What the FCLK? - Skylake GPU Performance

primetime

Supreme [H]ardness
Joined
Aug 17, 2005
Messages
6,605
So it somehow works with your dedicated card to boost performance? Sorry, to tired to read the link;)
 

Revdarian

2[H]4U
Joined
Aug 16, 2010
Messages
2,606
So it somehow works with your dedicated card to boost performance? Sorry, to tired to read the link;)

No.

edit to explain.
After leaving it overnight with a simple no, i will explain for the lazy.

FCLK is basically just another bus to overclock, this one relates to PCIe, the base clock is 800 mhz, you can overclock it to 1 ghz with the proper newer Bios of your motherboard, and this 25% bus overclock gives you some very slight performance boosts... usually in the 1% to 3% range, nothing to write home about. It isn't a "working with the dedicated card" deal, just a "instead of having a coffee break along the way, the mailman actually hands over the info on time".

Anandtech's average gain on a 980 was of 0.23%, which quite honestly could be just statistical error.
Edit to add:
After further consideration it actually makes sense that the 980 doesn't see too much of a gain, since on a typical workload for a 980 you will be trying to max out her "horsepower", in other words, you will be getting the gpu to do the most work possible, and thus usually it is the gpu itself the bottleneck, if any, for the performance, not the bus that delivers the data.
 
Last edited:

Nenu

[H]ardened
Joined
Apr 28, 2007
Messages
19,221
I tried reading up on fclk 2 days ago and came up with a fat nothing.
Typical he published it yesterday.
Thanks for posting.
 

defaultluser

[H]F Junkie
Joined
Jan 14, 2006
Messages
13,547
Anandtech's average gain on a 980 was of 0.23%, which quite honestly could be just statistical error.
Edit to add:
After further consideration it actually makes sense that the 980 doesn't see too much of a gain, since on a typical workload for a 980 you will be trying to max out her "horsepower", in other words, you will be getting the gpu to do the most work possible, and thus usually it is the gpu itself the bottleneck, if any, for the performance, not the bus that delivers the data.
Right, the cards that see the most gain are the ones that have the highest performance to VRAM memory ratio (R9 285 and GTX 770 have roughly similar performance, and same 2GB VRAM capacity). GPUs that get more done in a given time need access to more assets, so cards on the bleeding edge of memory capacity are the ones that will be most affected.

The GTX 980 and R9 290x have a luxurious 4GB VRAM (comparatively tons of memory), so they hardly ever use the bus to stream assets.

And this is a very small performance difference. I doubt that any of you would notice a 3% difference in real world, let-alone benchmark tests. And this would affect cards ABOUT THE SAME in similar price brackets, as they all tend to have the same amount of VRAM (or overkill VRAM like the 390). So in terms of benchmark validity, this is a non-issue.
 
Last edited:

LigTasm

Supreme [H]ardness
Joined
Jul 29, 2011
Messages
5,061
I tried reading up on fclk 2 days ago and came up with a fat nothing.
Typical he published it yesterday.
Thanks for posting.
Basically they couldn't get the microcode to work right, so instead of default being 1000mhz like Intel wanted it to be, it stayed at 800mhz. Its not really "overclocking", its a bug that causes lower than expected normal performance which is where people seem to be getting confused.
 

Nenu

[H]ardened
Joined
Apr 28, 2007
Messages
19,221
Once my board has bedded in fully and I know its issues, I'll test 1000MHz. (got this one running at the weekend)
The only benefit it might give is a decreased game loading time so I'll look for that.
If the game loading is primarily bound by PCI-E speed, it should knock a second or so off some games.
 
Top