AMD Kaveri Review: A10-7850K, A10-7700K and A8-7600
Overclocking
Overclocking on Kaveri brings very impressive results providing you focus on the GPU part. In our overclocking analysis we pushed the GPU part to the most we could and then re-ran all our GPU intensive benchmarks to see what kind of gains we could achieve. Below is a summary of the results we achieved but for the entire overclocking analysis and results discussion please see the full article here.
“AMD’s Kaveri A10-7850K is a solid overclocker. While we settled at 1028MHz on the GPU we were able to go another divider higher with slight bumps in voltage and the enthusiastic overclocker might even be able to go a couple more dividers up into the 1100-1200MHz region. What a difference the 308 MHz extra made. Pairing up our overclock with fast RAM saw us making large gains in most GPU-centric applications. Most games scaled well and we saw around 10% extra performance compared to stock GPU clocks. Best of all this overclocking only added about 20W to the total system power consumption meaning the A10-7850K with 2400MHz RAM and a hefty GPU overclock consumed less power than the stock last-generation A10-6800K with 1866MHz RAM – very impressive stuff indeed.” More here.
PCMark7 didn’t show very substantial gains as it is quite CPU-centric despite having a graphics segment to the test.
PCMark8 showed some very healthy gains thanks to having both a gaming segment to the test and offering OpenCL acceleration.
Basemark CL showed the largest gains of all the benchmarks because compute scales well with frequency increases. Anyone using the A10-7850K for compute applications will benefit hugely from overclocking, though you’ll need to ensure the overclock is rock-solid stable if you’re using it for work-based applications.
Compubench is another compute based application that shows excellent scaling due to AMD’s strong performance in OpenCL.
Luxmark showed very good scaling though when you add the CPU into the mix the scaling seems to be less impressive.
Photoshop CC performance was already ridiculously fast so I didn’t expect to see much of a change with the overclock, we saw about a 1 second improvement but as its within margin of error it really is too close to call.
Musemage loves AMD GPUs and so I’m not surprised to see a drastic jump in performance from the 42% frequency overclock. If you use Musemage for photo editing you can see the huge productivity potential here from overclocking.
In LibreOffice we saw around a 20% reduction in calculation times using OpenCL GPU acceleration so the benefits are substantial. Again I cannot stress enough how important I think this will be for AMD’s next generation FirePro APUs.
3DMark11 showed a significant jump of almost 30% – very impressive numbers indeed.
Into 3DMark and we saw a similarly high boost of about 25% which bodes well for gaming.
Game scaling perhaps wasn’t as impressive as 3DMark led me to believe it would be. Scaling was typically 10% or less and it’s easy to see that most games scale more from making the jump from 1866MHz RAM to 2400MHz RAM than they do from having a 720MHz to 1028MHz overclock – which to me does seem a bit strange but is easily explained. It would imply that the APU is still memory bandwidth limited at 2400MHz so overclocking makes little difference, this also explains why compute applications scale so well because they are not that bandwidth dependent and can take advantage of the increased GPU clock speed.
Power consumption was a strong point of the Kaveri A10-7850K. Even by lumping a 42% overclock onto the GPU the power consumption with a realistic combined system load only jumped 21 Watts. On a purely GPU load it jumped 31 Watts, which is a 40% jump so nearly aligns perfectly with the 42% frequency increase. However, with the overclock the A10-7850K was still more power efficient on a combined workload than the last generation A10-6800K APU (at stock clocks on both the CPU and GPU with slower DDR3-1866MHz memory).
may as well brand the A10 as the 720p chip, if your happy to game at that rez, still waiting for a capable APU that can play games at 1080p, the 512 cores just isn’t enough on its own, the chip needs around 650-1,000 cores, also, ram speed is critical for future apu performance, the quicker amd move to a ddr4 chipset, the better
Yes RAM is critical! We will be adding more about RAM with our RAM scaling featured coming up. Watch this space 🙂
excellent, also, love the way which the article summed up the cost of cpu+gpu V apu, it hits the nail on the head why you should buy an apu, however, I feel that AMD are missing trick here, a 1000 core apu with ddr4 support would allow many to build a slim steam machine that could compete with the consoles, as it is, you still need a gpu to provide an apu system with playable frame rates at 1080p, and overclocking an apu with a discrete gpu is no easy task, hybrid crossfire gives an apu system a much needed boost, as does ram speed, but it still leaves a lot to be desired, and the cost of a crossfired apu system v performance from an i5+discrete gpu puts many off, AMD need to make a huge leap in performance for the apu’s, I know why these chips are aimed at the budget segment, its to stop Intel grabbing market share, plus, a more powerful apu would probably eat into AMD’s own gpu business, I wish they showed more courage, and give us the one chip solution that the APU’s originally promised…
Could you maybe look at crossfire performance for the A10 ?…..can you get smooth 1080p gameplay ?, and is the cost V performance worth it ?
I am currently doing hybrid crossfire testing right now. Will also be doing overclocking and the RAM scaling I mentioned. We just wanted to get the base review out first. We will be adding the other bits over the coming week or so.
more tech goodness, thats great Ryan, cheers
1600x900p is the sweet spot for APUs right now.
DONT THINK ITS MADE FOR GAMERS… but for people who game on facebook pogo nad things like that will be fine
Did you even read the review? Kaveri is quite capable of playing games. Most people don’t have lots of $ to chuck at graphics cards and expensive processors, in those instances an AMD APU offers a great alternative, especially the A8-7600.
you can do a bit more than just FB bud, 720p performance is fine, as long as you keep to medium or low settings
Exactly, I’ve got a A8 APU in my girlfriends rig, play Resident Evil 5 at high-ultra graphics, 1080p and 30fps, not bad for an £80 chip IMO.
Completely agree, price for performance as absolutely outstanding
guess my insights were wrong game on
i seen a demo of the kaveri APu playing Battlefeild 4 on 720p with medium details it hand handle some games pretty well
I hate to say this, but even though I am a part of the kaveri’s intended market (which is *poor* entry level gamers and HTPC users), I don’t see why you would buy one. The Richland 6800k does comparably well to the 7700K but costs $30 less. I’m a poor gamer so that $30 does mean something to me. With the Kaveri’s dual graphics not coming anywhere close to the performance of the $120 260x, The new APUs don’t even make sense in the long run because you might as well pay for a 260x instead of spending that money on the 250 or 240. Power consumption vs. performance only improved significantly on the 7600, but that hasn’t released yet so the kaveri doesn’t really win in any case.
Well, I stand corrected on 3 issues so I guess I’m not really an expert like I thought I was. At first I didn’t see the power consumption chart on page 12, where both of the Kaveris in dual graphics mode beat the 6800k in power efficiency. I simply went by the TDP of the APUs (which dropped by 5W between generations). Then, I saw Newegg has $150 7700ks, so the margin between it and the Richland is now $20. Lastly, I learned that the Richland APU is really just an overclocked Trinity ($120), so the margin is still $30. Oh well, i guess I’m saving $10 because the power consumption doesn’t matter to me whatsoever.
As a reference, to convert a thermal margin to a processor temperature, you take the processor’s rated TDP and substract the thermal margin. So for a processor with a 95C TDP, the margins you posted are roughly equivalent to the temperatures measured. 95 – 57.9 = 37.1C for an A10-7800K at load. I think the reason they cannot specify the temperatures is because the TDP is configurable on those chips. Also keep in mind the temperature will likely rise dramatically with the GPU also on full load, because they both use the same cooling device. The stock fan will probably not handle the CPU and GPU at full load in turbo mode for too long, so I wouldnt cheap out on the cooler on these things.
Well this would have to be one (& only) of the most honest reviews about AMD ‘s Kaveri.
And with all the bias against AMD, its about time someone did a solid review about it’s latest APU.
Showing it to be a performer that exceeds all expectations with built in technology that’s starting to show even greater promise.
You forgot about ram speeds with a Pentium+250X you only need 1333mhz ram vs the high speed ram for the APU’s plus you forgot about the differences in performance from DDR3 to GDDR5 on the 250X
Trust me, there are a lot of things I could have included, I didn’t want to write 100,000 words. But yes you’re right.
just add 4 more cores then it beat haswell 4 cores but amd is doing the mainstream architure tweak first then release to FX version last if proves it can hurt haswell 8 cores monster..and steamroller is not it but its huge improvenment than piledriver in 4 cores alone….the last generation excavator is the fix and improve version of bulldozer..and 8 cores can mess intel haswell-e very well at the end even it took them awhile to come around from the back of intel …since carrizo apu has 4 way l1 data and 3 way l1 inst. code its only 3 modules with 6 cores total and 20nm is very small and less power than haswell-e………combined with a dual carrizo with 8-way l1 data, 6-way l1 inst. code total in last FX version with it can fit up 3-6 modules in that processor with 6 and 12 cores from 95 watts to 125 watts and 220 watts for 4.0ghz to 5ghz…
“In terms of temperatures with Kaveri it isn’t possible to judge. The temperature readouts on the Kaveri APUs are totally “borked” and the motherboard sensor readouts give ridiculously low temperatures which cannot be accurate – 37 degrees at full load for the A10-7850K? I think not.”
But you think that Haswell processors need more than 10W less at idle because of “the incredibly power frugal nature of the Haswell design and the advanced idle sleep states.”? Really funny. Dear author, please always try to use your brain. The difference between AMD and Intel at idle is hardly measurable because both need only a few watts then (normally less than 5W). It should be clear that a difference of 10W or more is only possible if the motherboards draw different amounts of power. Or do you think every motherboard draws the exact same amount of power, regardless of the processor?
Pure idle comparisons mostly don’t make much sense nowadays due to very advanced power saving mechanisms. We don’t have P4s anymore which really could draw a lot of power at idle. At least as long as you can’t guarantee the same power draw of the motherboards you won’t get meaningful values to compare.
Nice review!.
the concept of APU for gaming maybe didn’t look that interesting for western buyers where you could get better price/performance with separate CPU+GPU combo, but please don’t forget us that live at third world country where the price of an i5 is more than minimum wages in many parts of our country. With that in context, the A8-7600 combined with “Steam Sales” surely the best option for people who want enter the The Glorious PC Gaming Master Race at budget. and don’t forget the low power consumption too.
yeah, i know i’m late into this.
It is quite interesting when you consider it is a CPU with a GPU integrated onto the Processor Die. I am currently Running an AMD FX-8300 @4.1GHZ. My Daughter is running on an AMD A6-6400 at 3.8GHZ dual core with 8GB 2133MHZ RAM. Performance wise, SHe can run World of Warcraft on Medium settings at 40FPS easy and watch youtube videos in the background. With her only being young I am not going to spend a fortune on her first PC. These AMD APU’s are perfect and hit the right spot in bang for buck. people always tell me “for £100 more you could have had an Intel I7 4790k instead of the FX-8350” but in doing some research the AMD FX-8350 gave me the best bang for buck coupled with the R9 270X 2GB card
i was looking at my country’s local pc stores for the 480 amd vc which is supposedly 225 USD for the 8gb version but i was shocked to find it sellin 50% higher than the recommended price. then i compared prices of low to midend video cards to their US counterparts and i was saddened to see the same 50% markup from the recommended prices in the USA.
No wonder a lot of people here are just depending on IGP and APUS to play their games.
I settled for the AMD 7890k and got my self 16gb of memory since i cant get a price performance ratio for an intel cpu plus graphics card (overpriced in my country) that is equal to what other countries get.
The 7890k turned out to be the best option for me.