AMD A10-7850K processor review

We’ve been talking about the Kaveri technology in this A10-7850K for a long time now – about the promise of those new Steamroller cores, the addition of Graphics Core Next GPU cores, and the still rather intangible benefits of the heterogeneous system architecture. Now it’s here, in its shiny new heatshield, with those copper pins gleaming in the light of a new day, we’re still, in some ways, waiting for it to arrive.

The initial performance of this new, advanced silicon is somewhat immature and we’re still waiting for real-world instances of HSA software to take advantage of the new compatibility that’s been built into Kaveri. So despite the actual hardware being ready and available, the ecosystem as a whole just doesn’t seem to be ready. It’s a shame because we wanted to be jumping into our Kaveri testing with enthusiasm for AMD’s resurgence as a fighting force in the PC processor market.

Let’s just recap what that hardware is. The A10-7850K is rocking new, upgraded Bulldozer cores. Codenamed Steamroller, this update aims to fix a problem that’s always been holding back the AMD CPU architecture – weak individual core performance.

Steamroller gives each of those cores more of its own resources so it can operate more effectively as an individual unit, while still using its multi-threading abilities. This top Kaveri component is utilising a pair of Steamroller modules, effectively making it a quad-core part.

Next: Graphics!

The other half of the equation is the graphics. Well, not quite half, but at 47% of the Kaveri die being given over to Graphics Core Next silicon it’s a healthy boost over both its predecessor, Richland, and the latest Intel Haswell processors, which both dedicate less of their dies to graphics.

The move from the older VLIW4 graphics architecture to the modern GCN tech is important too, and, combined with a process shrink from 32nm down to 28nm, means we’ve got a rather impressive-looking 512 GCN cores inside the unassuming APU. Richland’s A10-6800K only had some 384 of the older cores inside, so Kaveri’s graphical power should be well in advance.

As for the CPU part though, AMD has decided to set the A10-7850K’s GPU clockspeed lower than its predecessor; here it’s 720MHz versus 844MHz, which offsets the performance gains to some degree. This is something that really confuses us – especially given that this is AMD’s top desktop APU part, and not a chip that really needs to fit inside a particular power envelope.

For all our talk of four CPU cores and 512 GCN cores though, AMD doesn’t want us to think about its APUs like that. AMD wants to talk about compute cores. Under that moniker it counts each CPU core as a compute core, and on the graphics side each compute unit gets the same treatment. The compute unit is made up of 64 GCN cores, giving the A10-7850K eight of them. In total, that means AMD wants to denote this APU as a 12-core part.

That might seem like marketing donkey dung, but in light of the software advantages of the new Kaveri architecture, it makes sense. The whole idea is that HSA software will essentially be blind to the CPU/GPU divide, and will just use whichever compute core is best suited to its task.

That’s possibly the most exciting part of Kaveri, and it could be a killer app for the new APU. The same can be said of the Mantle API that AMD is working on. That can allow for the APU to be used in conjunction with whatever discrete GPU you use to offset workloads like particle effects to boost gaming performance.

Present tense

But what about right now? This A10-7850K is not delivering the sort of performance we want from all the new tech that AMD has dropped into the APU. You can pick a bunch of synthetic benchmarks that focus directly on some key parts of the silicon alone, and you can play with a selection of early HSA tests that prove its eventual benefits, but in the real-world it’s not as big of an advance over Richland as we were hoping.

That can be partially explained away by the drop in clock speeds across the board. The A10-7850K is rated at 3.7GHz, Turbo-ing up to 4GHz when possible. We didn’t see it manage 4GHz on either of our testing motherboards, with the latest beta BIOSs direct from AMD. Richland, however, is happily running at 4.4GHz whatever CPU tests we throw at it. In Cinebench, the Kaveri chip is generally running at around 3.8GHz at best – that’s some 600MHz slower, and almost kills any advantage the Steamroller cores might offer.

We ran the A10-7850K at 4.4GHz though and saw a 10% boost in Cinebench, but that’s still a way off the 30% boost that AMD has been claiming. Gaming is where we see most of the performance boosts at stock speeds though, with the A10-7850K offering a 10-15% improvement over the A10-6800K in the CPU-intensive Battlefield 4. It still can’t keep up with Intel in that department, with even the Core i3-4130 beating Kaveri across the board when using a discrete GPU for gaming.

For the integrated GPU it’s a different story. The GCN graphics of Kaveri are a good deal quicker than the HD graphics of any of Intel’s desktop chips, and are actually capable of delivering a good 1080p gaming experience. It’s around 10% quicker than its Richland predecessor, with a slower clockspeed on the graphics cores, too. You can push that above 20% with a little light overclocking of both the system memory and GPU cores – we were hitting almost 40fps in GRID 2 at 1080p, on high settings with 4x AA.

So, where does that all leave us? Paired up with a discrete graphics card – and with nothing currently running the Mantle API – the A10-7850K is a difficult sell when it’s so close to the Core i5-4570 in price, and motherboard costs aren’t much different between H87 and A88X chipsets. Intel still has dominion over gaming PCs while the software still favours its hardware, but Kaveri is genuinely an option for an integrated system, sans discrete card. It’s faster than competing Intel GPU silicon and has a decent advantage over older AMD APUs too.

But at the moment we still can’t recommend it. It simply isn’t ready. The promise of both HSA software and Mantle is huge and tantalisingly close, but those are going to power the killer apps that make the AMD APU a really viable alternative to Intel in most markets. At the moment the integrated arena is the only place that AMD can claim a definite lead.

Price: $249

AMD A10-7850K Comparison and Performance Benchmark Results


AMD A10-7850K vs A10-6800k vs IS 4570 performance comparison.


Summary: The fastest integrated graphics around, but paired with CPU cores that are still disappointingly weak.

Value: 3.5 stars
Features: 4 stars
Performance: 3 stars

Overall Rating: 3.5 stars out of 5

  • Roy@AMD

    This is an APU, so why no mention of OpenCL? This test should use PCMark 8.2. Please.

  • monstercameron

    how can you state that “…but Kaveri is genuinely an option for an integrated system…” then go on to say “…at the moment we still can’t recommend it. It simply isn’t ready… ”
    it sounds like a great product from the review but you seem to be using your disappointment to color your conclusion.,

  • Graeme Willy

    So what does this mean in terms of CPU only? I’m going to building a Steam machine-like system, using a mini-itx board and a discrete graphics card..does the Richland still produce the best overall computational result? I want one that will not bottleneck my R9-280x. I’m confused because Steam Roller uses dedicated cores, while the Richland is Bulldozer inspired and is two cores per cache module. The Steam Rollers dedicated core setup was supposed to increase IPC…and what’s the max overclock achievable (roughly), on a Kaveri, without voltage bump?