performance in titles like Batman: Arkham City just about halves when running at the low quality preset. It looks like the integrated GPU suffers a memory bottleneck, where more memory would help performance at higher resolutions. A stronger helping hand in the form of a discrete card with GDDR5 RAM would improve performance. considerably. World of Warcraft has always been processor bound throughout the years with certain versions of the games becoming GPU-bottlenecked up to a point as quality settings were raised. It bodes well for isometric-view games like WOW and Diablo III, but perhaps it won’t be enough for more demanding titles like GTAIV or Starcraft II, where graphics takes a backseat slightly to processor performance.
Moving to the more demanding Elderscrolls: Skyrim and Diablo III, the GPU does show improvements over the outgoing Llano processor, with the improvements getting smaller and smaller as we move up to 1080p. On average the performance improvement is 15%, but its not exactly spectacular. Both these games favour a faster processor and the single-thread performance doesn’t matter much as both are multi-threaded. As before, a Radeon HD6670 would definitely help things here, running in tandem with the GPU on the APU’s die.
Finally, OpenCL support is slowly being supported in more and more apps across the desktop sphere. OpenCL uses the built-in graphics acceleration to accelerate certain tasks like compressing files or encoding movies to different formats. However there’s no benefit using the old Llano chip over the new Trinity family. In reality they perform the same with OpenCL, showing a slight decrease in performance even though its a better design. As before, I think that driver support still has to be worked on, so things aren’t completely sorted.
While Trinity chips are available in OEM machines from companies like Acer and HP, retail availability is only going to be in the next three months as third-party manufacturers get their FM2 motherboards sorted. Trinity is an improvement, but is only one that’s visible under benchmarks. In reality you’d be hard-pressed to pick out which is which in a blind test. So its worth the update, but hold onto your socket FM1 and socket AM2+/AM3 Phenom II-running boards for now.