Got a new graphics.

[peterfhayes]

Keith
I'm surprised you have got DSR to work in TS 2015 as it uses a technique called "deferred rendering" which usually means down-sampling may not work or is "hit and miss".
I've never had any luck with DSR, just the picture, text and detail gets smaller, frame rates fall, and using SSAA in TS2015 is a type of down-sampling anyway. I get very good results on many routes using NVI with this new compatibility bit.

pH

[class377fcc12]

Just a tip that I have also checked, the 960s GPU is nowhere near powerful enough to use 4 gigabytes of VRAM.

[peterfhayes]

Class

Just a tip that I have also checked, the 960s GPU is nowhere near powerful enough to use 4 gigabytes of VRAM.

Can you tell us where you got this information?
Have you reported it to NVidia and if so what is their comment.
What "power" are you taking about - bandwidth, texture fill rate, monitor size, number of monitors? etc, etc???
Thanks
pH

[deltic009]

Just a tip that I have also checked, the 960s GPU is nowhere near powerful enough to use 4 gigabytes of VRAM.

Oh dear oh dear, seems like we have an armchair expert. I've seen this card tested versus a 2Gb version and true, in most games at 1080P or 1440P there wasn't much difference but when ran at 4K resolution then it did - in one game it scored 21fps average vs 7fps average for the 2Gb version, so I would hardly call a 300% increase in performance as showing the that the card isn't powerful enough to utilise the extra VRAM

[gptech]

Conclusion: Will Games Use a 4GB GTX 960 Graphics Card?

The answer to the “is a 4GB video card worth it?” question is a decidedly boring “it depends.” As above, some games – like ACU – will actually make noteworthy gains with additional memory, while others are completely disinterested in the added capacity. For the most part, games are either optimized in such a fashion that additional VRAM offers no net gain or consume low enough capacities that it is largely irrelevant.

In the few above cases where an added 2GB vastly improves performance, it may be an arguable case that users instead invest in a higher-end GPU altogether, though that becomes a cost analysis argument that is based on the buyer's budgetary situation.

The 4GB GTX 960 ($240) can see noticeable, massive performance gains in the right situations. If you're playing the games that would see advantage from a 4GB card, it's worth considering an additional $30 for the purchase. For those who are interested in less demanding or better optimized titles (Metro, GRID), like some of those tested above, a 2GB Strix card ($210) sees effectively identical and impressive performance compared to the 4GB card.

- Steve “Lelldorianx” Burke.

from http://www.gamersnexus.net/guides/1888- ... 2gb/Page-2

The graphs shown in that mini-review rather debunk the claim that a 960 can't use 4GB of vRAM, so just where did you *check* that *tip*?

Sent not from an armchair, but laid in bed....am I more of an expert??

[ttjph]

Interesting article - thanks gptech.

Assassin's Creed Unity... regularly capped-out our available memory on the 4GB card and fully saturated the 2GB card. This saturation results in memory swapping between system RAM and the GPU's memory, causing the massive spikes reflected by the 1% low and 0.1% low numbers.

I'd be interested in seeing some maximum VRAM usage numbers for TS (obviously against resolution used) from people who have cards with enough VRAM not to saturate; my ageing 896 MB card can run close to 100% on Keith Ross routes at 1680x1050, but I wonder how much swapping it's doing? Could this be a significant contributor to the "tile loading" stutters?

[markpullinger]

Hi, it also doesn't take into account how much the card wants to use for the PhysX side as well as for graphics, my 1Gb 550ti really struggles with hardware PhysX enabled much above 720x1280.

[class377fcc12]

Sorry, but those are personal findings at 1080P. I have not tested at 1440P or 4K. I have two PCs, one with a 2gb 960 2-way SLI config, and the other running a single 4gb card.

[gptech]

Hi, it also doesn't take into account how much the card wants to use for the PhysX side as well as for graphics, my 1Gb 550ti......

It doesn't have to, TS is forced into using PhysX on the CPU---it doesn't support GPU processing.

(in nVidia control panel, force the PhysX display on or off with the '3D Settings Tab in the top menu bar to check in your own installation)

Secondly, what a 550 does or doesn't do wasn't/couldn't be part of a mini-review of 2Gb Vs 4GB vRAM equipped 960 cards.

EDIT: The recommended amount of vRAM for GPU based PhysX is:

In general, 512MB of graphics memory is recommended unless you have a GPU that is dedicated to PhysX.

(http://www.nvidia.com/object/physx_faq.html#q3) which suggests the overhead is quite small.

The number of CUDA cores and the clock speeds of the card matter much more than the amount of RAM it has.

[deltic009]

Sorry, but those are personal findings at 1080P. I have not tested at 1440P or 4K. I have two PCs, one with a 2gb 960 2-way SLI config, and the other running a single 4gb card.

So wait, you're comparing a single GPU setup with 4Gb versus a dual GPU setup of the same GPU with 2x2Gb? I would be shocked if the SLI didn;t come out on top, althoug I admittedly have zero experience of SLI setups and their uses.

But the higher you push the resolution, the more VRAM you'll need I believe.

[peterfhayes]

ttjph
Just single figures so not statistically Significant.
Taken from several routes and scenarios over a couple of years and more recently with the newer cards, which on my rigs use large chunks of VRAM
All settings in TS2015 full right, except mid shadows, no bloom, no lens-flare or LOD/DOF; Dynamic Stuff both ticked, Borderless. Measured with GPU-Z and MSI afterburner. All recommended resolution.
Rig 1) i5 3670 @ 3.4GHz – GTX 750TI, 2GB VRAM, Dell IPS Monitor 24" 1920 x 1200, Max VRAM in bursts 1.8Gb (Average over several routes 1.4 GB)
Rig 2) i7 2600K o/c to 4.4Ghz, GTX 960 4GB VRAM, AOC 27” GSync Monitor 1920 x 1080 @ 144 Hz, Max VRAM in bursts 2.1Gb but usually around 1.6GB.
Rig 3 i5 4670K o/c to 4.4GHz, GTX 970 4GB VRAM, ASUS ROG GSync monitor 2560 x 1440, 144Hz Max VRAM in bursts 2.8Gb
All seem to use the lowest VRAM of around 600 - 800 MB. Not all routes and Scenarios use the max VRAM quoted.

All 3 have nVidia Inspector settings highest possible for AA/AF
pH

[gptech]

...... I have two PCs, one with a 2gb 960 2-way SLI config, and the other running a single 4gb card........

Which one is named "Apples" and which one is named "Oranges"?

[gptech]

But the higher you push the resolution, the more VRAM you'll need I believe.

Yes Matthew, up to 1920x1080 on an average sized monitor---the kind that the majority of folk use--you can run TS with a 1GB equipped card. Higher than 2x1 or 1x2 SSAA is out though, unless you sacrifice other graphical options of course, but the game is definitely playable with cards that people can actually afford/are willing to 'push the boat out' on.
Once you start adding any nVidia DSR settings then yes, the more vRAM the better...simple maths; more width x height = more pixels to hold in memory therefore more memory is needed.

[Trev123]

Who cares technically what a graphics card can do, I don't, gave up on all that years ago. The proof in the pudding to me is what it looks like and runs like on screen. I am getting frame rates up to 90fps on some routes like the Riviera route and down to about 30fps on South London Network with lots of traffic on my new card.

[Trev123]

Just ran a scenario on DPS East Coast Mainline (North East) with the Class 180 getting 90fps graphics are set to Highest with everything on full, not even any scenery loading stutter. Vast improvement over my 560 TI 1gb card.