EVGA GeForce GTS 450 FTW Video Card Review

With the launch of AMD's next generation of graphics cards looming, most everyone's eyes are on the upcoming HD 6000 series. Today, however, we turn our attention to the NVIDIA camp. While it's still unclear what NVIDIA is preparing in response to this new generation, it's worth pointing out the GTS 450 is just barely a month old. As we witnessed in our last GTS 450 review, the overclocked variants can easily keep up with the HD 5770 and will still have a role to play in the mainstream segment for the foreseeable future. In this article, Benchmark Reviews looks at EVGA's fastest GTS 450: the GTS 450 FTW.

The For-the-Win edition represents EVGA's highest factory overclock, surpassing their Superclocked and Super-Superclocked models. It comes in at 920/1840 MHz for the core and shader clocks, and 4104MHz for the 1 GB of GDDR5 memory. This is comparable, and a bit faster when in terms of memory speed, to the ASUS ENGTS450 TOP that we looked at last month. Considering both cards represent the highest overclock for their respective companies, it's not surprising the clock speeds are so similar. Unlike the ASUS card, though, EVGA opted for the reference design.

Looking at the latest pricing on Newegg, the EVGA model is also a bit more expensive than the ASUS. At roughly $148, it's one of the more expensive GTS 450's on the market right now. As I mentioned, though, an overclocked GTS 450 can compete directly with an HD 5770. So, while it may face some pressure among other GTS 450's, the EVGA GTS 450 FTW is fairly competitive with the HD 5770 series, which tend to range from $140 to $160.

Manufacturer: EVGA Corporation
Product Name: EVGA GeForce GTS 450 FTW
Model Number: 01G-P3-1458-TR
Price As Tested:$147.99 at Newegg

Full Disclosure: The product sample used in this article has been provided by EVGA.

NVIDIA Fermi Features

In today's complex graphics, tessellation offers the means to store massive amounts of coarse geometry, with expand-on-demand functionality. In the NVIDIA GF104 GPU (GF100 series), tessellation also enables more complex animations. In terms of model scalability, dynamic Level of Detail (LOD) allows for quality and performance trade-offs whenever it can deliver better picture quality over performance without penalty. Comprised of three layers (original geometry, tessellation geometry, and displacement map), the final product is far more detailed in shade and data-expansion than if it were constructed with bump-map technology. In plain terms, tessellation gives the peaks and valleys with shadow detail in-between, while previous-generation technology (bump-mapping) would give the illusion of detail.

Stages of Tessellation with NVIDIA Fermi Graphics

Using GPU-based tessellation, a game developer can send a compact geometric representation of an object or character and the tessellation unit can produce the correct geometric complexity for the specific scene. Consider the "Imp" character illustrated above. On the far left we see the initial quad mesh used to model the general outline of the figure; this representation is quite compact even when compared to typical game assets. The two middle images of the character are created by finely tessellating the description at the left. The result is a very smooth appearance, free of any of the faceting that resulted from limited geometry. Unfortunately this character, while smooth, is no more detailed than the coarse mesh. The final image on the right was created by applying a displacement map to the smoothly tessellated third character to the left.

What's new in Fermi?

With any new technology, consumers want to know what's new in the product. The goal of this article is to share in-depth information surrounding the Fermi architecture, as well as the new functionality unlocked in GF100. For clarity, the 'GF' letters used in the GF100 GPU name are not an abbreviation for 'GeForce'; they actually denote that this GPU is a Graphics solution based on the Fermi architecture. The next generation of NVIDIA GeForce-series desktop video cards will use the GF100 to promote the following new features:

• Third Generation Streaming Multiprocessor (SM)
  o 32 CUDA cores per SM, 4x over GT200
  o 8x the peak double precision floating point performance over GT200
  o Dual Warp Scheduler simultaneously schedules and dispatches instructions from two independent warps
  o 64 KB of RAM with a configurable partitioning of shared memory and L1 cache
• Second Generation Parallel Thread Execution ISA
  o Unified Address Space with Full C++ Support
  o Optimized for OpenCL and DirectCompute
  o Full IEEE 754-2008 32-bit and 64-bit precision
  o Full 32-bit integer path with 64-bit extensions
  o Memory access instructions to support transition to 64-bit addressing
  o Improved Performance through Predication
• Improved Memory Subsystem
  o NVIDIA Parallel DataCache hierarchy with Configurable L1 and Unified L2 Caches
  o First GPU with ECC memory support
  o Greatly improved atomic memory operation performance
• NVIDIA GigaThread Engine
  o 10x faster application context switching
  o Concurrent kernel execution
  o Out of Order thread block execution
  o Dual overlapped memory transfer engines

Tessellation in DirectX-11

Control hull shaders run DX11 pre-expansion routines, and operates explicitly in parallel across all points. Domain shaders run post-expansion operations on maps (u/v or x/y/z/w) and is also implicitly parallel. Fixed function tessellation is configured by Level of Detail (LOD) based on output from the control hull shader, and can also produce triangles and lines if requested. Tessellation is something that is new to NVIDIA GPUs, and was not part of GT200 because of geometry bandwidth bottlenecks from sequential rendering/execution semantics.

In regard to the GF100-series graphics processor, NVIDIA has added a new PolyMorph and Raster engines to handle world-space processing (PolyMorph) and screen-space processing (Raster). There are eight PolyMorph engines and two Raster engines on the GF104, which depend on an improved L2 cache to keep buffered geometric data produced by the pipeline on-die.

Four-Offset Gather4

The texture unit on previous processor architectures operated at the core clock of the GPU. On GF104, the texture units run at a higher clock, leading to improved texturing performance for the same number of units. GF104's texture units now add support for DirectX-11's BC6H and BC7 texture compression formats, reducing the memory footprint of HDR textures and render targets.

The texture units also support jittered sampling through DirectX-11's four-offset Gather4 feature, allowing four texels to be fetched from a 128×128 pixel grid with a single texture instruction. NVIDIA GF100 series GPUs implements DirectX-11 four-offset Gather4 in hardware, greatly accelerating shadow mapping, ambient occlusion, and post processing algorithms. With jittered sampling, games can implement smoother soft shadows or custom texture filters efficiently. The previous GT200 GPU did not offer coverage samples, while the GF100-series can deliver 32x CSAA.

GF106 Compute for Gaming

As developers continue to search for novel ways to improve their graphics engines, the GPU will need to excel at a diverse and growing set of graphics algorithms. Since these algorithms are executed via general compute APIs, a robust compute architecture is fundamental to a GPU's graphical capabilities. In essence, one can think of compute as the new programmable shader. GF100's compute architecture is designed to address a wider range of algorithms and to facilitate more pervasive use of the GPU for solving parallel problems. Many algorithms, such as ray tracing, physics, and AI, cannot exploit shared memory-program memory locality is only revealed at runtime. GF106's cache architecture was designed with these problems in mind. With up to 48 KB of L1 cache per Streaming Multiprocessor (SM) and a global L2 cache, threads that access the same memory locations at runtime automatically run faster, irrespective of the choice of algorithm.

NVIDIA Codename NEXUS brings CPU and GPU code development together in Microsoft Visual Studio 2008 for a shared process timeline. NEXUS also introduces the first hardware-based shader debugger. NVIDIA GF100-series GPUs are the first to ever offer full C++ support, the programming language of choice among game developers. To ease the transition to GPU programming, NVIDIA developed Nexus, a Microsoft Visual Studio programming environment for the GPU. Together with new hardware features that provide better debugging support, developers will be able enjoy CPU-class application development on the GPU. The end results is C++ and Visual Studio integration that brings HPC users into the same platform of development. NVIDIA offers several paths to deliver compute functionality on the GF106 GPU, such as CUDA C++ for video games.

Image processing, simulation, and hybrid rendering are three primary functions of GPU compute for gaming. Using NVIDIA GF100-series GPUs, interactive ray tracing becomes possible for the first time on a standard PC. Ray tracing performance on the NVIDIA GF100 is roughly 4x faster than it was on the GT200 GPU, according to NVIDIA tests. AI/path finding is a compute intensive process well suited for GPUs. The NVIDIA GF100 can handle AI obstacles approximately 3x better than on the GT200. Benefits from this improvement are faster collision avoidance and shortest path searches for higher-performance path finding.

NVIDIA GigaThread Thread Scheduler

One of the most important technologies of the Fermi architecture is its two-level, distributed thread scheduler. At the chip level, a global work distribution engine schedules thread blocks to various SMs, while at the SM level, each warp scheduler distributes warps of 32 threads to its execution units. The first generation GigaThread engine introduced in G80 managed up to 12,288 threads in real-time. The Fermi architecture improves on this foundation by providing not only greater thread throughput, but dramatically faster context switching, concurrent kernel execution, and improved thread block scheduling.

NVIDIA GF106 GPU Fermi Architecture

Based on the Fermi architecture, NVIDIA's latest GPU is codenamed GF106 and is equipped on the GeForce GTS 450. In this article, Benchmark Reviews explains the technical architecture behind NVIDIA's GF106 graphics processor and offers an insight into upcoming Fermi-based GeForce video cards. For those who are not familiar, NVIDIA's GF100 GPU was their first graphics processor to support DirectX-11 hardware features such as tessellation and DirectCompute, while also adding heavy particle and turbulence effects. The GF100 GPU is also the successor to the GT200 graphics processor, which launched in the GeForce GTX 280 video card back in June 2008. NVIDIA has since redefined their focus, allowing subsequent GF100, GF104, and now GF106 GPUs to prove their dedication towards next generation gaming effects such as raytracing, order-independent transparency, and fluid simulations.

While processor cores have grown from 128 (G80) and 240 (GT200), they reach 512 in the GF100 and earn the title of NVIDIA CUDA (Compute Unified Device Architecture) cores. GF100 was not another incremental GPU step-up like we had going from G80 to GT200. GF100 featured 512 CUDA cores, while GF104 was capable of 336 cores. Effectively cutting the eight SMUs on GF104 in half, NVIDIA's GF106 is good for 192 CUDA cores from four SMUs. The key here is not only the name, but that the name now implies an emphasis on something more than just graphics. Each Fermi CUDA processor core has a fully pipelined integer arithmetic logic unit (ALU) and floating point unit (FPU). GF106 implements the IEEE 754-2008 floating-point standard, providing the fused multiply-add (FMA) instruction for both single and double precision arithmetic. FMA improves over a multiply-add (MAD) instruction by doing the multiplication and addition with a single final rounding step, with no loss of precision in the addition. FMA minimizes rendering errors in closely overlapping triangles.

GF106 implements 192 CUDA cores, organized as 8 SMs of 48 cores each. Each SM is a highly parallel multiprocessor supporting up to 32 warps at any given time (four Dispatch Units per SM deliver two dispatched instructions per warp for four total instructions per clock per SM). Each CUDA core is a unified processor core that executes vertex, pixel, geometry, and compute kernels. A unified L2 cache architecture (512KB on 1GB cards) services load, store, and texture operations. GF106 is designed to offer a total of 16 ROP units pixel blending, antialiasing, and atomic memory operations. The ROP units are organized in four groups of eight. Each group is serviced by a 64-bit memory controller. The memory controller, L2 cache, and ROP group are closely coupled-scaling one unit automatically scales the others.

NVIDIA Fermi GF106 Block Diagram

Based on Fermi's third-generation Streaming Multiprocessor (SM) architecture, GF106 could be considered a divided GF104. NVIDIA GeForce GF100-series Fermi GPUs are based on a scalable array of Graphics Processing Clusters (GPCs), Streaming Multiprocessors (SMs), and memory controllers. NVIDIA's GF100 GPU implemented four GPCs, sixteen SMs, and six memory controllers. GF104 implements two GPCs, eight SMs, and four memory controllers. Conversely, GF106 houses one GPC, four SMs, and two memory controllers. Where each SM contained 32 CUDA cores in the GF100, NVIDIA configured GF104 with 48 cores per SM... which has been repeated for GF106. As expected, NVIDIA Fermi-series products are launching with different configurations of GPCs, SMs, and memory controllers to address different price points.

CPU commands are read by the GPU via the Host Interface. The GigaThread Engine fetches the specified data from system memory and copies them to the frame buffer. GF106 implements two 64-bit GDDR5 memory controllers (128-bit total) to facilitate high bandwidth access to the frame buffer. The GigaThread Engine then creates and dispatches thread blocks to various SMs. Individual SMs in turn schedules warps (groups of 48 threads) to CUDA cores and other execution units. The GigaThread Engine also redistributes work to the SMs when work expansion occurs in the graphics pipeline, such as after the tessellation and rasterization stages.

GF106 Specifications

• Up to 192 CUDA Cores
• 4 Geometry Units
• 1 Raster Units
• 32 Texture Units
• 16 ROP Units
• 128-bit GDDR5
• DirectX-11 API Support

GeForce 400-Series Specifications

Closer Look: EVGA GTS 450 FTW

Apart from the high factory overclock, the EVGA GTS 450 FTW essentially follows the reference design. Let's take a closer look at what that entails. First up is the overall appearance of the card, the video ports it offers, and the heatsink and fan unit. Then, with the heatsink off, we'll move into some of the more technical details as we take a closer look at the chip itself, the memory, and the power circuitry.

In the photo above, we don't find many surprises. That's not necessarily a bad thing, though. The reference design is fairly straightforward and works well. The plastic fan cover is branded with EVGA's logo and graphic and, although not seen in the photo, there's another graphic that runs along the side of the cover. You may recall from my previous review that I'm not a big fan of these large stickers, but considering this is a mainstream card, I can live with it. I should point out that the peeling you see toward the top is the protective plastic film, not the sticker itself coming off. Also, it appears EVGA updated the sticker since shipping the samples. You can see the revised design on their website.

Looking at the other side of the card, we can see the output options. These, too, follow the reference design. Two DVI ports can clearly be seen and on the far left is a mini HDMI port. The mini port isn't ideal as it will require many users who have standard HDMI cables to buy a new cable or an adapter, but I suppose that's the price we pay to upgrade. I just wish they would have thrown in an adapter like some of the other manufacturers out there.

Continuing with our tour of the EVGA GTS 450 FTW, we look again at the heatsink and fan, this time with the unit removed. We can now get a better view of the heatsink itself, which is a fairly simple design. You won't find any heatpipes or even a copper base here. It's essentially an aluminum block with the popular forked-fin design to increase surface area. We'll see how it performs a bit later on during our temperature testing.

Beneath the heatsink, we find the GF106 chip, clearly marked GF106-250-KA-A1. No surprise there, as partners do not customize the chip itself. However, as we well know, the major difference here is the clock speed. While the reference chips run at 783 MHz, the cherry-picked chips used in the FTW model, ship out at 920 MHz. That's a 17.5% increase over stock and reason EVGA called this the For-the-Win edition.

The other reason is the increased memory speed. As you can see in the photo above, the EVGA GTS 450 FTW uses the same Samsung K4G10325FE-HC05 chips as the ASUS card we looked at. These are the reference chips NVIDA went with and are rated at 4000 MHz. However, while the reference design calls for 3600 MHz, the FTW model runs these at 4104 MHz. This is roughly 100 MHz more than ASUS, and 14% higher than stock.

The overclocked GF106 chip and GDD5 memory are powered by the reference VRM design. It consists of three power phases and utilizes solid capacitors. It's not quite the four-phase design we saw from ASUS' TOP card, but it should power the EVGA without a problem. That said, let's fire this card up and get to some benchmarks.

VGA Testing Methodology

The Microsoft DirectX-11 graphics API is native to the Microsoft Windows 7 Operating System, and will be the primary O/S for our test platform. DX11 is also available as a Microsoft Update for the Windows Vista O/S, so our test results apply to both versions of the Operating System. The majority of benchmark tests used in this article are comparative to DX11 performance, however some high-demand DX10 tests have also been included.

According to the Steam Hardware Survey published for the month ending May 2010, the most popular gaming resolution is 1280x1024 (17-19" standard LCD monitors). Our benchmark performance tests typically concentrate the more demanding 1680x1050 (22-24" widescreen LCD) and 2.30MP 1920x1200 (24-28" widescreen LCD monitors) resolutions. However, as this is a mid-range video card, we've dropped the 1920x1200 benchmarks and added 1280x1024 to better capture the intended use of this card.

In each benchmark test there is one 'cache run' that is conducted, followed by five recorded test runs. Results are collected at each setting with the highest and lowest results discarded. The remaining three results are averaged, and displayed in the performance charts on the following pages.

A combination of synthetic and video game benchmark tests have been used in this article to illustrate relative performance among graphics solutions. Our benchmark frame rate results are not intended to represent real-world graphics performance, as this experience would change based on supporting hardware and the perception of individuals playing the video game.

DX11 Cost to Performance Ratio

For this article Benchmark Reviews has included cost per FPS for graphics performance results. Only the least expensive product price is calculated, and does not consider tax, freight, promotional offers, or rebates into the cost. All prices reflect product series components, and do not represent any specific manufacturer, model, or brand. These retail prices for each product were obtained from NewEgg.com on 19-October-2010:

• $130 Radeon HD 4870 512MB
• $148 GeForce GTS 450 1GB
• $150 Radeon HD 5770 1GB

AMD Phenom II Test System

• Motherboard: ASUS M4A79T Deluxe (AMD 790FX/SB750)
• Processor: AMD Phenom II X4 955 BE Deneb 3.2GHz (HDZ955FBGIBOX)
• CPU Cooler: ZEROtherm Zen
• System Memory: Crucial 1333MHz CL 9-9-9-24 DDR3 (2x2GB)
• Primary Drive: OCZ Summit MLC SSD OCZSSD2-1SUM120G
• Power Supply Unit: Corsair CMPSU-850TX 850W
• Monitor: 23-Inch Widescreen LCD (up to 2048x1152@60Hz)

DirectX-10 Benchmark Applications

• 3DMark Vantage v1.02 (Extreme Quality, 8x Multisample Anti-Aliasing, 16x Anisotropic Filtering, 1:2 Scale)
• Resident Evil 5 Benchmark (DX10, Super-High Quality, 8x MSAA)

DirectX-11 Benchmark Applications

• Aliens vs Predator (Very High Quality, 4x AA, 16x AF, SSAO, Tessellation, Advanced Shadows)
• BattleForge v1.2 (Very High Quality, 8x Anti-Aliasing, Auto Multi-Thread)
• Mafia II (Antialiasing, 16x AF, High Shadow Quality, High Detail, Ambient Occlusion, No/Medium APEX PhysX)
• Unigine Heaven Benchmark 2.1 (DX11, Normal Tessellation, 16x AF, 4x AA)

Video Card Test Products

• ASUS HD4870 TOP (815 MHz GPU/925 MHz vRAM - ATI Catalyst Driver 10.8)
• XFX HD5770 (850 MHz GPU/1200 MHz vRAM - ATI Catalyst Driver 10.8)
• ASUS GTS 450 TOP (925 MHz GPU/1850 MHz Shader/1000 MHz vRAM - Forceware 260.52)
• EVGA GTS 450 FTW (920 MHz GPU/1840 MHz Shader/1026 MHz vRAM - Forceware 260.52)

DX10: 3DMark Vantage

3DMark Vantage is a PC benchmark suite designed to test the DirectX10 graphics card performance. FutureMark 3DMark Vantage is the latest addition the 3DMark benchmark series built by FutureMark corporation. Although 3DMark Vantage requires NVIDIA PhysX to be installed for program operation, only the CPU/Physics test relies on this technology.

3DMark Vantage offers benchmark tests focusing on GPU, CPU, and Physics performance. Benchmark Reviews uses the two GPU-specific tests for grading video card performance: Jane Nash and New Calico. These tests isolate graphical performance, and remove processor dependence from the benchmark results.

3DMark Vantage GPU Test: Jane Nash

Of the two GPU tests 3DMark Vantage offers, the Jane Nash performance benchmark is slightly less demanding. In a short video scene the special agent escapes a secret lair by water, nearly losing her shirt in the process. Benchmark Reviews tests this DirectX-10 scene at 1280x1024 and 1680x1050 resolutions, and uses Extreme quality settings with 8x anti-aliasing and 16x anisotropic filtering. The 1:2 scale is utilized, and is the highest this test allows. By maximizing the processing levels of this test, the scene creates the highest level of graphical demand possible and sorts the strong from the weak.

Cost Analysis: Jane Nash (1680x1050)

3DMark Vantage GPU Test: New Calico

New Calico is the second GPU test in the 3DMark Vantage test suite. Of the two GPU tests, New Calico is the most demanding. In a short video scene featuring a galactic battleground, there is a massive display of busy objects across the screen. Benchmark Reviews tests this DirectX-10 scene at 1280x1024 and 1680x1050 resolutions, and uses Extreme quality settings with 8x anti-aliasing and 16x anisotropic filtering. The 1:2 scale is utilized, and is the highest this test allows. Using the highest graphics processing level available allows our test products to separate themselves and stand out (if possible).

Cost Analysis: New Calico (1680x1050)

Test Summary: Before we get into the numbers, I'd like to highlight the fact that we pitted three overclocked cards against a HD5770 running at reference speeds. It just happened to be what I had on hand, but it's something I will take into consideration throughout the review. If you're interested seeing how the reference GTS 450 performs, be sure to check out our GTS 450 launch article and the corresponding SLI analysis. That said, let's look at the results. In the Jane Nash benchmark, all four cards appear to hold their own, with the HD 4870 dropping slightly behind the newer generation cards. However, the New Calico benchmark tells a different story. The GTS 450's have roughly a 20-25% lead over the HD4870 and HD5770, which battle it out for last place. I'm not prepared to draw any conclusions after just one benchmark, but so far things are looking pretty good for the GTS 450.

DX10: Resident Evil 5

Built upon an advanced version of Capcom's proprietary MT Framework game engine to deliver DirectX-10 graphic detail, Resident Evil 5 offers gamers non-stop action similar to Devil May Cry 4, Lost Planet, and Dead Rising. The MT Framework is an exclusive seventh generation game engine built to be used with games developed for the PlayStation 3 and Xbox 360, and PC ports. MT stands for "Multi-Thread", "Meta Tools" and "Multi-Target". Games using the MT Framework are originally developed on the PC and then ported to the other two console platforms.

On the PC version of Resident Evil 5, both DirectX 9 and DirectX-10 modes are available for Microsoft Windows XP and Vista Operating Systems. Microsoft Windows 7 will play Resident Evil with backwards compatible Direct3D APIs. Resident Evil 5 is branded with the NVIDIA The Way It's Meant to be Played (TWIMTBP) logo, and receives NVIDIA GeForce 3D Vision functionality enhancements.

NVIDIA and Capcom offer the Resident Evil 5 benchmark demo for free download from their website, and Benchmark Reviews encourages visitors to compare their own results to ours. Because the Capcom MT Framework game engine is very well optimized and produces high frame rates, Benchmark Reviews uses the DirectX-10 version of the test at 1280x1024 and 1680x1050 resolutions. Super-High quality settings are configured, with 8x MSAA post processing effects for maximum demand on the GPU. Results from the Scene #4 test are collected for the chart illustrated below.

Cost Analysis: Resident Evil 5 (Area 4)

Test Summary: The Resident Evil benchmark really turns things upside down, with the older HD4870 taking the lead. My guess would be that is the 256-bit memory interface showing its strength. The other three cards have more processing power, but are limited by 128-bit memory interface. The other three cards are left to fight it out for second place and are fairly well matched.

DX11: Aliens vs Predator

Aliens vs. Predator is a science fiction first-person shooter video game, developed by Rebellion, and published by Sega for Microsoft Windows, Sony PlayStation 3, and Microsoft Xbox 360. Aliens vs. Predator utilizes Rebellion's proprietary Asura game engine, which had previously found its way into Call of Duty: World at War and Rogue Warrior. The self-contained benchmark tool is used for our DirectX-11 tests, which push the Asura game engine to its limit.

In our benchmark tests, Aliens vs. Predator was configured to use the highest quality settings with 4x AA and 16x AF. DirectX-11 features such as Screen Space Ambient Occlusion (SSAO) and tessellation have also been included, along with advanced shadows.

Cost Analysis: Aliens vs Predator (1680x1050)

Test Summary: Now that we're into the DirectX 11 benchmarks, the HD4870 is out and we focus in a bit more on the reference HD5770 versus our overclocked GTS 450 cards. All three cards perform well, but the GTS 450's do manage to pull ahead just a bit.

DX11: BattleForge

BattleForge is free Massive Multiplayer Online Role Playing Game (MMORPG) developed by EA Phenomic with DirectX-11 graphics capability. Combining strategic cooperative battles, the community of MMO games, and trading card gameplay, BattleForge players are free to put their creatures, spells and buildings into combination's they see fit. These units are represented in the form of digital cards from which you build your own unique army. With minimal resources and a custom tech tree to manage, the gameplay is unbelievably accessible and action-packed.

Benchmark Reviews uses the built-in graphics benchmark to measure performance in BattleForge, using Very High quality settings (detail) and 8x anti-aliasing with auto multi-threading enabled. BattleForge is one of the first titles to take advantage of DirectX-11 in Windows 7, and offers a very robust color range throughout the busy battleground landscape. The charted results illustrate how performance measures-up between video cards when Screen Space Ambient Occlusion (SSAO) is enabled.

Cost Analysis: BattleForge (1680x1050)

Test Summary: The overclocked GTS 450 cards are the clear winner when it comes to BattleForge. At 1280x1024 they maintain a 24% advantage over the HD5770 and at 1680x1050, a 22% advantage.

DX9+SSAO: Mafia II

Mafia II is a single-player third-person action shooter developed by 2K Czech for 2K Games, and is the sequel to Mafia: The City of Lost Heaven released in 2002. Players assume the life of World War II veteran Vito Scaletta, the son of small Sicilian family who immigrates to Empire Bay. Growing up in the slums of Empire Bay teaches Vito about crime, and he's forced to join the Army in lieu of jail time. After sustaining wounds in the war, Vito returns home and quickly finds trouble as he again partners with his childhood friend and accomplice Joe Barbaro. Vito and Joe combine their passion for fame and riches to take on the city, and work their way to the top in Mafia II.

Mafia II is a DirectX-9/10/11 compatible PC video game built on 2K Czech's proprietary Illusion game engine, which succeeds the LS3D game engine used in Mafia: The City of Lost Heaven. In our Mafia-II Video Game Performance article, Benchmark Reviews explored characters and gameplay while illustrating how well this game delivers APEX PhysX features on both ATI and NVIDIA products. Thanks to DirectX-11 APEX PhysX extensions that can be processed by the system's CPU, Mafia II offers gamers equal access to high-detail physics regardless of video card manufacturer.

PhysX Disabled

Cost Analysis: Mafia II (1680x1050)

Medium PhysX

Cost Analysis: Mafia II (1680x1050)

Test Summary: The Mafia II benchmark offers a great opportunity to compare the effects of GPU-based and CPU-based PhysX. In the first benchmark, with PhysX disabled, the HD5770 manages a small, but observable lead over the GTS 450's. With medium PhysX enabled, however, that lead quickly and dramatically disappears as the HD5770 system cannot run the PhysX calculations on the GPU. I won't get into the open-vs-propriety physics debate here, but suffice it to say that GTS 450 will have a significant advantage over the HD5770 when PhysX is utilized.

DX11: Unigine Heaven 2.1

The Unigine "Heaven 2.1" benchmark is a free publicly available tool that grants the power to unleash the graphics capabilities in DirectX-11 for Windows 7 or updated Vista Operating Systems. It reveals the enchanting magic of floating islands with a tiny village hidden in the cloudy skies. With the interactive mode, emerging experience of exploring the intricate world is within reach. Through its advanced renderer, Unigine is one of the first to set precedence in showcasing the art assets with tessellation, bringing compelling visual finesse, utilizing the technology to the full extend and exhibiting the possibilities of enriching 3D gaming.

The distinguishing feature in the Unigine Heaven benchmark is a hardware tessellation that is a scalable technology aimed for automatic subdivision of polygons into smaller and finer pieces, so that developers can gain a more detailed look of their games almost free of charge in terms of performance. Thanks to this procedure, the elaboration of the rendered image finally approaches the boundary of veridical visual perception: the virtual reality transcends conjured by your hand. The "Heaven" benchmark excels at providing the following key features:

• Native support of OpenGL, DirectX 9, DirectX-10 and DirectX-11
• Comprehensive use of tessellation technology
• Advanced SSAO (screen-space ambient occlusion)
• Volumetric cumulonimbus clouds generated by a physically accurate algorithm
• Dynamic simulation of changing environment with high physical fidelity
• Interactive experience with fly/walk-through modes
• ATI Eyefinity support

Cost Analysis: Unigine Heaven (1680x1050)

Test Summary: Unigine Heaven is another clear winner for the overclocked GTS 450 based cards. The 28% increase in the 1280x1024 run and 25% increase in the 1680x1050 run speak for themselves. Considering the overclock on these cards is roughly 18%, a reference GTS 450 would likely still come out slightly ahead of the HD5770. In fact, this is exactly what we found during our Unigine Heaven benchmarking last month.

EVGA GTS 450 FTW Temperatures

Benchmark tests are always nice, so long as you care about comparing one product to another. But when you're an overclocker, gamer, or merely a PC hardware enthusiast who likes to tweak things on occasion, there's no substitute for good information. Benchmark Reviews has a very popular guide written on Overclocking Video Cards, which gives detailed instruction on how to tweak a graphics cards for better performance. Of course, not every video card has overclocking head room. Some products run so hot that they can't suffer any higher temperatures than they already do. This is why we measure the operating temperature of the video card products we test.

To begin testing, we use GPU-Z to measure the temperature at idle as reported by the GPU. Next we use FurMark's "Torture Test" to generate maximum thermal load and record GPU temperatures at high-power 3D mode. The ambient room temperature remained at a stable 23°C throughout testing.

FurMark does two things extremely well: drive the thermal output of any graphics processor higher than applications of video games realistically could, and it does so with consistency every time. Furmark works great for testing the stability of a GPU as the temperature rises to the highest possible output. The temperatures discussed below are absolute maximum values, and not representative of real-world performance.

The two GTS 450 cards are fairly matched at idle, with the HD5770 coming in slightly warmer, and the HD4870 running considerably hotter. Once a load is applied, though, the heatsink and fan unit on the HD4870 show its strength. Of course, it was the loudest of the three as well. The HD5770 clearly runs the hottest, followed by the EVGA GTS450 FTW. It's clear that the DirectCU cooler used by the ASUS ENGTS450 TOP has an advantage over the reference design EVGA went with.

VGA Power Consumption

Life is not as affordable as it used to be, and items such as gasoline, natural gas, and electricity all top the list of resources which have exploded in price over the past few years. Add to this the limit of non-renewable resources compared to current demands, and you can see that the prices are only going to get worse. Planet Earth is needs our help, and needs it badly. With forests becoming barren of vegetation and snow capped poles quickly turning brown, the technology industry has a new attitude towards turning "green". I'll spare you the powerful marketing hype that gets sent from various manufacturers every day, and get right to the point: your computer hasn't been doing much to help save energy... at least up until now.

A baseline test is taken with the system in an idle state. Once the baseline reading has been taken, the loaded power consumption reading is taken with the video card running a stress test using FurMark. Below is a chart with the total system power displayed in Watts for each specified test product:

Our power consumption tests tell a much different story than the temperatures. The HD4870 is clearly the power hog at idle, while the other three cards all sit within about 9% of each other. At load, however, the HD5770 consumes much less power than the other cards. What's interesting, though, is how much more energy the ASUS GTS 450 consumes over the EVGA model. Both are overclocked by a similar amount, but the higher core voltage of the ASUS card is definitely revealed in this test. We'll see how this voltage difference affects overclocking in the next section.

EVGA GTS 450 FTW Overclocking

As with the ASUS ENGTS450 TOP, overclocking the EVGA GTS 450 FTW was a challenge. EVGA's Precision software worked well, but I ran into the same limited headroom. I also wasn't able to adjust voltage, so I left it at the stock setting. Surprisingly, though, after several rounds of adjustments and testing, I found myself at the same overclock I was able to achieve with the ASUS card.

At 945MHz, the GPU overclocked to 2.7% higher than the stock speed for the GTS 450 FTW and 21% higher than NVIDIA's reference design. The memory, which started out at 4101 MHz, managed 4200MHz, a 2.3% increase over the FTW clock and 16% over reference. As with the ASUS card, I could have pushed it higher and still made it through the benchmarks, but at the cost of severe artifacting and possible damage over time. Here are the benchmark results of the stable clock speeds I was able to achieve:

The EVGA card ships with a higher memory clock than the ASUS, and both achieved the same overclock, so a smaller percent improvement was expected. The average increase for the EVGA GTS 450 FTW was 2.2%, while the ASUS was near 3.4%. Again, though, they both overclocked to the same speeds, which is fairly impressive for the EVGA considering it's a reference design.

NVIDIA GTS 450 Final Thoughts

The GTS is definitely a good move for NVIDIA. They really needed a current generation card that could compete in this price range. Looking at the benchmark results in this review, they clearly met this goal. Our two overclocked GTS 450's managed to outperform the Radeon HD5770 in the majority of the tests we ran. Of course, as I noted at the beginning of the benchmarks, the HD5770 we used ran at reference speeds. There are a number overclocked HD5770's on the market that could (feasibly) flip the scales back in favor of the HD5770.

Where the GTS 450 really differentiates itself, though, is in price. Most GTS 450's are selling for $120 - $140, while the HD5770's are still in the $140 - $160 range. That gives the GTS 450 based cards a nice edge in terms of value. As we saw throughout the benchmarking sections, our GTS 450 cards consistently had a lower cost per fps.

With AMD rolling out its new series throughout the next couple months, though, NVIDIA would be wise make the most of this advantage while it lasts. It's not clear yet whether AMD will be launching a new card in this category right away, but when they do, NVIDIA will need to be ready to counter it. Let's hope they have something equally impressive up their sleeves.

EVGA GTS 450 FTW Conclusion

The EVGA GTS 450 FTW performed extremely well, keeping pace with the top GTS 450 offering from ASUS and frequently surpassing our reference-clocked HD 5770. There's no doubt the 17.5% GPU overclock and 14% memory gave it an excellent boost. An overclocked HD 5770 could close up the gap and pull ahead in some benchmarks, but, overall, I'd say it's a fairly even match.

Performance aside, the GTS 450 FTW earns decent marks for appearance. However, it's not quite at the level of the highly customized ASUS card. EVGA stuck with the reference design, which is adequate for the mainstream segment that this card falls into, but as a top of the line model, I felt it warranted a little more effort.

The same sentiment can be applied to the construction quality. Again, EVGA stuck to the reference PCB and VRM design. While it's not a bad design, and and the card is well-built, considering it's one of the more expensive GTS 450's on the market and represents EVGA's fastest iteration, I would have like to see some extra attention paid to the overall design and component selection.

Despite the reference design, the EVGA GTS 450 FTW still managed to achieve the same overclock as the non-reference ASUS ENGTS450 TOP. Although voltage could not be adjusted, EVGA's Precision software worked very well and seems to be a bit more refined that ASUS' Smart Doctor. Their OC Scanner software also works well, and is great companion to Precision when overclocking.

The extra software does add value, but there's no getting around the fact that at $147.99 the EVGA GTS 450 FTW is on the upper end of the price scale among GTS 450's. Of course, it's also one of the highest factory overclocked GTS 450's on the market and has an excellent performance per dollar ratio. It's not quite as good a value as the top card from ASUS, but compared to other GTS 450's out there, it's definitely a good buy.

Pros:

+ Significant overclock beyond reference speeds
+ Good construction quality
+ Lower power consumption than comparable ASUS ENGTS450 TOP

Cons:

- No HDMI adapter

Ratings:

• Performance: 9.25
• Appearance: 9.00
• Construction: 9.00
• Functionality: 9.50
• Value: 8.50

Final Score: 9.05 out of 10.

Excellence Achievement: Benchmark Reviews Golden Tachometer Award.

Questions? Comments? Benchmark Reviews really wants your feedback. We invite you to leave your remarks in our Discussion Forum.

Related Articles:

• AMD Phenom-II X4-965 BE 125W CPU HDZ965FBK4DGM
• SandForce SF1500 Enterprise SSD Processor
• Fractal Design Node 304 Mini-ITX Case
• CyberPower CP1500AVRLCD 1500VA 900W LCD UPS
• Silverstone Kublai KL03B-W Mid-Tower ATX Case
• CyberPower CP1350AVRLCD 1350VA 810W UPS
• ASUS My Cinema-U3100Mini HDTV Tuner
• Lian Li PC-90 Computer Case
• Best CPU Cooler Performance LGA775 - Q4 2008
• RatPadz XT and GS Gaming Surface Mouse Pads