|New video card launches from AMD and NVIDIA are almost always reviewed on hardware less than 12 months old. That's not an arbitrary decision -- it helps reviewers make certain that GPU performance isn't held back by older CPUs and can be particularly important when evaluating the impact of new interfaces or bus designs.
The downside of this policy is that it leaves a gap in product coverage. Gamers with older systems often miss out on whether or not a new graphics card will be a meaningful upgrade for aging systems. That's particularly important as the speed of the desktop replacement cycle has slowed.
Here, we're going to compare the performance impact of upgrading the graphics card on an older system that doesn't have access to any of the substantial performance gains Intel introduced with Nehalem in late 2008. No integrated memory controller, 1600MHz DDR3, Hyper-Threading, or QPI.
Our upgrade card of choice is an EVGA GeForce GTX 660 SuperClocked card with 2GB of RAM, a 192-bit memory bus, and a core clock of 1046MHz / 1111MHz boost.
EVGA's GeForce GTX 660 SuperClocked - Our test vehicle
This card currently retails for $229 at NewEgg ($219 with rebate), which is in the so-called "sweet spot" of the GPU market. We figured the EVGA GeForce GTX 660 SuperClocked was exactly the type of mid-level graphics card someone trying to breathe new life into an aging desktop PC would consider.
So how much of an improvement could a buyer expect? Let's find out...
|Our Testbed, Game Detail Settings|
|Some of you may not remember the particulars of hardware from this far back, so we'll describe the testbed with a bit more detail than is typical. Our test system for this piece consisted of an Asus Rampage Formula motherboard, 3GB of DDR2-1066 RAM, Intel's Core 2 Q6600, and an NVIDIA GeForce GTX 260. Windows 7 64-bit w/ SP1 was installed on a 60GB Corsair Force 3 SSD.
The oldest component is the CPU. The Q6600 (65nm, 2.4GHz) was Intel's first consumer quad-core and debuted nearly six years ago, in early 2007. Its large L2 cache, hardware virtualization support, and excellent overclocking kept it popular even after 45nm Core 2 Quad CPUs based on Penryn became available.
The Asus Rampage Formula is a 2008 motherboard based on Intel's X48 chipset. The X48 chipset was the no-holds-barred enthusiast option for Core 2 Duo. The Geforce GTX 260 video card also launched in 2008; our GPU is an original GeForce GTX 260 on 65nm with 192 CUDA cores, not the "Core 216" variant that NVIDIA introduced later that same year.
The two components of our testbed that break with the 2006-2008 timeframe are the operating system (Windows 7 64-bit w/ SP1) and the 60GB SSD. Windows Vista has been abandoned by virtually everyone, which made it a poor choice for any modern comparison, and none of our tests touch the SSD.
Why an SSD, you ask? Because trying to load Shogun 2 off a hard drive is like trying to eat soup with a fork.
Who These Results Apply To:
These results should map directly to any gamer with a Core 2 Quad and an X38/X48 motherboard and at least 3GB of RAM. Gamers with P965, P35, or 975X boards should see very similar results, but these boards topped out at PCI-Express 1.1. At the time, there was no performance difference between PCIe 2.0 and PCIe 1.1; this may have changed since 2008.
These results don't necessarily apply if you've got an older Core 2 Duo. In 2008, precious few games used more than two threads. Quad-threaded titles are more prevalent than they used to be. It's possible that we'll revisit this question if enough of you are curious about the outcomes.
We tested Battlefield 3, Borderlands 2, Civilization V, and Shogun 2. For the first three games, we tested the same graphics settings across both cards to compare the performance gain. With Shogun 2, we took screenshots of the GTX 260 vs the GTX 660 as well as the GTX 660 at a custom detail level above the "Very High" preset.
Texture filtering was set to "High Quality" in all games, on both cards.
|Battlefield 3, Borderlands 2|
BF3 is known for maintaining a consistent frame rate and runs extremely well on older hardware. We benchmarked the game with custom detail settings -- everything was set to "High", save for MSAA, which was enabled at 4x. We benchmarked the Operation Firestorm map in a 64 player game.
Battlefield 3 picks up a solid 44% frame rate boost from our upgrade. The game is playable at 27fps but stutters when the action picks up. The GeForce GTX 660's 39 fps is noticeably smoother, particularly during high action fire fights.
Borderlands 2 doesn't have presets, so we've dropped in a handy screenshot with the settings we used.
In Borderlands 2, we forced FXAA on via the graphics driver as well as enabling it from in the in-game menu. This somewhat improved the visual quality but didn't impact performance much on either card.
Borderlands 2 played pretty well on the GeForce GTX 260, partly because it's a DX9 title. Slowdowns, however, were still quite noticeable when multiple enemies were on screen at the same time. Upgrading to the GTX 660 solved this problem and improved frame rates by 1.5x.
Civilization V was benchmarked with all settings on high with 4x MSAA enabled. We used the game's built-in Late Game View benchmark. The "Use the GPU for texture decompression" option was checked, but grayed out when testing on the GeForce GTX 260.
Civilization V doesn't run very well on the GeForce GTX 260, at least not in DX10/DX11 mode. The GTX 660 is nearly 3x faster. Part of the reason why may have to do with the option to use the GPU to decompress graphics textures. On the GTX 260, this box is checked and enabled by default, but the performance delta separating the GTX 260 and the GTX 660 in Civilization V's texture compression benchmark is significant to say the least.
According to the game's documentation, the texture compression test is "used to measure the performance of Compute Shading features via our variable bitrate compression technology."
Well, it's safe to say the GeForce GTX 660 is a little better than the GT200-era GPU. The performance jump between the two cards in Civilization V is something of an outlier for a game that's often been considered CPU-bound -- but upgrading the GPU clearly makes a major difference in DX11 mode.
|Shogun 2, Visual Upgrades|
|Shogun 2 is an immersive RTS game with gorgeous visuals, full support for DX11, and a wickedly long load time if you don't drop it on a solid state drive. We chose to "Benchmark Game Settings" using the "High" preset on both the GeForce GTX 260 and GTX 660. In this case, we also opted to test the GTX 660 using "Very High" details to include and additional data point and drive home the point that upgrading a graphics card may not only yield additional performance, but allow for better visuals as well.
Upgrading doubles the overall frame rate. Opting for Very High detail and enabling tessellation and soft shadows cuts the final frame rate somewhat, to 54 fps, but further improves the game's visuals.
In most of the games we tested, the quality difference between DX11 and DX10 wasn't noticeable. Shogun 2 is different. First, here's the GeForce GTX 260 against the GTX 660, both at High detail. This isn't the same frame, but it's as close as we were able to match, as shown by the movement of the leaf.
Look at the trees in the background. Notice how they're slightly blurred and shadowed in DX10 mode? This is one place where just shifting to DX11 makes a significant difference for these NVIDIA powered cards, even if no other detail settings are changed.
Here's a screenshot from the same sequence, at Very High Detail. Again, we matched the frame as closely as possible.
It's a nice argument for upgrading given that you get to keep most of the FPS increase, while also increasing graphics quality and improving the level of immersion possible with the game.
|Power Consumption, Conclusion|
|Just for fun, we also decided to compare idle and load power consumption between the two cards. Power consumption was measured in Shogun 2, using High Detail settings.
The 30W difference between the two cards is impressive considering that we've only swapped out a single component. What's really striking, however, is the power efficiency difference as measured in frames per watt. The GTX 260-equipped system draws 250W to output a 30 FPS frame rate; the EVGA GTX 660 system nearly doubles the FPS rate to 63, while drawing 12% less power. It also packs more than 2x the RAM and a roughly twice the number of transistors.
The Bottom Line:
The GeForce GTX 260 held up surprisingly well in every game but Civilization V. Borderlands 2 and Battlefield 3 were at least playable, if not exactly smooth. Upgrading the graphics card alone vastly improved the performance of every single title. The smallest gains were in Borderlands 2 and Battlefield 3, where the GTX 660 was ~1.5x faster than the GTX 260. In Shogun 2, the GeForce GTX 660 upgrade doubled the GTX 260's frame rate at the same quality level, but also allowed for higher image quality settings. Civilization V wasn't playable on the GTX 260 with the settings we used; performance improved by nearly 3x when we switched to a modern GPU.
If you've got an older system that loosely fits our specifications, is it worth upgrading? I think so. A new CPU, motherboard, and RAM will run substantially more than the $220 starting price of a GeForce GTX 660 like the EVGA card we've featured here. Moving from the GTX 260 to GTX 660 didn't just improve frame rates -- it gave us sufficient headroom to increase graphics fidelity without losing much in the way of performance.