970 memory allocation issue revisited

as it should be, turning off the pagefile is stupid and pointless.

 

Off

But now i have it on because i had crashes in Watch_Dogs.

 

Perhaps it was related to PEBKAC?

 

Would OpenCL be a good place to start?

 

Not likely.

 

I'm more inclined not to trust your post.

 

Why not? Everything posted on the internet is 100% credible isn`t it?

 

Ok, I think I got everything covered.....lol

The R9 295X2 does in fact have 8GB of VRAM physically installed on the card. All 8GB of VRAM is in fact useable. The only place where they aren't entirely truthful is in the fact that they never mention that data from GPU1's 4GB memory pool is mirrored into GPU2's 4GB memory pool. Other than that, they're truthful.

That may be true of some reviewers.... It was known "back in the day" that Intel paid off some reviewers to favor their products. However, I don't expect Hilbert or Anandtech would risk their reputations by participating in such deception.

Most reliable way is through gaming.....

Gigabyte did the same with their custom GeForce 6600GT dual GPU card.....NVidia did it with the GeForce 7950GX2, the GTX295, GTX590.... It's only false advertising, if the physically installed memory is less than advertised. Really no reason for anyone to bitch about AMD doing what NVidia has always done. Really no reason for people to be ditching now, seeing as how NVidia has always done things like this.....

If you want to load memory using a non-graphics oriented method, yes.

 

You have 16 GB of ram and Vram4GB (3.5 + 0,5GB) and using the swap file which is slower. PF is used before you load in all available physical memory on the computer.

 

Here we go again with PageFile = Off.

God knows how many other "optimizations" you have on your system.
99% of all you stutterers deserve all the stuttering you can get

 

From my perspective a "controlled" environment is the best way to test things. But never mind the OpenGL/CL thinking, I've been talking to a colleague and he said that it's going to be a headache as you need frameworks which could undermine the credibility/reliability of the result. I'm curious if the same results can be produced using something else then CUDA, as they could provide more insight.

 

You'll never get conclusive results. The best you can do, is compare-able data.

 

Hopefully, you do understand why this means almost nothing...?

The only thing that counts at the moment is what nVidia itself has stated, and that is:

*Both the 970 & the 980 suffer a huge fall-off in performance once the 3.5GB boundary is exceeded. The 980 drops 24% in the example nVidia used; the 970 drops 25% in the same game--in the same test, according to nVidia. In the other games, both nVidia products fall off > 40%+, which is a giganormous performance drop off. Not *just* the 970...nVIdia's admitted here that *both* cards suffer from exactly the same issue. THere's *no difference* between a 970 & 980 once the 3.5GB threshold has been eclipsed--performance takes a nosedive with both products. It's all right there in that green nVidia worksheet that has been reprinted so many times in this thread.

Neither card...actually provides a 4GB contiguous pool of VRAM. Neither one. 3.5GB & .5 GB pools is what you get with both cards. That's the incredible thing that nVidia has admitted here.

So, what's the the 1% & 3% difference in performance drop off about? Let me quote nVidia here:

"However the 970 has a different configuration of SMs than the 980, and fewer cross bar resources to the memory system."

There's your 1% & 3% difference between the two in performance drop-off after 3.5GBs. The 980 is only slightly faster than the 970 in terms of performance drop off because the 980 has a more optimal SM layout. Both cards really suffer when their respective 3.5GB pools of ram have been exceeded.

nVidia has designed both the 980 & the 970 with two pools of VRAM: 3.5GB and .5GB. Both of them. The information comes directly from nVidia and has been reprinted on many pages in this thread...

Wowza'! If I have interpreted the statement from nVidia erroneously, please instruct me as I am all ears...

 

Too many trolls these days..

 

Upcoming R9 3xx are looking better and better...

 

Tip: Don't buy "Nvidea" cards from a Chinese dollar store. I also heard that "SONYA" headphones are actually rather bad

 

But what if the stacked memory were to topple over...

 

This thread is sort of ridiculous, actually...so many raw emotions about something so obvious...I was amazed to see how many people got stuck at the 1% & 3% numbers, which are nearly irrelevant to the issue discussed, without understanding that the 24%&25%-40%+ performance drops with both products was far more interesting. That, and the fact that it was nVidia itself which divulged the information...

For what it's worth, I don't consider it a big deal because 90%+ of all the games anyone will play with either of these cards will not suffer at all because the games won't ask for more than 3.5GBs. Still, when you advertise 4GBs of ram...most people do indeed think of 4GBs in a contiguous pool...running at the advertised bus width, too...That part of it strikes me as fairly important information for a consumer to have.

 

Dude..The big drop(24/25%) is caused by the heavier resolution..Read NVIDIA's statement again please..

 

Very nice explanation there.

It looks like you did very well at simplifying NVidia's statement without losing important information.