Latest News Posts

Social
Latest Forum Posts

NVIDIA GeForce GTX 750 Ti Review: 1080p Gaming without a Power Connector
Bookmark and Share

NVIDIA GeForce GTX 750 Ti
Print
by Rob Williams on February 24, 2014 in NVIDIA-Based GPU

It’s often hard to get excited about a new $149 graphics card, but NVIDIA’s GeForce GTX 750 Ti becomes one of the rare exceptions. For starters, it doesn’t require a power connector, and it has half the TDP requirement of its nearest competitor – all despite promised performance improvements. What more can be said? Read on!

Test System & Methodology

At Techgage, we strive to make sure our results are as accurate as possible. Our testing is rigorous and time-consuming, but we feel the effort is worth it. In an attempt to leave no question unanswered, this page contains not only our test-bed specifications, but also a detailed look at how we conduct our testing.

Test Machine

The below table lists our testing machine’s hardware, which remains unchanged throughout all GPU testing, minus the GPU. Each card used for comparison is also listed here, along with the driver version used.

  Graphics Card Test System
Processors Intel Core i7-4960X – Six-Core @ 4.50GHz
Motherboard ASUS P9X79-E WS
Memory Kingston HyperX Beast 32GB (4x8GB) – DDR3-2133 11-12-11
Graphics AMD Radeon R7 260X 2GB – Catalyst 13.11 (GPU-Z)
NVIDIA GeForce GTX 750 Ti 2GB – GeForce 334.89 (GPU-Z)
Audio Onboard
Storage Kingston HyperX 240GB SSD
Power Supply Cooler Master Silent Pro Hybrid 1300W
Chassis Cooler Master Storm Trooper Full-Tower
Cooling Thermaltake WATER3.0 Extreme Liquid Cooler
Displays ASUS PB278Q 27″ 2560×1440
Dell P2210H 22″ 1920×1080 x 3
Et cetera Windows 7 Professional 64-bit

Notes About Our High-end System

The goal of our performance content is to show you as accurately as possible how one product compares to another – after all, you’re coming to us for advice, so we want to make sure we’re giving you the best possible information. Typically, one major step we take in ensuring that our performance results are accurate is to make sure that our test systems are void of all possible bottlenecks, so for that, high-end components must be used.

In the case of our graphics card test system, the processor chosen has six-cores and is overclocked far beyond reference clocks. Most games nowadays are not heavily CPU-bound, but by using such a chip, we feel that we completely rule it out as a potential bottleneck. The same can be said for the use of an SSD (as opposed to latency-ridden mechanical storage), and even our memory, which is clocked at the comfortable speed of DDR3-2133.

Why this matters to you: Our test PC is high-end, and it’s very likely that you’d encounter a bottleneck quicker than us. Our goals are to rid all possible bottlenecks, whereas yours is to build the PC you need. In our case, we need to go overboard to attain as accurate a representation of a graphic card’s performance as possible.

If your PC has at least a modern (~2-years-old) quad-core or better processor, and at least 8GB of fast memory (DDR3-1866+), that chances of you running into a bottleneck with today’s hottest game is admittedly low. If you’re using lower-end gear, you can absolutely expect that the rest of your system could be a bottleneck. It should be noted, though, that if you’re seeking out a lower-end graphics card, the importance of a bottleneck would of course be lessened.

Unfortunately, we’re not able to test a single card on multiple PC configurations; each single card we test takes at least 3 hours to test, with another 2 hours added on for each additional resolution, and at least another 1~2 hours for our Best Playable results (for up to 11 hours of mostly hands-on testing for a high-end model).

Please bear all of this in mind. If you’re unsure if your PC could prove to be a bottleneck, our comments section exists for such questions.

When preparing our test-beds for any type of performance testing, we follow these guidelines:

General Guidelines

  • No power-saving options are enabled in the motherboard’s BIOS.
  • No virus scanner or firewall is installed.
  • The OS is kept clean; no scrap files are left in between runs.
  • Machine has proper airflow and the room temperature is 80°F (27°C) or less.

To aid with the goal of keeping accurate and repeatable results, we alter certain services in Windows 7 from starting up at boot. This is due to the fact that these services have the tendency to start up in the background without notice, potentially causing inaccurate test results. For example, disabling “Windows Search” turns off the OS’ indexing which can at times utilize the hard drive and memory more than we’d like.

The services we disable are:

  • Windows Defender
  • Windows Firewall
  • Windows Search
  • Windows Update

For further fine-tuning, we also use Windows’ “Classic” desktop theme, which gets rid of the transparency that can sometimes utilize a GPU in the background.

Vendor Favortism

Sometimes, either AMD or NVIDIA will work with a game studio to help their development process along. As history has proven, this often results in a game that is tuned better for one vendor over the other, although sometimes the tides can change over time, resulting in the competing vendor offering the better experience.

One of our goals is to provide as neutral a benchmarking suite as possible, so while it’s impossible to avoid games sponsored by either of these companies, we can at least make an effort to achieve a blended list. As it stands, our current game list and their partners are:

(AMD) – Battlefield 4
(AMD) – Crysis 3
(AMD) – Sleeping Dogs
(NVIDIA) – Assassin’s Creed IV: Black Flag
(NVIDIA) – Metro: Last Light
(NVIDIA) – Splinter Cell Blacklist
(Neutral) – GRID 2
(Neutral) – Total War: SHOGUN 2

With that, let’s move on to a quick look at the game settings we use in our testing:

Assassin’s Creed IV: Black Flag

Assassin's Creed IV Black Flag Benchmark Settings

Battlefield 4

Battlefield 4 Benchmark Settings

Crysis 3

Crysis 3 Benchmark Settings

Crysis 3 Benchmark Settings

GRID 2

GRID 2 Benchmark Settings

GRID 2 Benchmark Settings

GRID 2 Benchmark Settings

Metro Last Light

Metro Last Light Benchmark Settings

Sleeping Dogs

Sleeping Dogs Benchmark Settings

Sleeping Dogs Benchmark Settings

Splinter Cell Blacklist

Splinter Cell Blacklist Benchmark Settings

Splinter Cell Blacklist Benchmark Settings

Total War: SHOGUN 2

Total War SHOGUN 2 Benchmark Settings

Unigine Heaven

Unigine Heaven 4 Benchmark Settings


  • Casecutter

    Rob, very nice write-up. Nice to see the “playable setting” something l’d like to see more at this level of card. Also I’ve never seen a review state the “Vendor Favoritism” to what group (AMD/Nvidia) that assisted in backing the release… kudos! One thing, I know this reasons you work from the i7 (and OC’d), but I believe it should be stated that those Max settings results, especially when it comes to minimum frames from either card would in actuality take a notable hit, to in some cases not offer playable results. Most buying this level card at top would mean working from some i5-4440 at minimum, while plenty of older i3 and Phenom II X4 like 945 Deneb 3.0GHz.

    As to power I’m surprised that the power delta under load wasn’t more, I mean it’s really only 10-12% difference. It’s good but IDK, considering the R7 260X gives you ZeroCore, which in today’s world when most “sleep” their computer that 3-5W drop over days will adds up… more than the 10% when gaming a few times a week.

    Another is the fact that the basic versions that held the $150 price point have evaporated (some say “sold-out” but either way I’d see them as a rare birds anymore) and now all that out ther are the AIB customs with as you put it the “beefier-looking coolers”. The problem with that is Newegg is pricing them at $170-180 now. Sure that perhaps the new-ness factor, let’s hope that price is tempered a little over the next few weeks. Heck with Nvidia work from a 7% smaller die they should be able to be more value oriented than the R7 260X, which today is like $120-130 even one at $110 working a $20 rebate. That 40% difference pays for a lot of electricity.

    Here my thinking you be better off dumping the old and most likely inefficient 300W (or less) PSU for something like the Corsair CX430M 80+ Bronze Modular Active PFC PSU that $30 –AR$20. Then if really entry gaming like a young teen a 260X is acceptable; want really more often higher settings/some AA see about a R7 265 or a good deal on a GTX660. If spending $170 to comprise on power as not buying as PSU is throwing good money after bad. This 750Ti is most sensible if building a HTPC, but it falls a little short on price for any gaming machine/upgrade.

    • http://techgage.com/ Rob Williams

      Thanks for the detailed comment, once again!

      “kudos”

      You deserve the kudos for actually noticing :-)

      “I know this reasons you work from the i7 (and OC’d)”

      I agree. I’ll add a note to the page soon about that, and keep that mention there in future content. I’ve been questioned about the decision to use high-end gear like that, but at the end of the day, the goal is to rid all bottlenecks (as it seems you are completely aware). I actually think we’re reaching a time where the CPU can be more of a bottleneck to a game than some people give credit, so it sounds like the premise for an article down the road.

      “As to power I’m surprised that the power delta under load wasn’t more”

      You’re not alone; basic logic would suggest that with the 260X being a 115W TDP card, and the 750 Ti a 60W one, we’d see more than a 31W delta, but not so. The reason could be that the reported TDPs are inaccurate, or there’s simply something else at play. Admittedly, I report the maximum value spotted during testing (twice over to verify), so we might very well see larger deltas if I were to record the wattage-over-time from a real-world game, and not a benchmark. Of course this would be in a perfect world; in my world I have a Kill-a-Watt.

      “Another is the fact that the basic versions that held the $150 price point have evaporated ”

      Ahh, fantastic =/ I looked at EVGA’s site and all of them have changed to “Auto-notify”. I’ll check with NVIDIA to see if I can get a reason for it, and see if a solution is en route, but I expect them to play coy as usual.

      With AMD’s inflation and now this, the GPU market has truly been put into a blender lately.

      I like your analysis at the end. It’s kind of frustrating just how much constant research people have to do to find the perfect GPU… things seem to change on a daily basis. When this article was pubbed, a $150 Ti was great; things are skewed when it becomes $180. Granted, the cards I see at Newegg are all overclocked, but even so… why on earth would there be stock of those and not the regular variants? Honestly, I find it odd that there’s OC variants of such a card at all… at $150 it’s already a bit overpriced; it just happens to offer unparalleled power consumption which helps negate that premium.

  • Casecutter

    Here’s how I see this, the 750Ti is what comprises the “entry, no 6-pin, plug-n-play market”; no different than the 5670 was back in beginning of 2010… so 4 years ago. Similar for that time 1680x was resolution of the day for the category, the 5670 could give you most titles on medium settings, but at that time it was a $75-80 upgrade with 1Gb GDDR5. In four years it’s at minimum 100% increase, that doesn’t fly!

    Against the 650Ti which MSRP for $150 I suppose it seems good, but that was overtly priced, as that used a 221mm die. Sure it was hard for Nvidia to get that down much more, but now it’s like 33% smaller and can’t provide some relief?

    If we look at what PC Perspective learned in their Upgrade Story we find that they couldn’t or didn’t feel they could provide the best graphic/playable experience most often with low settings, although Grid and Syrim provide medium that was with the best OEM box, a Core Gateway DX4885 with a i5-4440. I think working from that i5 machine or a Phenom II X4 like 945 Deneb 3.0GHz set-ups, and use a R7 250/7750 (no 6-pin), the R7 260X, and then find the best playable. I don’t consider the experience that comes across on the screen any much different between R7 250 and a GTX750Ti, while I’d say the GTX750Ti / R7 260X would basically spar with same settings and FpS. The difference the R7 260X leaves money for a nice Bronze+ PSU and Zerocore. If two twin machines… slept, browsed, and gamed identically over a month what either Kill-a-Watt record as total power used? That’s the story…

    • http://techgage.com/ Rob Williams

      You certainly remember things are lot better than I do; I curse my horrible memory sometimes. Once a new series comes out I quickly forget about the one before it.

      “Sure it was hard for Nvidia to get that down much more, but now it’s like 33% smaller and can’t provide some relief?”

      I think this comes back to the “Because it can” scenario, where it doesn’t feel compelled to lower its prices because people are paying what it’s asking. It’s better for the bottom-line, after all, to not discount prices when it’s not needed. Unfortunately, such a stance should prove to be a great thing for AMD, it not for the inflation issues. Once those pass, I’m sure NVIDIA will become price-competitive once again out of nowhere, as if nothing happened.

      I hadn’t heard about that PC Per article until now; it’s quite a good angle to tackle a card like this from. Given the way Ryan tested the systems, it’s pretty hard to compare his results to mine. That Gateway machine packs a pretty decent modern Intel quad-core (3.0GHz) with 8GB of 1600 RAM, so that to me shouldn’t prove to be too much of a bottleneck. But despite that, Crysis 3 was benchmarked @ Low, whereas I found Medium to be playable, and likewise, GRID 2 was tested at Medium, whereas I found it to be completely playable with almost maxed-out settings.

      Ryan might have been stuck between a rock and a hard place though, choosing presets that could be run across each setting. I’m not sure that gives the consumer a great idea of what the card could do when manual tweaking is involved, though. The problem with using presets is that certain settings can be applied that can cripple a game. In the case of a game like GRID 2, Ambient Occlusion and Global Illumination are sme real killers; so which would you prefer? GRID 2 @ Medium, or nearly max with 4xAA + GI/AO disabled? The same could be said for Crysis 3; I found Medium to be playable when Water, Shadows were put to Low and AA was disabled, while Ryan chose the Low preset.

      Either way, I don’t have those systems so I can’t claim that the Best Playable I found for this card would carry over perfectly to even that Gateway rig with ample Intel quad-core. It’s an interesting look, nonetheless.