Ramming the APU

[LiquidHard]

Howdy,for those with the APU,

Using the IGP on the APU here for light gaming e.g. BF3, Dead Island.

Whats better for the game performance, since the IGP uses the ram:

1866 / 1900 - CL10
OR
1333 / 1600 - CL9

 

[EarthDog]

Thats kind of a wash... but I would imagine the 1866 CL10 would be faster. Test it.

 

[LiquidHard]

Thats kind of a wash... but I would imagine the 1866 CL10 would be faster. Test it.

Man you're fast.

I thought 1866 up would be faster too.

Except, fraps is not giving me consistent test results for average fps and minimum fps.

Sometimes the 1866 would be higher, sometimes not.

 

[wagex]

which apu u using? idk how one would fare with the likes of bf3 :\

iirc timings help alot with the graphics too

 

[EarthDog]

Man you're fast.

I thought 1866 up would be faster too.

Except, fraps is not giving me consistent test results for average fps and minimum fps.

Sometimes the 1866 would be higher, sometimes not.

Seems like a wash then... problem is, in BF3, unless you are doing a single player scene, the FPS wont be consistent. I suggest you try it with a canned benchmark like 3dMark11 or something.

But yeah for BF3, I imagine high + no AA or HBAO/SSAO it would be playable at 1080p.

 

[LiquidHard]

single player.

hang on, let me load a scene and come back for u wagex

 

[LiquidHard]

BF3 settings on APU3670K IGP only, ram 1866, CL10, IGP stock 444 oc 800.

1600 X 900 60Hz
Fullscreen ON
Vertical Sync ON
Motion Blur 50%
AA deferred OFF
AA POST Medium
AF 4X
Ambient Occlusion SSAO
All other video settings on MEDIUM

Operation Guillotine, first 1minute of gamplay, in-game FRAPs about 22fps

Probably higher fraps on lower resolution, or like what EarthDog posted above

 

[EarthDog]

Turn off ambient occlusion and AA post.

 

[LiquidHard]

Turn off ambient occlusion and AA post.

Alright Earthie, did tat. In-game fps hanging in the same notch, or Im missing what to look for?

 

[TsunamiJuan]

The old APU's fared better with the lowest possible cas times and the highest possible bclk due to the multi for the IGP being tied directly to the bclk. This translated across to the the 3870k's aswell as the 3850k's which is was the primarily locked version.

The tightest timings also tended to make a big difference in the past, infact most of the boards would force timings to be stuff like 9-9-9 even it was 8-9-8 or or 9 10 11.

 

[LiquidHard]

The old APU's fared better with the lowest possible cas times and the highest possible bclk due to the multi for the IGP being tied directly to the bclk. This translated across to the the 3870k's aswell as the 3850k's which is was the primarily locked version.

The tightest timings also tended to make a big difference in the past, infact most of the boards would force timings to be stuff like 9-9-9 even it was 8-9-8 or or 9 10 11.

Thanks TJ.

1) That may explain y I was getting better readings from aida64 memory test from a higher bclk VS non-bclk OC, even though the ram speed and latency was kept the same. Memory latency test fared better in higher bclk (lower ns)

2) I also noticed OCing the IGP frequency solely on a 3670K did not help my games much, perhaps 1-2 fps gain in-game fraps reading. Ram timings made the biggest difference by 5-10 fps.

3) Any idea why the igp performance is so tied to bclk, even if it was a K apu, where we can change the igp frequency?

 

[TsunamiJuan]

for none K model cpu's the bclk is tied directly to the bclk, by an unsettable multiplier. In the case of the a8 chips its a multiplier of 6. So 6x100 = 600mhz igp speed. take that up to say 140bclk and you get 840mhz using the same equation. For the a6 cores it would make sense if the used a multiplier of 4. But i cant confirm if thats the case or not.

On the newer K model chips the 3870k for instances. The bios of most motherboards will have a section for IGP frequency. Most of the time this is set to auto. If you change it to manual it will allow you increase the igp speed somewhat independent of the cpu. However it will still be limited by the memory performance of the cpu. So if your using slower memory or a low bclk it will of course still be a limiting factor in the speed.

 

[ssjwizard]

Increasing the BCLK will increase the HT Link speed and CPU-NB speed which opens up more bandwidth between the RAM and APU thus even with a multiplier capable chip you can still benefit from some BCLK tuning also. Your black edition APU should have the ability to overclock the iGPU independently through the bios to further push that limit. Tight timings help alot with the iGPU and not so much for the CPU parts. I run my wifes A8-3870k setup @ 1866 CL 7-7-7-21 1T which seems to really help in DualGraphics mode since the iGPU can almost match the 6670(at stock) in clockspeed and memory availability.

 

[LiquidHard]

Thanks TJ.

For my A6 3670K BCLK / IGP multiplier: 5
i.e. 500mhz igp settings X 140 bclk = 700mhz igp = 200mhz increase in igp mhz

Mobo: Biostar ta75m+ Bio edition:412

 

[LiquidHard]

Increasing the BCLK will increase the HT Link speed and CPU-NB speed which opens up more bandwidth between the RAM and APU thus even with a multiplier capable chip you can still benefit from some BCLK tuning also. Your black edition APU should have the ability to overclock the iGPU independently through the bios to further push that limit. Tight timings help alot with the iGPU and not so much for the CPU parts. I run my wifes A8-3870k setup @ 1866 CL 7-7-7-21 1T which seems to really help in DualGraphics mode since the iGPU can almost match the 6670(at stock) in clockspeed and memory availability.

Thanks wizard. What rams are those?

That explains why high bclk seems to run game smoother, as opposed to lower bclk (ceterus paribus)
sidenote: On the biostar ta75m+, bios does not have any options to increasing the HT Link.

If we isolate the IGP mhz, why does an increase in IGP mhz not yield bigger gains in graphics performance for games e.g. more fps 5-10?

 

[TsunamiJuan]

none of the apu boards have the ability to increase htlink speed. Its a hard multiplier again in the chip.

As for isolating the igp. from the chip, part of the reason that we dont see better yield solo is that there is a huge amount of the cpu's memory bandwith being used by the igp. So even though you have increased the IGP speed the memory speed of the chip bottle necks it. without an increase or in memory speed, or a high initial memory speed to begin with.

Sadly the the boards dont really natively support memory speeds for 2400mhz, and at 2133mhz its still kinda iffy for support sometimes. But those definitely yield better performance.

 

[LiquidHard]

none of the apu boards have the ability to increase htlink speed. Its a hard multiplier again in the chip.

As for isolating the igp. from the chip, part of the reason that we dont see better yield solo is that there is a huge amount of the cpu's memory bandwith being used by the igp. So even though you have increased the IGP speed the memory speed of the chip bottle necks it. without an increase or in memory speed, or a high initial memory speed to begin with.

Sadly the the boards dont really natively support memory speeds for 2400mhz, and at 2133mhz its still kinda iffy for support sometimes. But those definitely yield better performance.

So we raise the bclk to increase htlink speed, as per ssjwizards post? Means APU platform HT Link is also tied to BCLK i.e. CPU-NB. Please correct me if this understanding is inappropriate.

The highest I can on tha TA75m+ ram settings is 1866, apu native support. Any further I hv to raise BCLK. No memory dividers options in bios, it tied to ram speed settings.