HSA Arrives: AMD’s Kaveri Hits the Desktop

HSA arrives on the desktop…

AMD A10 LogoSome thought, myself included, that this day would never come; but AMD’s first HSA enabled APU codenamed Kaveri has finally launched. For AMD today must be an extremely gratifying event as the company’s first implementation of their HSA vision is now available to consumers. Of course the benefits of this new architecture won’t be apparent until software developers do their part, but there is still quite a lot to like about Kaveri even without HSA enabled software.

Kaver review header

In this article we’ll cover the performance of AMD’s latest APU and compare it against Richland, AMD’s 2013 mainstream desktop APU, as well as Intel’s current generation Haswell APU.

AMD Kaveri intro slide

For more information on Kaveri’s unique HSA architecture, brand new Steamroller cores, and GCN GPU cores check out some of our older articles.

AMD Kaveri history

If you’d like more background on the lead up to this launch you can check out these articles like AMD’s initial Kaveri announcement, its first delay, second delay, death, and subsequent rebirth under a new engineering team.

With the architecture and historical background out of the way let’s take a quick look at our test system.

Test System Kaveri

Our test setup uses a 256GB Samsung 840 Pro SSD in RAPID mode running Windows 8.1 with the latest firmware. We’re also using stock CPU coolers as well as a standard five fan ATX mid-tower case. Our power supply is a 550 Watt Antec Basiq unit that is 80 plus certified and our RAM of choice is an 8 GB DDR3 2400Mhz kit from AMD. As far as motherboards are concerned we will be using MSI’s FM2-A85XA-G65 for our AMD Richland platform, Asrock’s FM2-A88X Extreme6+ for our Kaveri sample, and Intel’s own DZ87KL-75K for Hawsell testing.

amd kaveri (5 of 12)

Let’s take a look at the individual benchmarks in our testing suite for a moment. Our x246 encoding benchmark comes to us courtesy of the folks at Tech ARP. The Blender test we’re using is from EofW. Microsoft’s HTML 5 benchmarks have proven themselves to be a useful indicator of both browser and platform performance. In this review we’ll be using the Chalkboard benchmark in Chrome. Cinebench R15 and wPrime results are also included in our testing.

We have two APU compute benchmarks LuxMark, and ratGPU. Our Foobar2000 test uses the FLAC encoder to process and convert Arcade Fire’s Reflector Album to lossless FLAC files. With Handbrake we’re taking a 3 minute 720P MP4 video of motherboard and converting it using the iPad preset. The SunSpider Javascript benchmark is a popular benchmark we’ll be using with the Chrome browser. Finally TrueCrypt is a disk encryption application that offers a variety of different options including hardware accelerated AES256.

UPDATE 12/4/2016 10:30 AM CST: TrueCrypt is no longer maintained as of 2014. Here’s a list of alternative apps; I personally use VeraCrypt.

For our GPU testing we used the highest playable in-game settings that we found in our article on Kaveri’s performance at 45 Watts.

All of our testing data is once again available at Mega.

Kaveri benchmarks complete

There’s a lot of data here, so let’s go through it piece by piece starting with GPU performance. We all knew going into this review that AMD would have a performance advantage in video games, what we didn’t know is that it was going to be so pronounced. The A10-7850K offers between 1.8 and 4.5 times higher framerates than the HD 4600 graphics solution on Intel’s i7-4770K. That’s a massive performance advantage that averages out across our testing suite to 2.5 times higher framerates on AMD’s hardware versus Intel’s. Compared to the last generation A10-6800K the gap is less pronounced, but still significant.

Moving to single core CPU performance we see a much rougher picture of the A10-7850K where the fastest version of AMD’s new Steamroller offers only about 60 percent of the fastest version of Intel’s Haswell chip. We can also see that the A10-6800K is marginally faster than AMD’s latest and greatest APU in single threaded scenarios due to higher clock speeds.

In our multithreaded CPU testing things look a bit better for our A10-7850K as it’s able to put some distance between itself and the A10-6800K. Unfortunately it still only offers between 40 and 70 percent of the performance of Intel’s i7-4770K.

Moving to our CPU+GPU compute benchmarks it’s clear that the switch from the old VLIW4 GPU architecture to AMD’s new GCN architecture has paid off with a major performance boost over the A10-6800K. In fact the A10-7850K’s GPU cores are so fast now that AMD’s chip is able to more or less match Intel’s performance in our benchmarks.

Looking at our overall performance numbers it’s important to note that this number is an equally weighted average of the results from our four categories of performance. How you chose to weigh the value of each of those categories is a personal preference. But it’s interesting to note that for the first time since AMD launched it’s A-series of chips do we have a scenario in which lower CPU performance is negated by higher GPU performance in our benchmarking suite.

Of course that view papers over that fact that AMD’s new Steamroller core is effectively no faster than their prior generation Piledriver core at 95 Watts. Thus despite all the things there are to like about Kaveri and the new A10-7850K; this product launch will have little to no effect on the competitive positioning of Intel’s pricing stack. AMD’s Kaveri has brought with it a big bump in compute and graphics performance that will go out to store shelves with only a small price bump.

This is great news for consumers looking for bargain because you’ll be able to acquire one of AMD’s new chips for the same price as an equivalently performing Intel chip and then enjoy the benefits of much higher GPU  performance along with the novelty of owning a HSA enabled APU. Kaveri’s a good chip, but its launch on the desktop is going to have a hard time changing the status quo.S|A

The following two tabs change content below.
Thomas Ryan is a freelance technology writer and photographer from Seattle, living in Austin. You can also find his work on SemiAccurate and PCWorld. He has a BA in Geography from the University of Washington with a minor in Urban Design and Planning and specializes in geospatial data science. If you have a hardware performance question or an interesting data set Thomas has you covered.