Hybrid SLI dying?

No. The adaptive option has been disabled for quite some time. ((read earlier in this thread)) Its only been reimplemented since the drivers gave the ability to shut it off. People were unhappy that they had no control over it. See Nvnews thread which is an example of feedback I collected from various websites. So Nvidia disabled the feature all together in the early 18x.xx sets. Alot of people wanted the ability to toggle this feature off and on. The benefit is the GPUs will run several degrees cooler when running CPU limited applications. Those who play MMORPGs that are CPU limited won't have their GPUs running full clocks all the time when its not doing anything. But of course there are other examples. Thats just one apt to me.

If the GPU is actually under load. The drivers will run under full clocks. The toggle option. Simply allows the driver to force itself to full clocks all the time regardless of whether the GPU is under load in any app that uses DirectX/OpenGL. A "low" power mode is actually being discussed. Where the clocks are "always" throttled. But like anything. Theres a schedule for the way software gets turned/release and then it has to go through QA.

 

What exactly are you talking about? Are you discussing software controlled clock gating, or are you talking about actually reducing the operating frequency of the card (under driver/SW control)?

DK

 

Its hardware and software controlled. The driver determines the loads of the card, ((IE whether shader domains, ROPS, are idling)) Then will dynamically adjust their voltages/clocks dependent upon the actual load they are under. Your ROPS may be under full load but your shader core may be idling. So therefore the driver determines what is needed.

The result is lower power usage in essentially CPU limited games or low shader intensive games. However the driver will automatically switch back up if detects a cgange in GPU load. Some people did not like this feature. So Nvidia completely disabled it in 18.xx drivers, After being renabled in the 18x.xx drivers.

However the new 19.xx drivers allow you to toggle the feature on and off. If you leave it full performance all the time mode. Then whenever you load any kind of DX/OpenGL app your clocks will always be at max values of what the card support. It does require hardware support as only GT200 + GPUS support it.

Chris

 

Well its about time (spoken as one of those who waited 2 years in vain for a reply from nvidia/nzone moderator )

 

I'm sorry you feel that way. Whether it was me or another Nzone moderator. We keep pretty busy and alot of things slip over the radar. Also I have to choose which issues to address and which not. But I hate to say it. Sometimes we have to choose to work on what we can fix. Not try to fix every thing we can't. Not that whatever your issue was/is not important. But theres only so much the moderator team can do. At times I simply have to ask. "Is this a big enough issue to bother Nvidia with?". Like I said, Not trying to downplay whatever issue you may have. But like I said. Theres only so much we can do.

Your other option is to use Nvidia driver feedback. But Nvidia's QA team does look it over.

https://surveys.nvidia.com/index.jsp?pi=ad1c492c9d7488731f70e8d4356ba631

Chris

 

It wasnt so much the issue of whether it could be fixed or not, but just to have someone acknowledge our existence would of been so nice.

 

What is the granularity for this clock and voltage adjustment? I can imagine:
1. Single SM granularity
2. TPC granularity
3. Shader/ROP/setup/memory controller granularity

Without any information from NV, I would assume 3.

It also sounds like the only adjustment that's made is decreasing frequency and voltage.

So the big problem is that the driver's response may be sort of slow (say on the order of 10s of ms), creating a bit of a lag. I can understand why users would be annoyed and want to turn it off for certain applications (I would probably keep it on, but my idea of a taxing game is Medieval II or NWN2).

In addition, changing F and V isn't immediate unless you put a lot of effort into things (e.g. Foxton for Montecito or Foxton II in Tukwila). It probably takes a large number of cycles to make that shift.

DK

 

Is this somehow related to the feature that GTX2xx cards won't throttle to 2D clock speeds if two monitors are attached? I installed these latest drivers, but it's still running at high-power 3D speeds all the time, regardless of this new setting.

 

It's not fast enough. A 9800GTX+ (in my experience) is faster on its own doing physx + graphics than having an 8400gs add in. However, there was a stuttering present with just the 9800gtx+ alone (perhaps from the context switch penalty?) that wasn't there with the 8400gs.