For those who’re looking for a sub £100 graphics card, you’ve a hell of a lot of options. For AMD and Nvidia alike it’s one of the most lucrative sections of the graphics card market, and despite being a plethora of different options available for consumers who are looking for a budget deal, or the perfect HTC card, it can be a little tricky picking the right card.
In this review and benchmark we’ll be looking at two Gigabyte cards, one using AMD’s Radeon R7 250X chipset, and the other using Nvidia’s GT 740 GPU. Both cards are targeting the same budget range, so which of the two graphics cards are best for your cash?
Gigabyte R7 250X Specs & Details
The R7 250X isn’t based on new technology, but rather a re-badge of the Radeon HD 7770 (Cape Verde) from a few years ago, which isn’t necessarily a bad thing. We’ve already reviewed AMD’s R7 250 and found it a pretty powerful card considering the price range, and for the most part allows 1080P to run at medium settings (depending on the title). But while you might simply attribute the ‘X’ at the end of cards title to equate to a small bump in specs, in reality the R7 250 and the R7 250X are pretty different in terms of raw compute performance. While the R7 250 features only 384 Unified shaders, 8 ROPS and just 24 TMU’s, the R7 250X kicks things into high gear. We’re treated to a rather healthy bump of shader power by placing in 640 Unified Shaders, 16 ROPS and finally 40 TMU’s.
For this review sample, we’ll be using a GDDR5 version, default clock speed of 1125MHZ on a 128-Bit memory interface, providing 72GB/s memory bandwidth and 1GB of GDDR5 RAM. The motherboard’s PCI-E 75W isn’t enough for the card (which has a TDP of 80W), and thus requires an additional 6 Pin PCIE-E connection to the PSU. This particular model comes with 1GB of GDDR5, but there are 2GB models available should you decide to opt for the cash. I’d recommend this option if you plan on CrossFire down the line.
Gigabyte’s Nvidia Geforce GT 740 Specs
Nvidia’s GT 740 is a little different from AMD’s offering, as it comes in two distinct flavours of memory. The first is DDR3, and the second GDDR5, the latter clearly having a rather large memory bandwidth advantage over the former. Personally, I’d suggest steering well clear of any graphics card using DDR3, always opt for GDDR5 versions. The GDDR5 version in this case features 2GB of RAM. With the lower range cards typically lacking the raw compute power to push the settings up high enough to use up 2GB, it’s better to have ‘more’ ram than not I suppose.
Just like the R7 250X, the GT 740 will ask to be provided additional power too – a bit of a shame really. Fortunately, just like the R7, the 740 only asks for a single standard PCI-E 6 Pin cable. As for specs of the card itself, you’ll be looking at 384 CUDA cores, 32 TMU’s and finally 16 ROPS.
As I’ve said, both cards are by Gigabyte and both GPU’s came in rather attractive boxes (if you’re into that sort of thing). Both proudly display their specs, and it’s interesting how Nvidia are already marketing the GT 740 as supporting DirectX 12, albeit adding e note reading Feature level 11_0. This isn’t a lie, the cards do offer DX12 support, but then so do Intel’s HD 4400 series and AMD’s GCN architecture too.
On the subject of how the cards look – fortunately both cards are fairly small. In a comparison shot against a reference design GT 760 (I just happened to have one sitting on the desk) you can see the rather large size difference between both the GT 740 and the R7 250X. Both cards feature a non reference cooler, and just as importantly this cooler helps keep not just the GPU cool, but also the memory too. I will say the GT 740 felt a little heavier than the Radeon, but let’s be honest here – that’ll only bother folks who’re wanting to build a light weight LAN / portable HTC system, and the difference isn’t too much at all.
The system we used Radeon R7 250X and the GeForce GT 740 was our standard test setup. An Intel I-7 Haswell 4770K (slightly overclocked to 4.4GHZ), 16GB DDR3 memory (running at 1750MHZ), Windows 8.1 installed on an SSD. For drivers we’re using the latest Drivers at this time, which are the Catalyst 14.7 Beta’s for AMD, while Nvidia are using 340.52 WHQL. All of the games were using the latest versions (auto patched via their distribution platform (for example Steam). We’ll be targeting 1080P firmly with these tests and aim to find out just how well either GPU manages to fair against such demands.
For this test we decided to use the built in benchmark at two different quality settings – and of course targeting 1080P. The first test is full Ultra, DX11 and DDOF (Diffusion Depth Of Field) all enabled. R7 250X hits 25.32 FPS, which is below the ideal threshold – particularly if you were in the middle of a large fire fight. The High preset however allows the card to breath a little more, reaching 46.47 FPS.
The GTX 740 doesn’t do too badly, managing 24.09 FPS average for the Ultra, DX11 DDOF setting, and 39.99 FPS for the high setting.
|1920×1080 (average frame rate)||AMD R7 250X||GeForce GT 740|
|Metro: Last Light||24.00 FPS||19.41|
|Tomb Raider Custom Ultra + FXAA||38.1 FPS||26.1 FPS|
|Bioshock Infinite (high)||46.47 FPS||39.99 FPS|
|Bioshock Infinite (ultra, DX11, DDOF)||25.32 FPS||24.09 FPS|
|Batman Arkham Origins Max & FXAA High||39 FPS||41 FPS|
|Thief (Custom High / Medium)||.31.7 FPS (DX11)30.5 FPS (Mantle)||22.6 FPS|
|3d Mark FireStrike (Non Extreme)||3057||2014|
With Tomb Raider we play with the Ultimate Preset, and then crank down the shadows to Normal, and remove High Precision. Also gone is TressFX – it’s just too expensive for these cards to handle, and Anti-Aliasing is supplied via FXAA. AMD’s R7 250X starts things 38.1 FPS, and a min frame rate of 30 FPS. Pretty decent honestly. For some reason Nvidia’s GT 740 offers far worse performance, managing just 20.7 for min and only 26.1 FPS for the average.
Nvidia typically perform worse than AMD for Tomb Raider, but these results are certainly the furthest the GT 740 falls behind the R7 250X. It’s clear Tomb Raider isn’t taking advantage of the additional memory bandwidth the GT 740 offers, and the lower fill rate of the GT isn’t doing it favors. With that said, it’s still a commendable performance considering the price of the card.
Metro Last Light:
Unfortunately this review is taking place prior to the release of Metro Redux, so we can’t use that – but Metro Last Light is a punishing title either way. Neither card or offer here is capable of running Metro Last Light with everything at the highest settings, particularly settings such as SSAA. But with all of that said, we push the “quality” setting to Very High, Motion Blur Normal and set 16X AF. Nvidia’s hardware physx is naturally disabled, as is SSAA.
Tesselation is a little ambitious for either the R7 250X or the GT 740 however, and so we need to keep that disabled – naturally all of this is being done at the standard 1080P resolution.
Using the default built in benchmark, we manage to hit 19.41 average frames a second using the GT 740, while the R7 250X hits 24 FPS. In reality we’d recommend reducing the settings a little bit during gameplay (but the R7 250X might be okay with the heavy overclocked mentioned later on in this article) but we wanted to show off a rather intense benchmark on both cards.
Batman Arkham Origin:
Batman Arkham Origins isn’t super taxing, and both cards are more than capable of playing this at the highest settings. We do disable MSAA (settling for FXAA at highest) and of course have Nvidia’s Hardware physx disabled. The R7 hits 39 FPS average during the test, while the GT hits 41. Both cards hit the same min frame rate (27 FPS) and so there’s no difference there.
This is does give a slight nod to Nvidia’s Geforce GT 740.
Finally we have Thief, another title by Square Enix. Thief is a demanding title, and here we have custom image settings which are designed to push the cards to their limits. SSAA and Tessellation are disabled, Depth of Field (DOF) set to normal, and texture and shadow quality are set to high. We do leave Screen Space Reflections enabled, along with FXAA.
Nvidia’s GT 740 manages to eek out 22.6 FPS using these settings, which isn’t too bad considering what we’re asking the card to do. AMD’s R7 250X did manage a victory here – but it’s a bit of an odd one. Despite AMD’s R7 250X fully supporting Mantle, in our testing we’ve found a 1 FPS advantage to DirectX. We tried repeating the test a few times, but the difference was always the same. This flies in the face of previous Mantle tests conducted on an R9 280. If we had to guess it’s a simple case of a combination of the settings we’ve chosen, plus a budget GPU not allowing it to take advantage of Mantle’s features.
We will however perform some experimentation on this with BF4, Thief and other titles in the not too distant future and the low down. Regardless, Mantle manages to hit 30.5 FPS while DX11 manages 31.7.
Overclocking the Radeon R7 250X
Graphics card overclocking is typically about the easiest form of overclocking you can do – even years ago in the days of the 3DFX Voodoo 2’s all you had to do was download a small little utility – and nothing has changed a jot here. The good news, when we’re talking of the R7 250X is that is pretty damn overclockable. We didn’t do any real ‘tweaking’ here to find the highest overclocking settings possible – nor did we mess about with fan settings. Instead, we left the fan profile to default, cranked the power limit and voltage to their respective max and then played around with the GPU and Memory clock speeds.
Memory clock speeds were cranked up to 1500MHZ (not a typo!) – a huge leap from the default 1125MHZ of this particular model. We didn’t get any artifacts either – which is a testament to just how insane the memory potential on this card is. In fact, MSI’s AfterBurner wouldn’t let us push the sliders any higher, which is a bit of a shame. That’s pretty insane, it means we’re getting an effective 6000MHZ from the GDDR5 RAM. More importantly perhaps, this gives the card the shot in the arm it needs by increasing available memory bandwidth. The AMD’s R9 250X manages to reach 96GB/s of memory bandwidth, providing the overclocked card far more memory bandwidth than the 72GB/s of the default clocks.
The Core clock is slightly less impressive, but still isn’t too bad at all – particularly considering we’re leaving all of the fan settings to auto. We managed to hit 1245MHZ. We did go for higher settings but it crashed out in 3d Mark (but Tomb Raider ran fine). I’d say higher settings were possible if you were willing to mess about with the fan. To give you an idea of the type of raw GPU boost this provides, the standard clocks of 1050MHZ provide around 1.34 TFLOPS of computing power. However, if you go ahead and overclock the R7 250X to 1245MHZ you manage a smidgen under 1.6TFLOPS… a pretty damn significant boost in performance.
|1920×1080 (average frame rate)||AMD R7 250X OC||GeForce GT 740 OC|
|Metro: Last Light||28.03 FPS||20.80|
|Tomb Raider Custom Ultra + FXAA||46.6 FPS||28.0 FPS|
Overclocking the GeForce GT 740
Just like AMD’s offering, overclocking Nvidia’s cards is pretty much a snap. Once you’ve installed the graphics cards drivers your overclocking application of choice (we’re using AfterBurner from MSI) will automatically detect the default clock speeds, and you’re free to crank the clocks up to your liking. The Gigabyte GT 740 didn’t support either voltage adjustments or power limit adjustments (both were greyed out). This means the GPU cores plight of extra juice will go without answer, which has a natural and negative impact on overclocking.
While you shouldn’t buy a card for its overclocking or unlocking potential (though cards such as the old ATI Radeon 9500 which could be unlocked to a 9700 Pro were an exception), for those hoping to get extra juice out of the Kepler based GT 740 will be left disappointed. Interestingly enough, unlike just about any other Nvidia or AMD card I’ve had in the past few years, a crash on an overclock meant a restart. It wouldn’t lock the system up, the driver would partially recover to Windows desktop, but 3d applications would become unstable (even if you reset the clocks back to default). Regardless, after a few hours of playing about we managed to get a mild overclock of +90MHZ to the core and 100MHZ to the GDDR5 RAM.
Since the GT 740 is only using a 128 Bit bus, just like the R7 250X, the additional bandwidth is certainly welcome. The bandwidth raises from 80GB/s to 82.2 GB/s… hardly a revolution, but it’s certainly better than nothing. Pixel and texture fill rates naturally increase, and all of this provides a subtle bump in performance.
Performance scaled almost exactly how you’d expect, with Tomb Raider the average frame rate (using the same settings of course) manages to reach 28 FPS, which is a gain of about 2 FPS. It’s hardly an astounding improvement, but it’s better than nothing. Metro Last Light on the overclock GT 740 managed to crawl to 20.80 FPS. In other words, everything scaled perfectly given the clock speed.
GT 740 & R7 250X Tech Tribunal Verdict
Currently the R7 250X is available for $99 US dollars, which is the same price as the GT 740. Clearly unless you’ve a specific reason you want to buy an Nvidia card, the R7 250X is the faster of the two offerings. The biggest benefit we’ve found with this particular model of R7 250X was just how insanely overclockable the card is. Both cards are whisper quiet. We cranked the fans to 100 percent, and you could barely hear anything.
The black mark against both of these cards is they’re in an extremely competitive price bracket, where just 20 – 30 dollars can buy you a whole bunch of extra performance. The problem with this logic is if you plonk down the cash for the ‘next level up’, say a GTX 750 or a R7 260, another 30 dollars could net you the next level from that. In other words, if you’ve only so much cash, then you need to make a decision on the price range and stick with it.
Buying the cards:
If you’re planning on picking up either card, please consider using the below links so we get a few pennies for the Amazon referral. It costs you no extra, and provides us cash to help pay for bandwidth, hardware and buy us a cup of coffee.