As we look at the specifications, right off the bat we can see that EVGA has factory overclocked the card from reference design speeds. The reference design card has a base clock of 1020 MHz and a boost clock of 1085 MHz, but here we have 1189 MHz and 1268, respectively. I’d say that’s a pretty hefty overclock right from the start. The card also features 2 GB of GDDR5 memory set to a quad-pumped speed of 5400 MHz. Another added bonus is the inclusion of a 6-pin PCI-E power connection, whereas there is no such connector found on the reference design cards. Specifications provided by EVGA. A quick glance at GPU-Z confirms much of what we see above. History tells us that the actual boost clock usually comes in quite a bit higher than the official specification. This held true with this card, too, and it actually boosted up to 1345 MHz when under load. For features, we first have a list of those that are more related to NVIDIA and commonly found on most of their newer GPUs. Specifications Graphics Processing Clusters 1 Streaming Multiprocessors 5 CUDA Cores 640 Texture Units 40 ROP Units 16 Base Clock 1189 MHz Boost Clock 1268 MHz Memory Clock (Data Rate) 5400 MHz L2 Cache Size 2048 KB Total Video Memory 2048 MB GDDR5 Memory Interface 128-bit Total Memory Bandwidth 86.4 GB/s Texture Filtering Rate (Bilinear) 40.8 GigaTexels/s Fabrication Process 28 nm Transistor Count 1.87 Billion Connectors 1 x Dual-Link DVI 1 x HDMI 1 x DisplayPort 1.2 Form Factor 2 Slots Power Connectors 1 x 6-Pin PCI-E Recommended Power Supply 400 Watts/20A +12V Thermal Design Power (TDP) 60 Watts Thermal Threshold 95 ℃ We already mentioned the GTX 750 Ti FTW is outfitted with the ACX cooler. I’ve tested EVGA cards in the past that used the ACX cooler and have always came away impressed with how well it works. On the software side, EVGA’s Precision X is quickly becoming the go-to utility for overclocking NVIDIA based GPUs. OC Scanner X is another useful tool for checking overclock stability and monitoring information. When you buy into the EVGA product line, you get one of the best online community experiences found anywhere. From EVGA’s game servers and social networking to 24/7 tech support and forums, you can quickly find help when needed. KEY FEATURES NVIDIA TXAA Technology NVIDIA GPU Boost 2.0 NVIDIA PhysX Technology NVIDIA FXAA Technology NVIDIA Adaptive Vertical Sync NVIDIA Surround Support for three concurrent displays; dual-link DVI, HDMI, and DisplayPort 1.2 Microsoft DirectX 11.2 API (feature level 11_0) with Direct Compute 5.0 support NVIDIA 3D Vision-Ready NVIDIA CUDA Technology PCI Express 3.0 Support OpenGL 4.4 Support OpenCL Support NVIDIA SHIELD-Ready NVIDIA G-SYNC-Ready I attended a web conference with NVIDIA on February 13th, so they could present information on the new Maxwell platform. I’d be remiss if I didn’t at least pass on some of that information here. It does, in fact, bring quite a few new features to the GPU world. The first item of note is that the GTX 750 Ti and GTX 750 will be the first to use the new Maxwell GM107 GPU core. The GTX 750 Ti is aimed squarely at AMD’s R7 260X, while the GTX 750 stacks up against the R7 260. The GeForce GTX 750 Ti will replace the GeForce GTX 650 Ti in NVIDIA’s GPU lineup, but the GTX 650 and GTX 660 will continue to be produced. Buying into the NVIDIA GTX family of graphics cards allows the user to take advantage of several GTX Gaming technologies. ShadowPlay, G-Sync, and SHIELD are just a few of the unique NVIDIA technologies that promise enhanced gaming options not found elsewhere. Whether your intent is to stream game play, obtain smooth and stutter free visuals, or even make a video of your game session, you’ll find all the tools you need to accomplish this within the NVIDIA ecosystem. As NVIDIA put it during the web conference, “The soul of Maxwell is improving performance per watt.” The Maxwell architecture brings a new design to the Streaming Multiprocessor (SM) that improves performance per watt and performance per area. Logic partitioning, workload balancing, clock-gating granularity, number of instructions issued per clock cycle, and compiler-based scheduling are just a few of the many improvements over the Kepler architecture. The number of SMs has increased to five compared to Kepler’s two and done so with only a 25% increase in die area. The L2 cache sees a huge increase from Kepler’s 256 KB to Maxwell’s 2048 KB. With more L2 cache on chip, there is a large drop in requests to the DRAM, which reduces power demand and improves overall performance. To further maximize energy efficiency, NVIDIA states the engineers have aggressively tuned the implementation of each unit in the Maxwell GPU down to the transistor level. In a nutshell, this all boils down to NVIDIA’s claim that Maxwell-based GM107 GPUs can deliver two times better performance per watt when compared to a Kepler based GK107 GPU, while continuing to use the same 28nm manufacturing process. One GPC, five Maxwell Streaming Multiprocessors (SMM), and two 64-bit memory controllers (128-bit total) make up the contents of the GM107 GPU. This is the uncut implementation of the GM107 and is what’s found on the GTX 750 Ti. Below is the full chip block diagram showing the GPU configuration. For a down and dirty comparison, here is a chart summarizing the major improvements that the Maxwell GPU provides over the Kepler GPU. Keep in mind, a few of these numbers vary depending on NVIDIA partner card designs, especially when it comes to clock speeds. I wanted to touch on two other aspects I found appealing about the Maxwell based GTX 750 Ti GPU. The first being how it stacks up against the R7 260X in performance and efficiency. Based on the following graphs, NVIDIA claims the GTX 750 Ti outperforms the R7 260X with just a fraction of the power. GPU GK107 (Kepler) GM107 (Maxwell) CUDA Cores 384 640 Base Clock 1058 MHz 1020 MHz GPU Boost Clock N/A 1085 MHz GFLOPs 812.5 1305.6 Texture Units 32 40 Texel fill-rate 33.9 GigaTexels/s 40.8 GigaTexels/s Memory Clock 5000 MHz 5400 MHz Memory Bandwidth 80 GB/s 86.4 GB/s ROPs 16 16 L2 Cache Size 256 KB 2048 KB TDP 64W 60W Transistors 1.3 Billion 1.87 Billion Die Size 118 mm2 148 mm2 Manufacturing Process 28 nm 28 nm Secondly, the extremely low power draw makes it an attractive option for HTPC, mITX, or basic store-bought systems that may have smaller wattage PSUs. If you buy a reference design GTX 750 Ti, there won’t be a 6-pin power connector needed, which adds a wider variety of potential usage options. Currently, the GTX 750 Ti is the fastest available graphics card that does not have a power connector. As is the case with today’s review sample, you’ll probably find that NVIDIA partners will add a 6-pin power connector to some of their Maxwell models. The other item that makes the GTX 750 Ti an easy upgrade for a wide variety of systems is the card’s physical size. In its reference form, the card only measures 5.75′′ long. Even the EVGA GTX 750 Ti FTW only measures out at a smidgen over 9′′ long. Either way, the card will fit into almost any system design out there. As you can see, the new Maxwell GPU brings a lot of new technologies to the table, and it should make for interesting times ahead! HARDWARE The information on the box does a nice job of explaining the product found within. Looking at the box front, you see a few high-level features mentioned, as well as that the card is in the “FTW” family. Around back, the features and specifications we talked about above are displayed. The box sides are a placard for additional branding and a multilingual list of key features. Inside the box, you will find the GTX 750 Ti FTW well protected in a bubble wrap envelope. Also included are lots of documentation, a DVI to VGA adapter, a PCI-E adapter cable, and the driver/utility CD. EVGA also tosses in a poster, case badge, and a couple different stickers. The first thing worth noting is the lack of SLI support on the GTX 750 Ti series of cards. It’s understandable that SLI wouldn’t be a feature of this card, as it’s probably not in the class of cards most would consider for that type of setup anyway. The card supports up to three concurrent displays through the available dual-link DVI, HDMI, and DisplayPort connections. As we mentioned earlier, EVGA opted to include a 6-pin PCI-E power connector. Adding this power connector is said to provide a 25 watt boost (30% increase) in power delivery, which should improve overclocking potential. The ACX cooler features a dual fan design that sits on the aluminum fin stack. There are two large copper heat pipes that pass through the cooling block as they weave their way through the fin stack. This isn’t the most robust implementation of the ACX cooler I’ve seen, but given the smaller size of the card and the low power draw that the Maxwell GPU has, it should work just fine. We’ll find out later in the review. With the ACX cooler removed, we can get a clear view of the PCB layout. It appears we have a 3+1 power phase design implemented here (3 GPU + 1 Memory). The 2 GB of onboard memory are courtesy of the Samsung K4G41325FC-HC03 GDDR5 memory ICs. Finally, we have a close-up look at the Maxwell GM107 GPU. SOFTWARE Precision X has evolved into one of the better GPU overclocking utilities over the past couple of years. It allows for real time monitoring of vital GPU information, overclocking capabilities, and the ability to control the ACX cooler fans. The fan control option allows for setting the speed manually or by using the fan curve option. In the case of this particular video card, there is no power target adjustment available. However, we do have the option to raise the temperature target as high as 95 °C. There is a slight core voltage adjustment available as well, but it’s a mere +31 mV. Another item of note is the ability to save up to ten different profiles. To compliment Precision X, you might want to get a copy of OC Scanner X, too. It’s a quick way to test GPU overclocking stability and check for any artifacting issues. It even comes with its own monitoring capabilities. Plates 41-45 are images of the EVGA GTX 750 Ti FTW taken from different angles as well as the Precision X overclocking software. We’ll have a close-up look at the card as the review progresses. OVERCLOCKING It was mentioned by NVIDIA that the GTX 750 Ti would overclock to 1250 base clock without much trouble, and that’s exactly where it landed, while still being able to complete all of the tests in our suite. That might not sound like a lot compared to the 1189 MHz clock the card is factory overclocked to, but when compared to the reference clock of 1020 MHz, it’s actually pretty impressive. The actual boost clock landed at 1407 MHz when the GPU was overclocked to 1250 MHz. On the memory side, I was able to push it to an additional 200 MHz, which landed us at 1550 MHz (6200 MHz quad pumped). Nothing to complain about there either! Ok, now that we have our 24/7 stable overclock established, let’s get to the benchmarks! System Components Motherboard ASUS Maximus VI Formula CPU Intel i7 4770K Haswell Memory G.SKill TridentX DD3-2666 MHz 2x4 GB SSD Samsung EVO 500 GB Power Supply Corsair HX 1050 Professional Series Video Card EVGA GTX 750 Ti FTW w/ ACX Cooler Cooling Swiftech Apogee HD CPU Water Block - 360 mm Radiator - MCP35X Pump For the sake of comparison, I’ve chosen four different new generation AMD cards. Because NVIDIA claims better performance than the R7 260 series cards, we’ll use the R7 260X and R7 260 in the comparison graphs. Taking a step up in price and performance, I also used the R9 270 and R9 270X for comparison. How close can the mid-range level GTX 750 Ti FTW come to the higher-priced R9 270 series cards? Let’s find out. Synthetic Tests 3DMark Vantage DirectX 10 benchmark running at 1280X1024 – Performance preset 3DMark 11 DirectX 11 benchmark running at 1280X720 – Performance preset 3DMark Fire Strike DirectX 11 benchmark running 1920X1080 – Standard test (not extreme) Unigine Heaven (HWBot) DX11 Benchmark – Extreme setting Our synthetic tests show a pecking order that actually held true throughout our entire test suite. The EVGA GTX 750 Ti FTW had no trouble topping the R7 260 and R7 260X in all the tests. Once it was overclocked, it closed the gap substantially when compared to the R9 270 and R9 270X. Game Tests Batman: Arkham Origins Vantage 1920X1080, 8x MSAA, PhysX off, V-Sync off, The rest set to on or DX11 enhanced Battlefield 4 1920X1080, Ultra Preset, V-Sync off Bioshock Infinite 1920X1080, Ultra DX11 preset, DOF on Crysis 3 1920X1080, Very high settings, 16x AF, 8x MSAA, V-Sync off Final Fantasy XIV: A Realm Reborn 1920X1080, Maximum preset Grid 2 1920X1080, 8x MSAA, Intel specific options off, everything else set to highest available option Metro Last Light 1920X1080, DX11 preset, SSAA on, Tessellation very high, PhysX off The game benchmarks show the same pattern with the EVGA GTX 750 Ti FTW topping the R7 260 series cards handily. Of note here are the Bioshock Infinite results, where the EVGA GTX 750 Ti FTW actually topped the R9 270 at stock speed. Then, when overclocked, it topped the R9 270X just for good measure! The Battlefield 4 and Crysis 3 results also show this card holding very close to the R9 270 series cards. If you were paying attention, you will have noticed that every game surpassed the 30 FPS threshold we call “playable,” except for Crysis 3 and Metro: Last Light. That’s some pretty impressive stuff for a card in this class. In the end, a pretty sweet showing here for the EVGA GTX 750 Ti FTW! POWER CONSUMPTION AND TEMPERATURES With the aid of a Kill-A-Watt meter, I used HWBot Heaven and the Combined Test within 3DMark 11 to get the maximum power draw from the video card. I’ve said it before, and I’ll say it again, I’m amazed when I see how little power these modern PCs use. Even with the EVGA GTX 750 Ti FTW overclocked and under load, the total system draw still remained under 200 watts. When idle, the entire system used no more wattage than your standard 100 watt light bulb. Yea, amazing! Equally impressive are the temperatures this card runs at. With temperature readings normalized to 25°C ambient, the card never even sniffed the threshold temperatures. While it’s true we didn’t have a lot of voltage manipulation at our disposal, it’s easy to see that the thermal improvements that Maxwell GPUs bring forward do indeed work quite well. With temperature readings like these, there is no reason to take the fan off of auto control... even if overclocked. The ACX cooler performs quite admirably, to say the least. PUSHING THE LIMITS I wasn’t able to get a whole lot more out of the card before stability issues arose, but I did manage a 3DMark Fire Strike run with an additional 15 MHz added to the GPU and 25 MHz added to the memory speed. This gave us just short of another 100 points added to our previous overclocked score. I must admit, I was thoroughly impressed with the EVGA GTX 750 Ti FTW from the moment I started working with it. From impressive overclocking to the great performance of the ACX cooler, everything worked terrific right out of the box. The improvements that the Maxwell GPU brings are impressive, especially on the performance per watt front. EVGA has done a great job with their first implementation of a Maxwell-based graphics card. The card is aesthetically pleasing and will fit into just about any PC application I can think of, which allows a good gaming experience to be brought to systems that couldn’t before handle the power demands.