What to Upgrade?

[Athanasa]

I'm unable to run games at super shiny quality. It's a tragedy, right? Thing is, a few years ago (before my GPU died, I got it replaced, and it died again) I could run stuff at Awesome Mega Shiny mode. Giving up on the repeatedly dying GPU, I replaced it with something... cheaper, as I'd sat on my RMA for the GPU for too long close to the end of it's warranty. That and I think the cat fished it out of the box I'd put it in for returning and molested it even further. Ho hum, we live and learn.

 

ANYWAY! My question is, what's bottlenecking my ability to have nice shadows in games? Should I get an entirely new GPU, or should I cross-fire / SLI or whatever it is with my current one? What would get me the most bang-for-buck? Can my PSU handle it if I do that? Christmas has given me an extra bit of money to throw about.

 

 

Current Specs

MoBo: Gigabyte Z68AP-D3 (I should replace this - my tower unit turned out to be a variant 2/3 of the front ports as USB 3.0, which my MoBo doesn't have the sockets for... but it works, so I'm being lazy.)

CPU: i5 2500k @ 3.30 GHz (average running heat when idle is 35C - do I have room to overclock? HOW do I overclock? It only has the basic factory fan currently, but with cooling paste.)

RAM: 16GB Dual-Channel DDR3 (It was on special offer.)

GPU: 2047MB NVIDIA GeForce GTX 750 Ti (ASUS flavour)

PSU: CORS 850W CMPSU-850TXV2 PSU

Primary Screen: Acer G245HQ (23.6", 1920 x 1080, 60 Mhz)

Secondary Screen: DELL E207WFP from 2008 or something. (20.1", 1680 x 1050, 60Mhz) - predates HDMI

Original GPU: 3GB ASUS HD7950 (RIP, sweet prince... but you failed me too often. And now I miss your power.)

Anyway, I was looking at the NVIDIA GTX 960 model, assuming that the GPU is the weak point in my PC. They're well within budget (under £200). However, I don't know which of the 'Brands' to get - Asus normally seems pretty good with their warranties and returns?

Also, what - in layman's speak - is the difference between these three? I have no idea what half these numbers even mean. Does VRAM give me an edge for HD textures and whatnot?

£205 Most Expensive - Listed as High End.

4GB Asus GTX 960 STRIX DirectCU II OC, PCIe 3.0, 7010MHz GDDR5, GPU 1228MHz, Boost 1317MHz, Cores 1024, 3x DP/HDMI/DVI-I

£169 2GB, High End - Listed as high end on Scan.

2GB Asus GTX 960 STRIX DirectCU II OC, PCIe 3.0, 7200MHz GDDR5, GPU 1253MHz, Boost 1317MHz, Cores 1024, 3x DP/HDMI/DVI-I

£169 4GB VRAM - Listed as mid-range on Scan.

4GB Asus GTX 960 TURBO OC(?), PCIe 3.0 (x16), 7010MHz GDDR5, GPU 1190MHz, Boost 1253MHz, Cores 1024, 3x DP/HDMI/DVI-I DL

I really can't tell the differences here.

Edited December 28, 2015 by Athanasa

[obobski]

The GPU is very likely your bottleneck - GTX 750 isn't a bad card but it isn't as powerful as HD 7950 or other top-tier performers, especially nowadays. The i5 shouldn't be too much of a problem as-is, but you could replace the heatsink if you felt so inclined (the temperatures at idle aren't bad, but its probably getting quite a bit warmer under load; not that it's likely in any dangerous place, the bigger advantage to a replacement heatsink would be quieter operation).

 

The GTX 960 is a good bit faster, and would probably address your desires. Ditto for other similar cards, like the Radeon R9 280 or 280X, or R9 290/290X/390/390X, or GeForce GTX 770 or 970.

 

I'm entirely unfamiliar with "CORS" as a PSU maker - I hate to say it but it could be some no-name gutless wonder, in which case replacing it may be a good idea. You'd have to do some more research on that to find out though. Basically the problem here could be, that no-name/cheapo PSUs usually lie on their ratings, so they'll claim they can do say 1000W out, but in reality its the guts of a 400W PSU and when the system tries to load it to 1000W it may blow up, it may shut down, etc. A PSU from a reputable/quality manufacturer (like EVGA, Thermaltake, Antec, PC Power & Cooling, Zippy, Seasonic, etc) should not behave in that way, and will provide its rated power (and often at higher than normal temperatures, which is a good thing too). Use jonnyguru as a resource for PSU reviews, as they're very thorough. Don't, however, run from the monitor screaming in fear that your PSU is bad - I'm located in the US (and therefore know US brands and retailers), but I know there are different brands in Europe and Asia that are also reputable (in some cases just different trade dress of the same product), so for all I know "CORS" is tip-top stuff for your region of the world, and perfectly fine.

 

As far as the numbers and whatnot: I didn't actually know GTX 960s were made as 2GB variants. This isn't a bad thing though. VRAM does help with performance, but like system memory, it only becomes a problem when you have not enough (and applications don't directly access/"see" VRAM so "not enough" isn't cut and dry). 2GB should be considered appropriate, but if the 4GB cards aren't any extra money, why not? As far as the other numbers, they're different clockspeeds - faster is better. The middle one looks like its the fastest (and that may be why it has 2GB instead of 4GB; because it has faster RAM it may be they had to use lower density memory chips), elsewise the top one; the bottom one would be slightly slower. I'd probably go for the bottom one given the prices listed - it has 4GB of VRAM over 2GB (for the same exact price as the 2GB model), and the clock difference isn't worth ~$50 let alone 50 pounds.

 

If you want to look at some competition to Asus, give Gigabyte, PNY, and EVGA a look. If you're looking at Radeon cards as well, add Sapphire, XFX, and PowerColor (TUL Corporation) to the mix.

Edited December 28, 2015 by obobski

[Athanasa]

I'm entirely unfamiliar with "CORS" as a PSU maker - I hate to say it but it could be some no-name gutless wonder, in which case replacing it may be a good idea. You'd have to do some more research on that to find out though. Basically the problem here could be, that no-name/cheapo PSUs usually lie on their ratings, so they'll claim they can do say 1000W out, but in reality its the guts of a 400W PSU and when the system tries to load it to 1000W it may blow up, it may shut down, etc. A PSU from a reputable/quality manufacturer (like EVGA, Thermaltake, Antec, PC Power & Cooling, Zippy, Seasonic, etc) should not behave in that way, and will provide its rated power (and often at higher than normal temperatures, which is a good thing too). Use jonnyguru as a resource for PSU reviews, as they're very thorough. Don't, however, run from the monitor screaming in fear that your PSU is bad - I'm located in the US (and therefore know US brands and retailers), but I know there are different brands in Europe and Asia that are also reputable (in some cases just different trade dress of the same product), so for all I know "CORS" is tip-top stuff for your region of the world, and perfectly fine.

 

If you want to look at some competition to Asus, give Gigabyte, PNY, and EVGA a look. If you're looking at Radeon cards as well, add Sapphire, XFX, and PowerColor (TUL Corporation) to the mix.

 

Ah. CORS = Corsair, I think. Although there's somehow a dead moth in mine now. Corsair seems to have a good reputation for PSUs? I think - but I don't remember - that I picked my PSU specifically to be 'growth friendly' so I could expand my computer further at a later date. The same way I grabbed a load of RAM on special offer.

 

I'm not much of a fan of Gigabyte. Mostly because my MoBo had faults, and they were pretty unhelpful with it. The USB ports used to flicker rapidly, turning on-off-on-off. I thought it might be my peripherals, but anything in the affected ports would do it. Luckily, it stopped after a couple of months, but it crippled my keyboard / mouse when it happened.

 

I'm still curious as to why Scan rates the 4GB cheaper version of the GPUs as mid-range while the others are "high end". There's hardly any difference between it and the top one that I can tell. 38 less MHz... oh, I guess the boost is lower too. Hmm. What does the Boost even do?

 

 

Obsessive colour coding time!

 

£205 Most Expensive - Listed as High End.

4GB Asus GTX 960 STRIX DirectCU II OC, PCIe 3.0, 7010MHz GDDR5, GPU 1228MHz, Boost 1317MHz, Cores 1024, 3x DP/HDMI/DVI-I

2GB more VRAM than 2GB. 190 fewer Mhz GDDR5 (VRAM speed?) than the 2GB. 25 MHz slower GPU than 2GB. 64MHz Faster 'Boost' than Mid-Range (whatever that is)

£169 2GB, High End - Listed as high end on Scan.

2GB Asus GTX 960 STRIX DirectCU II OC, PCIe 3.0, 7200MHz GDDR5, GPU 1253MHz, Boost 1317MHz, Cores 1024, 3x DP/HDMI/DVI-I

2GB less VRAM than 4GB. 190MHz faster VRAM speed. Fastest GPU speed. Equal Boost to £206, 64MHz faster 'boost' than Mid-Range.

£169 4GB VRAM - Listed as mid-range on Scan.

4GB Asus GTX 960 TURBO OC(?), PCIe 3.0 (x16), 7010MHz GDDR5, GPU 1190MHz, Boost 1253MHz, Cores 1024, 3x DP/HDMI/DVI-I DL

WTF is Turbo OC? 2GB more VRAM. Slower VRAM than 2GB. Slowest GPU speed of the lot, 109MHz down from £205. Lowest Boost. (Whatever THAT does)

Edited December 28, 2015 by Athanasa

[obobski]

 

 

Ah. CORS = Corsair, I think. Although there's somehow a dead moth in mine now.

 

I'm not much of a fan of Gigabyte. Mostly because my MoBo had faults, and they were pretty unhelpful with it. The USB ports used to flicker rapidly, turning on-off-on-off. I thought it might be my peripherals, but anything in the affected ports would do it. Luckily, it stopped after a couple of months, but it crippled my keyboard / mouse when it happened.

 

I'm still curious as to why Scan rates the 4GB cheaper version of the GPUs as mid-range while the others are "high end". There's hardly any difference between it and the top one that I can tell. 38 less MHz... oh, I guess the boost is lower too. Hmm. What does the Boost even do?

 

 

Oh, Corsair is totally fine - that PSU should be no problem, unless the moth damaged something. :laugh:

 

No pressure on Gigabyte - they're one of the "better" brands but there are alternatives. Since you're in Europe (and I always forget this exists), you should have Gainward and Leadtek too - also fantastic choices for nVidia cards.

 

No idea why Scan rates what it does like it does; there really isn't a significant difference between those cards. As far as "Boost" - newer nVidia GPUs have a feature known as "GPU Boost." It's a dynamic clocking scheme that acts as a power and thermal management mechanism - the GPU will lower its clocks when idle or not doing very much (e.g. play an old game), and will raise its clocks when working harder. The first # listed as GPU clockspeed is the nominal clockspeed that Boost will target, the second number is the maximum number it will target - it will aim for the maximum until/unless it reaches a thermal limit at which point it will back off of it (trying first the nominal value, and then lower values as needed to cool the GPU). With a decent cooling solution it may run at the Boost clock more or less continuously under load, if the load is demanding enough (my GTX 660 could achieve this without fuss, for example). It'd be less confusing if they didn't provide the nominal clocks, but my guess is they do it for legal reasons, since the Boost clock is considered "an ideal situation" and may not be possible in all settings/scenarios/loads/etc.

 

nVidia has a page about GPU Boost on their website if you want to know more:

http://www.geforce.com/hardware/technology/gpu-boost-2/technology (they're using "base" to refer to "nominal" but I consider "base" inaccurate - the actual clock floor is much lower, and that's what the GPU will target when its idle or not working very hard; with newer nVidia drivers this is done at a very granular level, so it can make adjustments more or less MHz-by-MHz whereas older drivers tended to move in larger "steps" (e.g. it'd jump 300MHz at a time), similar to how older technologies handled this functionality).

[Athanasa]

No idea why Scan rates what it does like it does; there really isn't a significant difference between those cards. As far as "Boost" - newer nVidia GPUs have a feature known as "GPU Boost." It's a dynamic clocking scheme that acts as a power and thermal management mechanism - the GPU will lower its clocks when idle or not doing very much (e.g. play an old game), and will raise its clocks when working harder. The first # listed as GPU clockspeed is the nominal clockspeed that Boost will target, the second number is the maximum number it will target - it will aim for the maximum until/unless it reaches a thermal limit at which point it will back off of it (trying first the nominal value, and then lower values as needed to cool the GPU). With a decent cooling solution it may run at the Boost clock more or less continuously under load, if the load is demanding enough (my GTX 660 could achieve this without fuss, for example). It'd be less confusing if they didn't provide the nominal clocks, but my guess is they do it for legal reasons, since the Boost clock is considered "an ideal situation" and may not be possible in all settings/scenarios/loads/etc.

 

Ooooh, that makes sense. So there's not much point getting the really expensive one - it's basically diminishing returns, and I probably won't need the power anyway. And the extra VRAM on the middle one (4GB) should pad out any differences with clock speeds or something?

 

Noice, thankies.

[obobski]

 

 

Ooooh, that makes sense. So there's not much point getting the really expensive one - it's basically diminishing returns, and I probably won't need the power anyway. And the extra VRAM on the middle one (4GB) should pad out any differences with clock speeds or something?

 

Noice, thankies.

 

 

Yes it is diminishing returns for the >£200 one; you may be able to get the next-tier GTX 970 for that kind of price (I don't know you'd have to look), which would be faster. The extra RAM will not, however, "pad out the differences" - the overclocked 2GB model will perform slightly faster than the same-priced 4GB model (and on-par with the overclocked 4GB model, if not a bit faster, owing to its faster memory), but it makes more sense to me to go for the extra memory than the slightly higher clocks. That said, on a 1680x1050 monitor, your memory requirements will be somewhat lower than a more common 1080p monitor would want.

 

Give this comparison a look:

http://www.gamersnexus.net/guides/1888-evga-supersc-4gb-960-benchmark-vs-2gb/Page-2

 

For the few games they had that actually needed more VRAM at high resolutions/settings, the 4GB card had a slight edge, but otherwise the performance is similar (and notice at lower resolutions the differences get even smaller; and their "low resolutions" tend to be 1080p or 1440p; at 1680x1050 you're somewhat below that too).

 

Those cards also don't have equal clocks, which will impact performance results. The EVGA is clocked 1279/7010 with a GPU boost of 1342; the Asus is clocked 1253/7200 with a GPU boost of 1317 (using the Asus OC software; stock it is 1228/1291) - the differences in some of the Gamers Nexus results are likely due to the EVGA card being slightly faster (e.g. the 1-2 FPS variations at 1080p in some games), whereas the big differences in other games can likely be (properly) attributed to VRAM (e.g. AC Unity). Those small differences serve as a reasonable illustration of the diminishing returns on the overclocked card.

Edited December 28, 2015 by obobski

[Thaneize]

I often read that people cant decide between the 960 and the 380X

 

380X:

- 2048 vs. 1024 Stream Cores

- 256 Bit vs. 128 Bit

- OpenGL 4.5 vs. OpenGL 4.4

- 4K is possible (if it makes sense is another thing^^)

 

960:

- 7000 MHz VRAM vs. 6000 MHz

- 1228 MHz GPU Speed vs. 1040 MHz

- Max Power 120W vs. 220W

- PhysiX

 

Both cost the same so i would go with the most expensive one of the 960s because 4GB VRAM are very nice to have if you look at the future, Devs tend to oversize their textures either because they are lazy or want to push sales of VCards, with a 2GB card you arent really futureproof because there are games which require a 2GB card for minimum today.

[Athanasa]

Huh. So... which protects me better from my framerates tanking random dumps on me? Which can I run with super shiny shadows, godrays and particles out the wazoo?

[obobski]

Huh. So... which protects me better from my framerates tanking random dumps on me? Which can I run with super shiny shadows, godrays and particles out the wazoo?

 

It depends on the game and settings and blah blah - the 960 is a fine performer, but performance with mods cannot be guaranteed for any platform.

[Athanasa]

It depends on the game and settings and blah blah - the 960 is a fine performer, but performance with mods cannot be guaranteed for any platform.

 

Awesome. So, double checking my understanding - 4GB VRAM is better for if I want to play with shiny HD textures? But do stuff like sunshafts and shadows come from clock speed?

Edited December 28, 2015 by Athanasa