Can AMD counter Nvidia's Optimus?

[sulphademus]

Originally Posted by Drunkenmaster

Theres several different reasons, theres things like bios's being set for X speed in situation Y, I think in general the first cards had the most problems because, they didn't adjust for dual screens. The drivers SHOULD in most situations from, somewhere around 10.1/10.2 I think set 400/900 or 400/1000 speeds for idle when you use dual screens, but at first with initial drivers and initial bios's this would be 157/300 or so, the lowest clock you can get, but with dual screens, something happens, timing with memory and , blah, blah, blah. Not quite sure why, Nvidia saw this and had 6 months to have a fix from the start, you plug in a second screen and Nvidia cards don't drop below a fairly high speed.

...

As you said, another issue is transistor cost, which to a certain degree will go down with process nodes, but then it also increases as you increase the number of things you want to gate off. IE at 40nm say you have 20 SIMD's in a 5870, do you gate off just each group of 10, or each SIMD individually, do you find a way to shut off all but one bit of the memory controller, or each bit individually.

So AMD's drivers need to be aware of double, triple, quad, quint, and hex monitor configurations (remember Eyefinity) and adjust GPU idle speed accordingly. The system also needs to increase speed as needed for when GPGPU Flash is being used, but this may not require the full power of the card. Yes, this is much easier said than done.

Im thinking of throttling each SIMD similar to how CPU cores are now done. If I am reading this page on a single monitor setup on a 5870, perhaps only 1 of the SIMD blocks could be running 200MHz while the other 19 are in a sleep state. I jump to Youtube and it first throttles up the live SIMD block to 800MHz and if thats not enough power it fires up a second.
Also for multimonitor, keep 1 SIMD block awake per monitor connected to ensure there is enough graphical power. Make it known that each additional monitor increases idle power draw by X%.

[integrated]

http://www.anandtech.com/show/6243/a...hics-levels-up

It took a long time, but it looks like AMD may actually be able to compete in portable discrete-GPU notebooks again...

[HighTech4US]

Originally Posted by integrated

http://www.anandtech.com/show/6243/a...hics-levels-up

It took a long time, but it looks like AMD may actually be able to compete in portable discrete-GPU notebooks again...

Reading the comments on the article you posted show that many of the 7970M owners are up-in-arms over Enduro and in fact doubt AMD can fix the problems they are experiencing.

On NBR it has been concluded for the most part. Even though on paper the 680M and the 7970M are equal, it makes far more sense to pay an extra $300 because 680M works. 7970M getting less performance on half the games compared to last gen's 580M?

[distinctively]

Originally Posted by HighTech4US

Reading the comments on the article you posted show that many of the 7970M owners are up-in-arms over Enduro and in fact doubt AMD can fix the problems they are experiencing.

Sounds like Enduro has been a real flop so far but I don't think the Anand comments section is a good source to decide if AMD can fix the problems.