65 replies to this topic

[xWiredx]

Yes, that's nice. You do realize that this fact does not change the already-measured performance of the card, right?

[Oderint dum Metuant]

gaIaxor, on 13 March 2015 - 01:02 PM, said:

And there you just have to look at the frametimes. That explains why the driver tries to avoid using more than 3.5GB



It can use both segments, but not at the same time:
"accessing the weaker 512MB segment blocks access to the stronger 3.5GB segment"

You don't see the contradiction in what you just posted?

If it could not access both parts at the same time the card would not exceed 3.5GB Period.

It can therefore it does.

[Catamount]

Being able to store in both portions wouldn't necessarily mean it could perform read/write operations to both portions simultaneously. I read Galaxor's cited source, who I assume know what they're talking about, as saying exactly that. "Access" means just that. If that's the case it'd be no wonder the 970 chunks badly, because every operation to read from the painfully slow 512mb section would not only be, well, painfully slow, but would hold up the 3.5GB.

The problem with the 970 isn't the measured performance. When it released it basically looked like a 290x (a Titan even?) for less money and really good, if somewhat understated, power consumption. The problem is that that measured performance is not actually available to the end user in all situations, because graphics settings you'd be able to push smoothly with them make the card chunk as the memory system holds things up.

In other words, we see a review with X performance, enabling gameplay at the target framerate for a given title with A settings and resolution, but really, you're only getting X-h and A-h, where h is the divide between what the 970 has the horsepower to run smoothly, and where you have to put settings to avoid hitting the memory bottleneck. SoM is a no-go with high resolutions and textures, Star Citizen is being reported to be a VRAM hog so that's probably the same thing (I'll have to get with the 970 owners I know to do some testing later, but I'd say my friend is out being able to use the consumer Rift like he was planning), so you're not buying that already-measured performance, but rather that already measured performance minus bottlenecks. At absolute best, that's a gamble that you're really getting the performance advertised for your money... most of the time.

Edited by Catamount, 14 March 2015 - 06:48 AM.

[Oderint dum Metuant]

Catamount, on 14 March 2015 - 06:26 AM, said:

Being able to store in both portions wouldn't necessarily mean it could perform read/write operations to both portions simultaneously. I read Galaxor's cited source, who I assume know what they're talking about, as saying exactly that. "Access" means just that. If that's the case it'd be no wonder the 970 chunks badly, because every operation to read from the painfully slow 512mb section would not only be, well, painfully slow, but would hold up the 3.5GB.

The problem with the 970 isn't the measured performance. When it released it basically looked like a 290x (a Titan even?) for less money and really good, if somewhat understated, power consumption. The problem is that that measured performance is not actually available to the end user in all situations, because graphics settings you'd be able to push smoothly with them make the card chunk as the memory system holds things up.

In other words, we see a review with X performance, enabling gameplay at the target framerate for a given title with A settings and resolution, but really, you're only getting X-h and A-h, where h is the divide between what the 970 has the horsepower to run smoothly, and where you have to put settings to avoid hitting the memory bottleneck. SoM is a no-go with high resolutions and textures, Star Citizen is being reported to be a VRAM hog so that's probably the same thing (I'll have to get with the 970 owners I know to do some testing later, but I'd say my friend is out being able to use the consumer Rift like he was planning), so you're not buying that already-measured performance, but rather that already measured performance minus bottlenecks. At absolute best, that's a gamble that you're really getting the performance advertised for your money... most of the time.

You have to go with the language available to you. Currently that is either Anandtech's it cant use both sections to PC gamers

Quote

[color=#000000]The GTX 970 does have 4GB of VRAM, and it can use all of it, but accessing those last 500MB will decrease performance.[/color]

http://www.pcgamer.c...than-35gb-vram/

Now we've touched in this before in this thread i think from what i have seen of end users tests elsewhere, the card does use the total amount, I just don't see on a technical level how it can't once it has to use past that 3.5GB.

Now does the card try as hard as it can to not go above that 3.5GB, damn sure it does and that's to prevent the bottlenecking issue when it uses the .5GB section and 3.5GB.

Now i may be wrong, but i just can't see a technical way of the card being able to continue its current operations if it can't access both parts at the same time.

[DustySkunk]

Hi all,

I just wanted to thank you all again for your help and support in going through this process with me. I learned much from what all of you had to contribute. I just pulled the trigger on the parts today and will update the OP with the final specs when I get home this evening.

[DustySkunk]

Well she's built and she's beautiful!!! A few minor changes to the build in the OP (primarily the case... I grabbed an Antec 1100 v2). It's like a whole new game. When I booted up up the game for the first time and dropped into a match I was killed because I was zoomed in and staring at the enemy mechs

My old system was below minimum spec. I had every trick in the book going to get it running at a "playable" framerate. I took a few new screenies to show you all the results I'm getting from the new system. All of these are taken at 1920x1080 with DX11, all settings on high.


This first one is significant because... such sun! and *gasp* shadows! and in the past when I rounded this corner my FPS would tank hard. Look at the FPS counter...

Spoiler

HPG is another map that would tank me badly. The Bog as well.... not anymore....

Spoiler

Crazy action doesn't drop FPS all that badly either. I rarely drop much below 60. Here's a screenie of me spectating another player eating a faceful of LRMs. (something that would have turned my game into a slideshow before). Notice FPS...

Spoiler

Finally, my performance is getting better as a result. It's so much easier to target damaged sections on mechs (or for that matter, target at all lol). My kill/damage ration is much better. It used to take me twice the damage to get this many kills...

Spoiler

Anyway. It's a whole new game. Thank you all again for your help!!!