The staggered birth of Kaveri has been an interesting story to cover but it has been difficult to keep all the pieces right in the forefront of memory. The initial launch in January 2014 saw a small number of SKUs such as the A10-7850K and the A8-7600 at first and since then we have had a small trickle at a rate of one or two new models a quarter hitting the shelves. We've seen 65W SKUs, such as in the form of the A10-7800, which offer 45W modes as well. Today we're reviewing the most recent Kaveri processor to hit the market, the A8-7650K rated at 95W and officially priced at $105/$95.

AMDs APU Strategy

Integrated graphics is one of the cornerstones of both the mobile and the desktop space. Despite the love we might harbor for a fully discrete graphics solution, the truth of the matter is that most people and most places still have that integrated platform in both consumer and business. Whenever I meet with AMD, the question from them is always simple - when you build a system, what would you get from AMD/Intel at a similar price point? The APU series tackles the sub-$200 price bracket from head to toe:

CPU/APU Comparion
AMD Kaveri Amazon Price on 5/12
Intel Haswell
(4C/4T, 88W)
3.5-3.9 GHz
HD 4600
    $199 i5-4590
(4C/4T, 84W)
3.3-3.7 GHz
HD 4600
    $189 i5-4460
(4C/4T, 84W)
3.2-3.4 GHz
HD 4600
3.7-4.0 GHz
512 SPs
(2M/4T, 95W)
$140 i3-4330
(2C/4T, 54W)
3.5 GHz
HD 4600
3.5-3.9 GHz
512 SPs
(2M/4T, 65W)
3.4-3.8 GHz
384 SPs
(2M/4T, 95W)
$120 i3-4130
(2C/4T, 54W)
3.4 GHz
HD 4400
3.3-3.8 GHz
384 SPs
(2M/4T, 95W)
3.1-3.8 GHz
384 SPs
(2M/4T, 65W)
$96 Pentium G3430
(2C/2T, 53W)
3.3 GHz
HD (Haswell)
3.7-4.0 GHz
X4 860K
(2M/4T, 95W)
    $70 Pentium G3258
(2C/2T, 53W)
3.2 GHz
HD (Haswell)
3.5-3.9 GHz
256 SPs
(1M/2T, 65W)
$64 Celeron G1830
(2C/2T, 53W)
2.8 GHz
HD (Haswell)

I first created this table with launch pricing, and it had some of the APUs/CPUs moved around. But since the release dates of these processors varies on both sides, the prices of individual SKUs has been adjusted to compete.  Perhaps appropriately, we get a number of direct matchups including the A10-7700K and the Core i3-4130 at $120 right now. This table is by no means complete, due to Intel’s 20+ other SKUs that fight around same price points but vary slightly in frequency, but that tells a lot about each sides attack on the market. Some of AMD's recently announced price cuts are here, but for consistency our results tables will list the launch pricing as we have no mechanism for dynamic pricing.

Testing AMDs APUs over the years has provided results that these are not necessarily targeted to the high end when it comes to multi-GPU systems that total $2000+, although AMD wouldn't mind if you built a high end system with one. The key element to the APU has always been the integrated graphics, and the ability to offer more performance or percentage of transistors to graphics than the competition does at various price points (irrespective of TDP). Ultimately AMD likes to promote that for a similarly priced Intel+NVIDIA solution, a user can enable dual graphics with an APU+R7 discrete card for better performance. That being said, the high-end APUs have also historically been considered when it comes to single discrete GPU gaming when the most expensive thing in the system is the GPU as we showed in our last gaming CPU roundup, although we need to test for a new one of those soon.

Part of the new set of tests for this review is to highlight the usefulness of dual graphics, as well as comparing both AMD and NVIDIA graphics for low, mild-mannered and high end gaming arrangements.

The A8-7650K

The new APU fits in the stack between the 65W A8-7600 and before we get into the A10 models with the A10-7700K. It offers a slightly reduced clock speed than the A10, but it is built (in part) for overclocking with the K moniker. The integrated graphics under the hood provide 384 SPs at 720 MHz, being part of AMDs 4+6 compute core strategy. The A8-7650K is designed to fill out the processor stack to that end.

AMD Kaveri Lineup
Price $140 $135 $120 $104 $96 $83 $64
Modules 2 2 2 2 2 2 1
Threads 4 4 4 4 4 4 2
Core Freq. (GHz) 3.7-4.0 3.5-3.9 3.4-3.8 3.3-3.8 3.1-3.8 3.7-4.0 3.5-3.9
Compute Units 4+8 4+8 4+6 4+6 4+6 4+0 2+4
512 512 384 384 384 N/A 256
IGP Freq. (MHz) 720 720 720 720 720 N/A 756
TDP 95W 65W 95W 95W 65W 95W 65W
2133 2133 2133 2133 2133 1866 1866
L2 Cache 2x2MB 2x2MB 2x2MB 2x2MB 2x2MB 2x2MB 1MB

At a list price of $105 (current $104), we were at a quandary with what to test against it from team blue. The Pentium G3258 sits at $72 with two cores at 3.2 GHz and HD (Haswell) GT1 graphics. The next one up the stack is the i3-4130, a dual core with hyperthreading and HD4400, but sits at $120. Ultimately there is no direct price competitor, but AMD assured us they were confident in the positing of the SKUs, particularly when gaming is concerned. Due to what I have in my testing lab, the nearest competitor to this is the i3-4330, a model with a larger L3 cache which has a list price of $138, or the i3-4130T which is a low power SKU.

New Testing Methodology
Comments Locked


View All Comments

  • jabber - Tuesday, May 12, 2015 - link


    "Yayyyy I use 7Zip all day long! "

    Said no one...ever.

    I don't even know why people still compact files? Are they still using floppies? Man, poor devils.
  • Gigaplex - Tuesday, May 12, 2015 - link

    I've been getting BSODs lately due to a bad Windows Update. The Microsoftie asked me to upload a complete memory crash dump. There's no way I can upload a 16GB dump file in a reasonable timeframe on a ~800kbps upload connection, especially when my machine BSODs every 24 hours. Compression brought that down to a much more manageable 4GB.
  • galta - Tuesday, May 12, 2015 - link

    So it makes perfect sense for yoy to stay with AMD...
  • NeatOman - Wednesday, May 13, 2015 - link

    I use it everyday :( rocking a FX-8320@4.5Ghz for the last 3 years.. I picked it up for $180 with the CPU and!! Motherboard. I was about to pick up a 3770k too, saved about $200 but am about 15-20% down on performance. And if you're worried about electrical cost, you're walking over dollars to pick up pennies.

    I do it to send pictures of work I do, and a good SSD is key :)
  • UtilityMax - Tuesday, May 12, 2015 - link

    If you look at the WinRAR benchmark, then that result strongly suggests that WinRAR is multi-threaded. I mean, two core two thread Pentium is clearly slower than the two core but four thread Core i3, and quad-core i5 is clearly faster than Core i3, and Core i7 with its eight threads is clearly faster than Core i5. Hence galta's comment that AMD FX with 8 cores is probably even faster, but he says that this is not normal usage.
  • TheJian - Thursday, May 14, 2015 - link

    There is an actual checkbox in winrar for multithreading for ages now. ROFL. 95% of usenet uses winrar, as does most of the web. That doesn't mean I don't have 7zip installed, just saying it is only installed for the once in 6 months I find a file that uses it.

    You apparently didn't even read what he said. He clearly states he's using winrar and finds FX is much faster using 8 cores of FX in winrar. You're like, wrong on all fronts. He's using winrar (can't read?), he's using FX (why suggest it? Can't read?) AND there is a freaking check-box to turn on multi-threading in the app. Not sure whether you're shilling for AMD here or 7zip, but...jeez.
  • galta - Saturday, May 16, 2015 - link

    Last AMD CPU I had was the old and venerable 386DX@40Mhz. Where any of you alive back in the early 90s?
    Ever since I've been using Intel.
    Of course there were some brief moments during this time when AMD had the upper hand, but the last it happened was some 10 years ago when Athlom and its two cores were a revolution and smashed Pentium Ds. It's just that during that particular moment I wasn't looking for an upgrade so I've Intel ever since.
    Having said that, I have to add that I don't understand why we are spending so much time discussing compression of files.
    Of course that the more cores you have the better, and AMD happens to have the least expensive 8 core processor on the market, BUT most users spend something like 0.15% of their time compressing files, making this particular shinny performance irrelevant for most of us.
    Because most of other software does not scale so good in multithreading (and for games, it has nothing to do with DX12 as someone said elsewhere), we are most likely interested in performance per core, and Intel clearly has the lead here.
  • NeatOman - Wednesday, May 13, 2015 - link

    Truth is the average user won't be able to tell the difference on a system with a i3 running on a ssd and a A6-7400k on a ssd or even a A10-7850k which would be more direct competition to the i3. I build about 2-4 new Intel and AMD systems a month and the only time I myself notice is when I'm setting them up, after that they all feel relitivly close in speed due to the SSD which was the largest bottleneck to have been overcome in the last 10 years.

    So Intel might feel snappier but are still not much faster in day to day use of heavy browsing and media consumtion as long as you have enough ram and a decent SSD.
  • mapesdhs - Tuesday, May 12, 2015 - link

    Ian Cutress wrote:
    > Being a scaling benchmark, C-Ray prefers threads and seems more designed for Intel."

    It was never specifically designed for Intel. John told me it was, " extremely
    small program I did one day to figure out how would the simplest raytracer program
    look like in the least amount of code lines."

    The default simple scene doesn't make use of any main RAM at all (some systems
    could hold it entirely in L1 cache). The larger test is more useful, but it's still wise to
    bare in mind to what extent the test is applicable to general performance comparisons.
    John confirmed this, saying, "This thing only measures 'floating point CPU performance'
    and nothing more, and it's good that nothing else affects the results. A real rendering
    program/scene would be still CPU-limited meaning that by far the major part of the time
    spent would be CPU time in the fpu, but it would have more overhead for disk I/O, shader
    parsing, more strain for the memory bandwidth, and various other things. So it's a good
    approximation being a renderer itself, but it's definitely not representative."

    As a benchmark though, c-ray's scalability is incredibly useful, in theory only limited by
    the no. of lines in an image, so testing a system with dozens of CPUs is easy.

    Thanks for using the correct link btw! 8)


    PS. Ian, which c-ray test file/image are you using, and with what settings? ie. how many
    threads? Just wondered if it's one of the stated tests on my page, or one of those defined
    by Phoronix. The Phoronix page says they use 16 threads per core, 8x AA and 1600x1200
    output, but not which test file is used (scene or sphfract; probably the latter I expect, as
    'scene's incredibly simple).
  • Ian Cutress - Tuesday, May 12, 2015 - link

    It's the c-ray hard test on Linux-Bench, using

    cat sphfract | ./c-ray-mt -t $threads -s 3840x2160 -r 8 > foo.ppm

    I guess saying it preferred Intel is a little harsh. Many programs are just written the way people understand how to code, and it ends up sheer luck if they're better on one platform by default than the other, such as with 3DPM.

Log in

Don't have an account? Sign up now