I must confess that over the past few years, I’ve found the CPU market to be considerably less interesting than the fast paced and rapidly changing GPU market; and for a number of reasons. Until recently, games have relied heavily on just a single core for performance – which is changing now Mantle is out, and with DX12 and Vulkan looking on the horizon, and secondly the market hasn’t really seen a really big shakeup.
But now Windows 10’s release is tantalizing close, I jumped at the opportunity to review an AMD FX 8350 processor. In this review, of course we’ll be looking at single thread performance (as, it’s still important) but with a focus on the future – in other words, how does the processor fair on heavy parallel workloads – such as DX12’s API overhead testing, 3d rendering across multiple cores, compression and so on – and how does it compare against Intel?
An Introduction to the FX 8350
At it’s heart, the FX 8350 Vishera is based on the Piledriver architecture, a successor to Bulldozer. The chip is built on a 32nm process (by GlobalFoundries). A complete Vishera processor measures 315 square mm, and packs in 1.2 billion transistors. The Piledriver architecture is based on ‘modules’, with each module comprised of two Integer Cores and one Floating point core. According to AMD, about 90 percent of the ‘average’ desktop calculations will be Integer based. Remember, computers operate on pure maths, and Integers are whole numbers, so for example, “2” is an integer, while something like 2.13414 would be a float.
So, in the of the FX 8350, the processor handles eight threads and runs at 4GHZ (assuming no overclocking). AMD have opted to include 8MB of level 3 cache, and 1024 Cache (KB/core) for the level 2 cache. The FX 8350 is fully unlocked, and just like Intel’s ‘k’ series (for example, the 4670K) signifies you’re free to manipulate the processors base clock several ways – including raising or lowering the processors internal multiplier.
What is the FX 8350’s Target Market?
If you’re buying AMD’s FX lineup for multithreading, is it a good purchase? AMD’s pricing for their products has been rather aggressive of late – with their R9 2xx series cards reduced heavily in price, and currently the FX8350 is available on Amazon for about 127 pounds, it makes a very compelling case for itself compared to Intel’s £178 I5 4690K.
AMD have made it rather clear they’re targeting Intel’s I5 range (both in pricing and their reviewer’s guide). With the new generation of API’s out, it’s actually helped AMD’s processors become good choices for gamer’s looking to upgrade this year and have their rigs last for some time, particularly if they’re on a budget, and also would like to do content creation (for example).
Review Setup and Notes
As mentioned in the opening paragraphs of this review, we’ll be focusing primarily on Multi-threaded performance, but naturally we’ll include tests designed to stress a single core too – including games. We’ll be running Windows 10 pre-release, since it grants us access to DX12, and DX11 performance is identical to older versions of Windows. For the rest of the PC, we’re running a Gigabyte 990FXA-UD3 motherboard (it’s a really nice bit of kit for the money, review coming soon), 8GB DDR3 1866MHZ and as for software – all games and software are their latest patched version.
We’ll also be running the default Heatsink and Fan for this build, which does a pretty respectable job of cooling the processor. We’ll tinker with overclocking in this article, but we’ll focus on it in another article and choose a more robust cooling solution then.
Comparison systems – we’ll be using a couple of comparison systems – based around Intel’s I7 4770K. Unfortunately, because of time constraints we’d not had time to redo some of the 4770K tests which were done when the CPU was overclocked to 4.2GHZ, so please bare in mind that Intel’s platform technically has a slight clock speed advantage in these tests.
While some people buy a high powered PC for just gaming (you lucky soul), many use their PC for work too. With that in mind, we’ve put together a series of benchmarks which primarily focus on multi-threading performance.
Starting out with 7-zip, which is used for compression or decompression. It’s highly multithreaded, and focuses on Integer workloads (which as we’ve discussed above, is the Vishera’s speciality); all while removing the IO system from the equation. The built in benchmark uses a 32MB test as a ‘standard’ and thus the higher the MIPS the better. The FX-8350 reached 23272 MIPS, slightly below that of a 4770K (at 4.1GHZ) which scored 23930 in this test.
|AMD FX-8350||23272 MIPS|
|Intel I7 4770K (OC to 4.2GHZ)||23930 MIPS|
|Intel I5 2500K||14392 MIPS|
Sisoftware Sandra 2013
Processor Arithmetic – I think it’s pretty self explanatory: how fast can the processor calculate maths. A similar story as above – the FX sits right between Intel’s two popular choices.
|AMD FX-8350||76.72GOPS / 105.75GIPS|
|Intel I7 4770K||94.33GOPS / 123.78GIPS|
|Intel I5 2500K||50.8GOPS / 87.2GIPS|
And now Sisoft’s Sandra’s Multi-Media benchmarking:
|AMD FX-8350||321.69 Int X16 / 200.42 Floating Point / 112 / Double X8|
|Intel I7 4770K||401.11 INT X16 / 393.98 Floating Point / 226.11 Double X8|
|Intel I5 2500K||157.4 INT X16 / 198.9 Floating Point / 113.9 Double X8|
Cinebench Version 11.5
|Intel I7 4770K||8.12|
|Intel I5 2500K||5.4|
CineBench Version 15
|Intel I7 4770K||791|
|Intel I5 4690K||592|
A popular ray tracing application used by artists to create beautiful and realistic 3d images; also used by reviewers who need to test out a processors ability to render said images. The application is extremely processor intensive, and in this instance we used the standard, built in benchmark but set it across multiple threads.
|Intel I7 4770K (OC to 4.2GHZ)||1535.4|
|Intel I5 2500K||1011.0|
3DMark API Overhead Testing
I’d love to include DirectX 12 games in this benchmark (but can’t, because they’re not released yet…) but 3DMark is the main reason we opted to use the Radeon R9 290X for our testing. We’ve conducted extensive DX12 API overhead testing and we saw what happens when you’ve multiple CPU cores to push data to the GPU. In a nutshell, the test asks the CPU to send a load of draw calls to the graphics card, and the GPU will then be asked to draw a bunch of unique blocks on screen as quickly as it can. There are a few oddities in this test however – for example, in the test we’d noticed that I7 4770K scores higher with a GeForce GPU when HyperThreading is DISABLED. The reverse is true of the R9 290X, hence the reason we’re using it for the FX-8350.
The FX 8350 scores 14,622,913 Draw Calls at default clock speed (compared to 17,600,402 for an Overclocked I7 4770K, OR 14,000,385 for an overclocked 4670K). It’s still a little early to draw conclusions with this at the moment, because the drivers are still in beta stage – but the results are interesting.
We run the games / tests at their highest settings, but run at only 1920×1080 to ensure that the GPU isn’t the limiting factor. The primary goal of course is to see how the FX handles multi threading tasks thanks to Mantle, but we also throw in a few single threaded games too, just for the sake of variety.
While FireStrike is certainly more interested in punishing your GPU, we’ve included it in this little test just for the sake of completeness. The graphics tests of the FX hit 59 and 45 FPS compared to 57 and 43 of the I7, which demonstrates what many already knew – when your system is GPU bound, additional CPU performance won’t make the difference.
From a personal standpoint, I loved Tomb Raider, despite wishing it had just a few more puzzles. For the benchmark, we ran everything at the highest settings at 1080P, with TressFX enabled – but we disabled any in-game anti-aliasing. The end result is both the I7 4770K and the FX-8350 are within just a few frames of each other, the I7 reaching an average of 77.9 and the FX-8350 hitting 81.2. Such small differences are mostly likely coming down to improved driver revisions and the margin of error.
There was a lot of criticism with the launch of Square’s Sneak’em up, gamer’s didn’t appreciate the smaller and tighter levels, instead preferring the greater freedom and flexibility of the original series. Using Mantle, AMD’s FX manages to achieve 81.7 FPS as an average, compared to the I7’s 96.4 FPS.
AMD FX-8350 Conclusion
The FX-8350 is a compelling piece of kit, particularly because of the price. It’s single threaded performance isn’t quite up to the levels of Intel’s (something Zen will hopefully fix) but it’s multithreaded performance is damn impressive. AMD’s CPU makes a great case for itself, coming in considerably cheaper than Intel’s I5 range, and performing slightly better in many creative applications. For DX11 games, it’s clear that Intel’s CPU’s have an advantage in lower resolutions, but when an application punishes the GPU (such as a game running at higher resolution or say a benchmark) the GPU then becomes the limiting factor.
I’ve no qualms in recommending the FX-8350 at all, and we’ll be putting out another article on how to overclock the processor soon too. We’ll also be using the system for one of our main work and capture systems, replacing an older Intel model (if that’s a recommendation or not… well, only you can decide).