Help me pick out a new video card for my computer!

[SGGG2]

Ash, I didn't mean to imply you should post there -- it's an excellent resource (seriously, did you check it out?) and I wanted to make it available to anyone who's unfamiliar. I recently built a gaming PC for stereoscopic 3D gaming plus PCSX2 and Dolphin and it helped me out tremendously.

[evil_ash_xero]

Ash, I didn't mean to imply you should post there -- it's an excellent resource (seriously, did you check it out?) and I wanted to make it available to anyone who's unfamiliar. I recently built a gaming PC for stereoscopic 3D gaming plus PCSX2 and Dolphin and it helped me out tremendously.

Oh, it's cool. I got banned from there, and I'm still bitter.

They do have their informative threads though, yes.

[Xyga]

Still not settled on a GPU *sigh* it's tricky when you don't know exactly what 'power' and features you need. ><

I've asked on PCSX2 forums about the i5 K + GTX760 config I'm thinking of and they replied it's pretty good for most games (not at the highest visual settings but I don't care) and I can always overclock later for the really heavy/problem games.

The thing is for my other intended use I have no idea if that 760 will be okay too;

As mentioned on the previous pages, I plan to purchase the much anticipated Acer XB270HU (2560x1440 IPS 144Hz G-Sync) or similar competing model to come, in order to take full advantage of all 8/16/32/64bit + arcade emulation advances and tricks (doubled framerates/ULMB, fluid v-sync-less synchronization, most advanced shaders), and I don't know if I will ba able to do all that at native 2560x1440 with a GTX 760...

I know the 960 will be out in two days, but many people already bitch about the smaller bus (128-bit), when other say it doesn't matter since the different architecture using compression allows for better (?) performance.

Some members here do fiddle with HLSL/CRT_shaders at 1440, but I don't know what kind of gear they use... any advice welcome.

[Ed Oscuro]

If you have them "on the line," I'd be interested to ask what impact, if any, a wider bus would have on Dolphin performance, and whether GPUs with the OpenGL "Buffer Storage" extension and plenty of RAM reduce the importance of this requirement somewhat.

I don't want to spread any misinformation, but I still want to say that my belief is that having a huge bus probably won't make much, if any, difference in Dolphin, so long as the rest of the GPU performance remains unchanged. From the FAQ:
https://dolphin-emu.org/docs/faq/#which-gpu-do-i-need

"Pixels and enhancements," as well as the discussion of special features of newer DirectX versions used to more accurately emulate the fixed-function hardware in the GameCube, all seem to indicate requirements of the GPU itself. So, what's left? If we're talking about a PC game, there certainly is a lot of coordination done between the CPU (and memory) and the GPU, which can readily eat up that bandwidth with tons of texture data, swapping out of complicated models, and other unusual tasks like tessellation or physics. For an emulated game, a lot of the traditional things (textures, coordinating placement of 3D objects, etc.) shouldn't require much bandwidth in comparison.

The one thing that does stand out are the shaders used to program the GPU. But, once again, how much bandwidth does this really require? I unfortunately don't see any explanation of this topic. Something like this random page, which does talk about the performance of shaders being impacted by bandwidth, doesn't seem possible to use in direct comparison with Dolphin - because Dolphin's requirements are very different from 3DMark. There is one line that I found interesting, however: "The 4870 actually understates the advantage of memory bandwidth, because the shaders are less capable than the 5850’s – about 12% lower in terms of GFLOP/s." Funny, I have both those GPUs - but anyhow, the key point here is that a higher performing shader will result in higher performance overall, all other things held equal. If the workload isn't so bandwidth-intensive, but is highly shader-intensive, having a newer and faster GPU may well turn out to be more important than bandwidth.

It's true that the Maxwell bandwidth is better than previous types - but this appears to be specialized in color compression to reduce the amount of bandwidth required to transmit color information. I don't think this is the most important issue in Dolphin performance, and I don't think this will help other types of data transfer across this bus.

My expectation is that a slower bus may lead to slight stutters when beginning emulation, as shaders are loaded onto the GPU, but performance afterward shouldn't differ from GPUs with bigger busses. Likewise, I don't think that a bigger bus will magically erase these slight loading hiccups. In no case would I expect ongoing stutters or slowdown in gameplay, so long as the GPU is fairly recent. How much money is it worth spending to erase a few hiccups? Component developers are well aware that maximizing CPU and GPU performance requires the bandwidth to be fairly good between them, so I think the question of getting a very powerful CPU and GPU but then starving them for bandwidth may be somewhat academic. Maybe the 760 Ti is a special case, but I don't think that bandwidth alone is going to magically turn around any performance problems in the most demanding games, when the GPU shader units still need to perform more quickly.

[Xyga]

Now that the 960's reviews are up I'll ask them the question in a more simplistic way;

"Overall the 960 scores a little bit higher than the 760 in PC games (about 10%, some say 20% thu), but because of its tiny bus the memory bandwidth tops at 112GB/s (149GB/s if the 'color compression' architecture works) when the 760 offers a solid 192GB/s.

So what's more important for emulators like PCSX2 and Dolphin ? those added 10% or the larger bandwidth ?"

I'm tempted by the 960 since it's more modern, reviewers complain that it's can't handle 1440p and 4K at high details levels, but I don't care about that as long as it handles heavy MAME + shaders at 1440p (which is the only use I'll have of that resolution later).

[Ed Oscuro]

More modern features might tip it your way. I don't think combining the two variables (bus vs. shader performance, though that's not actually what that review score will be showing for the 10%) actually makes it simpler. Probably nobody affiliated with the project yet has the GPU so you might consider waiting until that information is out there, for a more solid basis to answer your question. That said, I think the theoretical answer about bandwidth would be keen...maybe I'll finally register and tell them about Space Raiders' savestate problem.

Edit: This page has some comparisons in AutoDesk CAD and Maya applications:
http://www.tomshardware.com/reviews/nvi ... 038-7.html

It's interesting to note that the 960 drops a bit behind, apparently due in part to bandwidth, in DirectX. OpenGL leveled things out for the 960, but note that Dolphin still is, last I checked, much better performing under DirectX.

[Xyga]

Now some say even a 750 Ti is enough for most PCSX2 emulated games even at 4x internal resolution...

It's a significantly cheaper option compared to the 960 and even 760, but I need one with DP and there are fewer options around;

• . EVGA GeForce GTX 750 Ti FTW ACX (02G-P4-3757-KR)
  . ASUS GeForce GTX 750 Ti STRIX OC (STRIX-GTX750TI-OC-2GD5)

Supposedly both DP 1.2 and therefore should support G-Sync fully... but I need a few confirmations (on DP version and driver version) to be sure.

[Ed Oscuro]

PCSX2 does basically nothing with graphics hardware. On my ancient laptop, it was the dual-core CPU holding back Castlevania LoI from being 100%, not the very old GeForce 8600M GT. Though I was only running at original resolution there (filtered) there certainly will be enough rendering grunt to increase the rendering rez.

G-SYNC support will be in the nVidia drivers already, so no worries there.

It sounds like you want to hold off on a purchase until you find the monitor you want, like me

[Xyga]

It sounds like you want to hold off on a purchase until you find the monitor you want, like me

In a way yes, I'm looking for a GPU that will handle the 'new generation' of displays, like the upcoming Acer XB270HU.
So G-Sync support becomes kind of mandatory, and the juice too but I don't need barrels since my rig will be an 'emulation' rig.

One thing in favor of the 960 would be HDMI 2.0 support, if one would consider to include TV's in his search for the 'right' display.

*sigh*

The 960 is a bit overkill for my needs and I'd like to limit cost, but I also don't want to limit my display options in a few months... so maybe I'll wait until March-April and get both GPU and new display at the same time, like you.

It's my first time really having to choose a GPU and there are much more options than I ever imagined, this is a really fucked up market.

[Ed Oscuro]

And to mix things up a bit more:
http://www.digitaltrends.com/computing/ ... n-r9-380x/

It is interesting to see that AMD is going to better memory so soon. This might go a long ways in reducing the lead from core improvements nVidia has shown with the Maxwell series, though it will depend on usage case. I hope AMD also pairs this with some core improvements. But it's just another reason to hold on a little longer if possible.

[Lord of Pirates]

And to mix things up a bit more:
http://www.digitaltrends.com/computing/ ... n-r9-380x/

It is interesting to see that AMD is going to better memory so soon. This might go a long ways in reducing the lead from core improvements nVidia has shown with the Maxwell series, though it will depend on usage case. I hope AMD also pairs this with some core improvements. But it's just another reason to hold on a little longer if possible.

I'm still not fond of the smaller bus, short of a larger one actually being detrimental to performance I don't think there's a reason to cut it down, certainly not on the high end or flagship cards like the 970/80. The average power and temp. drops are nice though. It would be nice to see nVidia (or AMD/someone else)develop some manner of shader based AA, that doesn't degrade the image like FXAA/SMAA or lessening it significantly, HardOCP mentioned it in their MFAA article and I couldn't agree more.

[Xan]

It's as fast as a 3 year old HD 7950/R9 280, which has 1 GB more, but needs over 100W more juice and doesn't even support Freesync for gaming. 960 Ti with 3 GB/192-bit would be the next logical step.

[Ed Oscuro]

@ Pirates Lord: I think even the 980 is going to be enthusiast mid/higher range. I'm expecting the "big Maxwell" rumors will have more to them, as it stands to reason there will be a Titan release (or Titan-style at least). For AA, I like the downsampling idea, though it unfortunately doesn't seem universally applicable. About the bus, what card are you referring to? It's also worth stating that just throwing power at a bus for the sake of having a big bus doesn't make sense to gamers unless that helps the card run faster in real-world tasks. It might help in some of our unusual usage cases, but I'm still waiting to hear about that.

@ Xan: Same question there; what card do you mean?

[Xan]

The GTX 960.

[Xyga]

It's a '128bit bus -gate', everyone is looking at this particular spec and overall saying the 960 doesn't belong in the '900' series and they should make a Ti w/ 256bit bus right away.

It's my first time ever being in the GPU mess, and I believe they're right lol.

[Ed Oscuro]

But again, if the most stringent use case is Dolphin, it would be best to ask them directly about what they think this means, and under different games. I've not yet tried The Last Story, which is apparently one of the most difficult games to run in emulation. If I knew hard cases like that wouldn't be on the table, I'd be a lot less interested in the performance debate, period.

Keep in mind my 5 year old Radeon HD 5850 has run almost every game, and especially since the vast performance improvements from the last year or more, pretty much flawlessly. Doubtful losing a bit of VRAM or a bit of bus bandwidth is really going to sink things. At the same time, apparently the powers that be decided this helped them hit their preferred power and performance targets - and balance them off if need be.

For running PC games "on the cheap," like I tend to do, bus speed probably doesn't matter much. I've always tried to "underclock" performance by running things with the least intense settings, to save wear and power use even when my hardware is straight from the land of intensity. Choosing less capable card will just restrict you to this plan involuntarily Of course with Dolphin you need to meet minimums and there's little wiggle room, but the new cards may well have boosted the entry level to far more power than Dolphin needs in many cases.

[Lord of Pirates]

@ Pirates Lord: I think even the 980 is going to be enthusiast mid/higher range. I'm expecting the "big Maxwell" rumors will have more to them, as it stands to reason there will be a Titan release (or Titan-style at least). For AA, I like the downsampling idea, though it unfortunately doesn't seem universally applicable. About the bus, what card are you referring to? It's also worth stating that just throwing power at a bus for the sake of having a big bus doesn't make sense to gamers unless that helps the card run faster in real-world tasks. It might help in some of our unusual usage cases, but I'm still waiting to hear about that.

@ Xan: Same question there; what card do you mean?

I imagine the next Titan card is going to be just as useless for the price as the previous one unless there have been drastic improvements. Titan Z was beaten by a dual card costing half as much. The whole Maxwell line, it's always relevant for AA and I don't see it as a necessary cut unless it was affecting things negatively. I wasn't mentioning it in relation to Xyga's intended use, sorry for derailing the thread a bit.

[Xyga]

Okay I think I've found my GPU: http://www.gigabyte.us/products/product ... TOC-2GL#ov

It's definitely not the biggest factory OC 750 Ti, but it's got DP 1.2 (which I need for future G-Sync use) and is significantly cheaper than any other of my previous options (140€).

Should be enough for PCSX2 at least, hopefully enough for 1440p MAME with shaders and both v-sync and g-sync, hopefully enough for Dolphin as well.

Well, you'll know when I get the rig and test the shit our of it.

[Xyga]

Okay I think I've found my GPU: http://www.gigabyte.us/products/product ... TOC-2GL#ov.

So I got that GPU and it's really okay in terms of muscle it can easily handle both PCSX2 and Dolphin at high upscaled internal resolutions with various AA and filtering on it.

The only issue i get it's that I get tearing with pretty much every application and whatever the app's settings or nVidia control panel sync settings I use, there's always degree of tearing.

Also I'm not sure everything's working perfectly because on DVI the signal almost always disappears seconds after powering the computer.
When using HDMI instead (with a DVI>HDMI adapter) I don't lose the picture and there's less tearing (but it's still there).

Yup geforce drivers are the newest.
Dunno what's the issue, maybe it's because of my mobo's bios/drivers/firmwares that aren't all the latest. The MSI stuff is kind of fucked-up and let's not even talk about their 'Live Update' app...

[Xan]

Well, what did you expect? Isn't that why we are waiting for Freesync/G-Sync monitors to arrive? There is plenty of stuff to experiment with in the control panel, like framerate limiters/adaptive VSync/triple buffering (OpenGL only though I believe). Doubt it will give a perfect experience though. I'd also recommend checking out Nvidia Inspector for some custom AA modes.

As for the other issue, you mean you don't see the boot screen on your DVI display when you power the PC? Sometimes when another display is hooked up via HDMI the boot screen is displayed there instead of DVI, in my experience.

[Xyga]

Well I'd expect that when V-Sync is set on 'Active' I don't get tearing. I've owned other older GPU's from both nVidia and AMD, and V-Sync just worked when enabled/forced.
It's appalling to see that today on more modern hardware and software a simple command such as this just doesn't work...
And yes of course I've tried all other settings including app-controlled, adaptive, triple buffering, etc. No changes.

Regarding the DVI issue: I see the MSI boot pic, the windows loading bar, and the the monitor looses the signal.
Sometimes it goes up to the windows login screen and disappears before I get to type my password.
Strange. Anyway it works on HDMI...

EDIT: also tried VGA from the mobo, but it doesn't work at all. Maybe the mobo's VGA driver is too old or wrong, I've tried to update it with MSI's crappy app and it didn't work.

[Xan]

If you want to try it with the integrated GPU, uninstall the Nvidia drivers and install the newest Intel ones manually from their site. Also be sure that the integrated GPU is enabled in the EFI/BIOS.

[Xyga]

Nevermind I've found the culprit: I had Aero disabled, which forces V-Sync OFF on absolutely everything, and has priority over any GPU.

Switched back to Aero theme and now everything's working fine.

It's so stupid I just can't believe it.

[Ed Oscuro]

Yuck. I don't like running Aero either. Are there any visual / performance changes of note?

[Xyga]

Transparent and animated windows. That's it.
Uses about 200 or 300mb of memory just for that.
That's the stupid price to keep v-sync working.

[Ed Oscuro]

Did you check options to disable those extra effects in Computer - > Properties - > Advanced System Settings - > Performance?

A really cursory Reddit search mentions that desktop compositing is on all the time in Windows 8 - no big deal - but I read a concerning post that software (i.e., games) that load into the front buffer only will not work. Maybe a compatibility mode / fullscreen fixes that, but now there's something else to watch out for, potentially.

[Xyga]

Yeah so it's something like double buffering + vsync, disabling Aero or just desktop composition cancels that (I went to performance settings but I bet there aren't much things you can remove without disabling DWM).
They say it doesn't matter in fullscreen there shouldn't be as much lag... does it mean 'DWM' dynamically changes its drawing method and kind of hands it to the GPU when you go fullscreen ?