APU Wars - Hybrid Cross Fire

[AtenRa]

This is an evaluation of Hybrid Cross Fire using the Kaveri A10-7850K in Hybrid Cross Fire mode with the R7 250 2GB DDR-3 Cape Verde GPU.

There are three different configurations I have used for this review, the APU alone using the iGPU, the APU + dGPU using only the R7 250 and the APU Hybrid CrossFire mode when both the iGPU and the dGPU are working together.

With all three methods we can evaluate the performance, perf/$ and perf/watt of each mode.

I have included the Core i3 4330 with iGPU and using the R7 250 alone for reference.
Also, the HD7790(1075MHz) 1GB DDR5(5000MHz) was used at 1080p. This can simulate an R7 260X.



Hardware and software used

Socket FM2+
APU : AMD Kaveri A10-7850K
Motherboard : ASUS A88XM-Plus
Memory : 2x 4GB 2133MHz Kingston Genesis 11-12-11 1.65V
HDD : Seagate Barracuda ST1000DM003 1TB SATA-6
PSU : Thermal Take TR2 380W 80plus Bronze

AMD Catalyst 14:12 Omega
Windows 8.1 64bit

Socket 1150
CPU : Intel Core i3 4330
Motherboard : ASUS B85E-M
Memory : 2x 4GB 2133MHz Kingston Genesis (1600MHz at 9-9-9 1.5V)
HDD : Seagate Barracuda ST1000DM003 1TB SATA-6
PSU : Thermal Take TR2 380W 80plus Bronze

Intel Graphics Driver 15.36.14.64.4080
Windows 8.1 64bit

Graphics Cards
GigaByte R7 250 2GB DDR-3, 512 Shaders, 720MHz core, 800MHz memory
GigaByte HD7790 OC GDDR-5, 1075MHz core, 5000MHz memory

Overclocking
The AMD A10-7850K was overclocked in two different modes. When the APU was working with its iGPU alone the CPU was at 4.35GHz, the iGPU at 970MHz, the memory was run at 2424MHz 11-13-13, NB was at 2000MHz and bus speed at 101MHz.
In Overclocked Hybrid CrossFire mode with the R7 250 the iGPU was lowered to 856MHz, the rest remained the same as above.

That was made because the Overclocked R7 250 was also working at 855MHz for the GPU core and 830MHz for the Memory.

So in order to understand the different modes here are the exact specs.

A10-7850K iGPU = Default A10-7850K, using only the iGPU

A10-7850K iGPU OC = Overclocked A10-7850K, CPU core at 4,35GHz, iGPU at 970MHz, memory at 2424MHz, NB at 2000MHz (using only the iGPU).

A10-7850K + R7 250 = Default APU and dGPU, using only the R7 250 and the A10-7850K as a CPU only, simulating the Athlon 860K with a dGPU.
A10-7850K + R7 250 OC = the APU CPU is at 4.35GHz and the R7 250 at 855MHz core, 830MHz memory (dGPU graphics only). Simulating an OC Athlon 860K with a OC dGPU.

A10-7850K Hybrid CF = Default APU and R7 250 in Hybrid Cross Fire mode, both iGPU and dGPU graphics.
A10-7850K Hybrid CF OC = Overclocked A10-7850K at 4,35GHz for the CPU Cores, 856MHz for the iGPU and 2424MHz memory with NB at 2000MHz. R7 250 OCed to 855MHz core, 830MHz memory. Both iGPU and dGPU graphics.


Mantle and CrossFire.
I have used three games that support Mantle, those are the Civilization Beyond Earth the Dragon Age : Inquisition and Sniper Elite III.
Only one of them can support CrossFire using the Mantle API and that is the Civilization Beyond Earth. You can read about Civ BE Mantle CrossFire in this Anandtech article.
For the rest of the games, when in Hybrid Cross Fire mode the benchmark was run in DX-11 (iGPU and dGPU only modes were run with Mantle).

Thief also suppose to have Mantle Cross Fire support but it was not working in Hybrid CF mode with my hardware, so i havent included that game in the end.

 

[AtenRa]

Bioshock Infinity

-----------------------------

Civilization Beyond Earth

 

[AtenRa]

Dragon Age : Inquisition

-----------------

Formula 1 2014

 

[AtenRa]

Hitman Absolution

-------------------

Metro Last Light Redux

 

[AtenRa]

Sleeping Dogs

-------------------

Sniper Elite III

 

[AtenRa]

Total War : Rome II

Well I think that Hybrid Cross Fire is much better today than last year. Its working without any problems in every DX-11 game but there are problems or no support at all in the majority of the Games that support Mantle (in Mantle mode). It is a pity because Hybrid CrossFire in Mantle could be a game changer. I hope with Windows 10 and DX-12 things will only get better for Hybrid Cross Fire not only in desktop but especially in Mobile.

As of now Hybrid Cross Fire is better suited for 1080p as it makes all current games playable with an APU and a low end dGPU at that resolution. The question is, when do you recommend a Hybrid Cross Fire setup. Personally in its current state, I could only use it in Laptops where the iGPUs are very week and higher performance dGPUs are very expensive.

In Desktop you can always get a cheap Quad Core Athlon and pair it with a dGPU for maximum performance and perf/$. As of now with current prices of CPU/APUs and dGPUs I could only see a use of Hybrid CF if you already have an APU like A10-7850K and you want to upgrade the graphics performance but you are on a low budget. Otherwise a faster dGPU will get you better performance and performance/price. This could change in the near future with DX-12 though but until that time my recommendation remains the above.

 

[USER8000]

Nice review!:thumbsup:

I wonder how the A8 7600 would fair with an R7 250?? They have the same number of shaders and the A8 7600 is fare cheaper than an A10 7850K.

 

[AtenRa]

Nice review!:thumbsup:

I wonder how the A8 7600 would fair with an R7 250?? They have the same number of shaders and the A8 7600 is fare cheaper than an A10 7850K.

thx,

Actually the R7 250 with DDR-3 has 512 Shaders same as A10-7850K(i will update the OP with GPUz later), the R7 250 with GDDR-5 memory has 384 Shaders same number as A8-7600.

 

[NTMBK]

Those drops in minimum framerate are a little worrying :\

 

[sm625]

Looks like overall, the hybrid CF approach eviscerates minimum framerates. I hope DX12 improves this.

 

[LTC8K6]

Will DX12 allow us to use the Intel IGP like GT4e, along with a video card?

 

[AtenRa]

Just upload a short 3,5 minutes video playing Dragon Age : Inquisition at 1080p Medium. This one is at DX-11.

https://www.youtube.com/watch?v=XHO0PQDXfGk&list=PLPPlscE2CXdFD3sh6m6sNiPB9GdtPU0wU&index=8

And one more with A10-7850K + HD7790 with Mantle.

https://www.youtube.com/watch?v=ZIAJLShAlow&list=PLPPlscE2CXdFD3sh6m6sNiPB9GdtPU0wU&index=9

I believe with DX-12 in Mobile space, Hybrid CF will gain a lot of momentum.


edit: double up
http://www.youtubedoubler.com/?vide...dtPU0wU&index=9&start2=&authorName=snoopsagan

 

[TheELF]

Just upload a short 3,5 minutes video playing Dragon Age : Inquisition at 1080p Medium. This one is at DX-11.

And one more with A10-7850K + HD7790 with Mantle.

I believe with DX-12 in Mobile space, Hybrid CF will gain a lot of momentum.

So this is the combo that was supposed to get 38,8 fps minimums right?

Thanks for discrediting yourself,saves us the work.
Do you see now that gpu benchmarks that come with games are totally irrelevant to actual gameplay?

 

[BSim500]

Those drops in minimum framerate are a little worrying :\

That's always been true of DDR3 bottlenecked APU's. It's also why despite the hype, "relative better perf per $" is still utterly meaningless if it ends up below the absolute threshold of acceptable gaming. It's like arguing over 7fps Intel iGPU vs 12fps AMD APU's "gaming" all over again... Trying to crossfire a $130 APU with a slow $80-$90 R7 250 (total $210-$220) simply to make up an arbitrary "$200 budget" is also poor value for money (especially given modern 2-3 year split CPU/GPU upgrade cycles and the faster upgrade "churn" of APU's). The same money could buy a $70 G3258 / $80 X4 760K + GTX 750Ti / R7 260X or $100 FX-6300 / $115 i3 + GTX 750 / R7 260 with far higher min fps (due to DDR5 vs DDR3 alone) and no crossfire issues (as in OP's "I couldn't benchmark Thief"). And "Yes, but if you already own an A10-7850K and you wanted to add performance" falls under the "buy cheap, buy twice" category, and is an open admission that 2015 APU's are inadequate for 2015 gaming (and still for many 2010-2014 games assuming 30fps is a baseline, even dropping resolution / quality settings to "low").

I applaud the effort AtenRa puts into benchmarking these, but little is going to change until HBM takes off, and even then AMD really need to double the core performance yet still remain below 120w total CPU + GPU for these to be taken seriously in SFF's & gaming notebooks. "Total system power draw 225w with an A10 + 7790"? Not trying to be funny, but I've measured an i5-3570 + GTX 960 (R9 285 equivalent performance) peaking at only 172w under Prime + Furmark with over double the overall GFX performance of this 225w "APU/7790 combo", an i5-4670S + 750Ti pull under 120w, and an i3-4160T + 750Ti barely break the 100w barrier at same overall performance whilst being fed off a 160w Pico-PSU power brick. (Not to mention compared to the typical 120-160w of 7790-7850 GFX class of XB1 / PS4 consoles).

 

[mutantmagnet]

Nice post. As someone else correctly pointed out you made an error on minimum frames. What I find to be the most interesting is that the the results with the best average Hybrid Crossfire results had terrible minimum frames while the games with underwhelming scaling had very good minimum frames. I think that speaks to the priorities of the dev teams working on these games when it comes to optimization.

 

[AtenRa]

So this is the combo that was supposed to get 38,8 fps minimums right?

Thanks for discrediting yourself,saves us the work.
Do you see now that gpu benchmarks that come with games are totally irrelevant to actual gameplay?

Unless you believe that A8-7600(see video link bellow) is faster/better than the A10-7850K, then you would understand that this fps drop in the 7850K video was a random occurrence and it only happened once in a 3 minutes game play. It could happen from windows accessing the internet or the system accessing the HDD for some reason etc etc. If the fps drop would occur more than once, then we could say that this is related to the game and it would make the in-game benchmark not reliable.


A8-7600 + HD7790 (cat 15.5beta)

https://www.youtube.com/watch?v=IgvacMHh8Bc&feature=youtu.be


Also to point out that, the in-game benchmarks are made to give you a glance of what performance to expect out of your hardware. There may be some situations where you could find lower performance in actual game play or even higher performance than what you get in the in-game benchmark.

But even if you benchmark by directly playing the game, you may get different performance in one area/place of the game and a different performance in another area/place.

So at the end, your assertion falls short of technical evidence especially after we can see that the A8-7600 doesn’t exhibit a fps drop like the one we show in the A10-7850K video. So the in-game benchmark can be used to evaluate the hardware performance.

 

[TheELF]

It could happen from windows accessing the internet or the system accessing the HDD for some reason etc etc. If the fps drop would occur more than once, then we could say that this is related to the game and it would make the in-game benchmark not reliable.

Sure,but when you see a benchmark with high avg fps and low minimum fps on a pentium then it is 100% core related,right?
Like the GtaV benches you like to refer to all the time.
Because face it to get a high avg you can't have a lot of drops.

this fps drop in the 7850K video was a random occurrence and it only happened once in a 3 minutes game play

Plenty of drops below 38 at 1:04 as well,so at least twice in only 3 minutes of gameplay.

 

[AtenRa]

and no crossfire issues (as in OP's "I couldn't benchmark Thief").

That was for CrossFire in Mantle mode, in DX-11 CrossFire was working just fine.

 

[AtenRa]

Sure,but when you see a benchmark with high avg fps and low minimum fps on a pentium then it is 100% core related,right?
Like the GtaV benches you like to refer to all the time.
Because face it to get a high avg you can't have a lot of drops.

I havent bench the GTA V so i have no personal opinion on the subject, but i can tell you in Thief, the dual core Pentium although it had high average fps it was stuttering all the time. And i did run the benchmark multiply times to be sure it was the CPU and not something else.
And if im not mistaken, Toms Hardware mentioned exactly the same thing with Pentium and Thief.

Also i can tell you that in Metro Last Light Redux, you may see low fps in the graphs, like in Hybrid Cross Fire at 900p Low, but that low 12,94fps only occurred once (frame 8) and the benchmark was almost always above 30fps. Especially in the Overclocked benchmark run.

A10-7850K Default Clocks Hybrid Cross Fire

A10-7850K Overclocked Hybrid CrossFire