A10-7850K Performance Review

I think it's worth mentioning that the CPU portion is lower-clocked, 5w cooler and still pulls ahead in nearly all benches. Imagine the Richland style 'refresh' down the road. I'm more interested in overclocked figures.
 
We already know its 20% faster per clock vs. Vishera so that is all the information you'd need to get ball park or in-field estimate of performance at a given clock speed.
 
mesyn191 said:
We already know its 20% faster per clock vs. Vishera so that is all the information you'd need to get ball park or in-field estimate of performance at a given clock speed.
Click to expand...

the benches arent showing that actually.

it looks like now due to the dual instruction queues, it's going to be better on multicore, but each core stays the same really, so on single core benches it's actually slower due to lower clocks.

Here is a percent summation of how Steamroller compares to Piledriver on these benches, not clock adjusted (so, 3.7 vs 4.0, which is 7.5% less clock for Steamroller)

Geekbench 3 Multi core:

Memory -6.5 %
Integer +4.2%
Floating point +1.3%
Geekbench score +2.1%

Geekbench 3 single core:

Memory -13.3%
Integer -6.3%
Floating Point -3.7%
Geekbench score -6.5%

Cinebench CPU rendering

Multiple CPU -6%
Single CPU -9.4%

PC Mark 08

Microsoft Office score +7.6%
Overall score +10.9%



You can see if you average all those out Steamroller/Kaveri is going to be slower than Richland/Vishera. But like I said, more or less it's likely less bottlenecked on multi core so it will do better there, but each core is the same so single core decreases with clock. Not sure why the memory benchmarks are so much lower, maybe the Kaveri is artificially handicapped there somehow more testing needed.

It's actually INCREDIBLY unscientific, but as an average across those 12 benches Kaveri is 2.1% SLOWER than Richland at CPU stuff. Which it's clocked 7.5% lower, so there's your IPC improvement.

They said the clock decreased on the CPU because they moved from a CPU optimized process to a balanced APU optimized one. So if you could move the Steamroller enhancements to Vishera, the clocks would presumably stay the same and you'd just get the (supposed) 20% IPC increase, so it'd look better there. In other words an 8350 with Steamroller would stay at 4.0 ghz and just get the 20% IPC.

But it would still by the looks of it get no improvement at all in single core, which is a major weakness of the Bulldozer arch since the beginning. You'd just see the 8350 get more benefit from it's 8 cores than it does currently on multithreaded loads.
 
Last edited:
Shark974 said:
the benches arent showing that actually. Here is a percent summation of how Steamroller compares to Piledriver on these benches, not clock adjusted (so, 3.7 vs 4.0, which is 7.5% less clock for Steamroller)
Click to expand...

1)You have adjust the scores for clock difference otherwise you're not keeping things apples to apples and 2)those are mostly synthetic benches which are useless.
 
mesyn191 said:
1)You have adjust the scores for clock difference otherwise you're not keeping things apples to apples and 2)those are mostly synthetic benches which are useless.
Click to expand...

well do the calculations to adjust yourself, the info is there.

synthetic benches are not useless thats why they exist. at any rate they're all we have.

i agree we should wait for better benches and hopefully things will look better, i think they will judging by pc mark.
 
Like others I'd like to see a review when released that has both Kaveri and Richland with the same cpu clocks and memory speed across a variety of benchmarks. That should give us the overall IPC/graphics improvement.
 
AMD has been really gungho on multi-thread idealogy since the start of bulldozer.

Even jaguar went down the same path with 4 weak-cores.
 
mesyn191 said:
Hahaha passive aggressive much.


Aaaand logical fallacy. All kinds of useless crap and even straight up scams exist. Just because its there doesn't mean its legit or useful.
Click to expand...

But in the benchmarking field synthetics serve a purpose as you know well. Let us rephrase to "that is why they are used in many reviews".

Annnnd logic fail by you again :p
 
omega-x said:
AMD has been really gungho on multi-thread idealogy since the start of bulldozer.

Even jaguar went down the same path with 4 weak-cores.
Click to expand...

That works for me though. I don't really play any high end games and it's undeniable that software will eventually get with the program and move towards multi core. We are already seeing it some of the games and software.
 
Shark974 said:
But in the benchmarking field synthetics serve a purpose as you know well.
Click to expand...
What exact purpose do they serve? How exactly do GeekBench scores translate into fps when running Skyrim or encoding a video or any other software for that matter?

Shark974 said:
Let us rephrase to "that is why they are used in many reviews".
Click to expand...
They're used in reviews because reviewers get lots of requests to see them by people who don't understand much if anything about benchmarks other than "higher numbers = better". All you've restated is appeal to popularity.
 
mesyn191 said:
What exact purpose do they serve? How exactly do GeekBench scores translate into fps when running Skyrim or encoding a video or any other software for that matter?
Click to expand...
How exactly do Skyrim frame rates translate into GeekBench scores?
 
Synthetic benchmarks matter to the companies trying to sell the apu's, cpu's and video cards. It's often those metrics placed on the retail boxes for the masses and in ads. Real benchmarks matter to us geeks/gamers. Point is they both matter...
 
It doesn't really matter. There are so many metrics we look at which aren't even consequential to the performance of other applications that singling out a synthetic benchmark seems pointless. Frame rates in Skyrim, while "real-world", don't necessarily tell the story about performance in any other application (like, say, GeekBench). Even Skyrim frame rates don't tell the whole performance story of Skyrim itself.
 
If you optimize for benchmark X then say benchmark X does a gazilion billion fps/other stupid score. Then it still says nothing ;)
 
wonderfield said:
How exactly do Skyrim frame rates translate into GeekBench scores?
Click to expand...
They don't and can't. I was asking a rhetorical question. omega-x has it right.

pcjunkie said:
Point is they both matter...
Click to expand...
There is a big difference between useless marketing, which is at best what synthetic benchmarks are when used by OEM's or IHV's, and actual in application performance numbers.

wonderfield said:
Even Skyrim frame rates don't tell the whole performance story of Skyrim itself.
Click to expand...
Absolutely correct. Performance is actually very hard to quantify in a meaningful way and trying to boil a complex situation down to 1 number is usually a gross over simplification that usually does more to hurt your understanding of what is exactly going on rather than help.

That being said actual FPS from a game is still far more useful and informative than some arbitrary snythetic benchmark score.
 
Last edited:
I agree entirely that they're a more useful metric: real-world — when you can design your benchmark properly and get meaningful data out of it — trumps synthetic every time, but I'll still argue that there's potential value in synthetic benchmarks. They can sometimes point to architectural differences that manifest in performance differences in certain types of applications, but they aren't in of themselves particularly meaningful.

Still, they're data points, and synthetic scores can still correlate to real-world performance metrics (and do so often, if the tests are designed correctly).
 
The CPU-Z 1.68 update added support for other Kaveri models than the two initially to be released: Besides the A10-7850K and A10-7700K, there are also other models listed.
AMD Kaveri APUs (A10-7850K, A10-7800, A10-7700K, A8-7600, A6-7400K, A4-7300).
Intel i7-4790, i5-4690, i5-4590, i5-4460 (Haswell refresh).
Intel Celeron Haswell (G1830, G1820).
Intel serie 9 chipset (Z97).
Click to expand...
 
Last edited:
mesyn191 said:
They don't and can't. I was asking a rhetorical question. omega-x has it right.


There is a big difference between useless marketing, which is at best what synthetic benchmarks are when used by OEM's or IHV's, and actual in application performance numbers.


Absolutely correct. Performance is actually very hard to quantify in a meaningful way and trying to boil a complex situation down to 1 number is usually a gross over simplification that usually does more to hurt your understanding of what is exactly going on rather than help.

That being said actual FPS from a game is still far more useful and informative than some arbitrary snythetic benchmark score.
Click to expand...

Excuse me but that useless marketing is what sells large volumes of units and keeps companies in business. I'm not sure what you are trying to prove on your high horse but its time to get off.
 
Only benchmarks I pay attention to is Winrar and the video encoding ones as they are doing something useful. Everything else is for fun and to debate with others as to how it might relate to performance.
 
cageymaru said:
Only benchmarks I pay attention to is Winrar and the video encoding ones as they are doing something useful. Everything else is for fun and to debate with others as to how it might relate to performance.
Click to expand...

Me too. Especially x264 encoding, since i use it and i consider it the best "real world" test of multithreaded performance of a CPU. It literally puts the CPU to its knees crying for help. Well, that and also the fact that i don't know/care about the game benches, which at any case, i consider dependent on the way the company coded the game. For instance, often i have read "FX 6300 runs better in games than 1090T". Well, maybe so, but this for me has no value. In x264 the 1090T beats the FX6300 (both at stock). Barely, but it does. So it only means that the game is using some code optimization for FX and none for Phenom or something that has no real meaning about which CPU is the strongest.
 
pcjunkie said:
Excuse me but that useless marketing is what sells large volumes of units and keeps companies in business.
Click to expand...
Just because it moves product doesn't mean its good to BS people by taking advantage of their ignorance. Do you working in marketing or what?
 
tonyftw said:
Some are already on sale at newegg and tigerdirect btw. Little pricier than I expected them to be.
Click to expand...

The price leaks I found indicated >$150 price for the 7850k and ~$150 price for the 7700K.

Not good prices for the pure "game" performance of these units, but AMD should emphasize there's more than pure CPU in these units and Mantle,HSA,Trueaudio etc. is part of the price.
Just a shame you can't sum up all the CPU "extras" in a review showing X fps in game a and X:YY min of encoding time.

Speaking of which, reviews should start showing up, if the reviewers can get through all the CES articles they have on their desk.
 
there are a lot of FM2+ articles on techpowerup, i don't suppose today is launch day for Kaveri?

in which case we might expect reviews to pop up some time later today...
 
Yeah I read it the only thing that shows Kaveri power is the OpenCL bit the rest is hit and miss.

They got 1 thing right that AMD promises HSA, HQ and Huma but delivering it is another point.

The rest is so so out of pity they paired the Intel cpu with a radeon graphics card.
 
Credit to AMD: this is the first time I've ever been impressed with their performance at 65w (now the major falloff is at 45w). You used to have a noticeable drop in game performance going 65w! Part of that is due to having more GPU than they need: you can clearly see they are bandwidth-starved when the cut-down 65w parts get within a stones throw of the 95w parts in games.

Not the greatest CPU release, but impressive refactoring of the product lineup.
 
Even the 45w numbers have me hopeful for the eventual 35w mobile's.

Should be an insane bump from a a10-5750m
 
Most of the reviews so far like Anandtech are pretty useless since they're refusing for some reason to publish A10-7850K Battlefield 4 performance data. Techreport at least published it for A8-7600 45W and 65W which is near 30fps at 1080p. Great little inexpensive setup for someone who plays at lower resolution.
 
Came away with two things. For mobile this chip should be great. For a new desktop cpu this is pretty much a bust as far as expectations go. If you already have a Richland cpu stick with it...Mantle could change it a bit but you could get the same performance with a Pentium cpu and 240/250 video card.
 
The price isn't bad compared to A10-6800k for $139, but it stinks compared to the $88 a10-5800k i recently picked up.
 
You must log in or register to reply here.
Back
Top