Thursday, October 7, 2010

The Game Rundown: Finding CPU/GPU Bottlenecks, Part 1


Wondering if your favorite game wants more CPU or GPU muscle? We tested 20 different titles using a GeForce GTX 460 and a Core i5 processor (one of our favorite configs) in order to figure out where a solid mainstream machine gets hung up.

There are several questions we're often asked about CPUs and GPUs that we wanted to answer in this article:

  • Do I need a quad-core CPU or is dual-core enough?
  • What would be the performance impact of mixing a powerful graphics card with a weak CPU?
  • Should I invest my money in a better processor or faster graphics?
  • How much GPU and CPU performance do I need for HD-class gaming?


Of course, we already have a wealth of information on the site covering these questions indirectly. You can start with Paul Henningsen's Building A Balanced Gaming PC series, and read back through Parts 1-4, all of which are linked here.

For this article, we go after each question head-on. We decided to abandon our usual benchmarking style and instead focus solely on actual in-game results, experienced first-hand. Throughout multiple runs, we tried to keep the real-time gaming consistent, but a slight margin of error was unavoidable. Some slight fuzziness in the results was a fair trade-off in order to reach our true goal: analyzing the CPU and GPU under actual gaming conditions to show just how those two components can hold each other back, influence results, and even help each other.

Zoom

To answer the questions above, we modified our BIOS settings in order to test different CPU performance levels--so you're not just going to see the results with a stock Core i5 processor. We settled on a base clock frequency of 3 GHz, as this should be within reach for many of today's overclockable mainstream CPUs. We then forced the Core i5 to run in single-, dual-, and quad-core configurations to test and verify multi-core optimizations in our games. As a best-case scenario, the quad-core Core i5 was overclocked to 4 GHz--a reasonable clock rate for this particular model given heat, power consumption, voltage, and stability considerations.

The CPU was paired with a GeForce GTX 460 (768 MB), which is sufficient for most games at HD resolutions. In a few cases, we used the more powerful AMD Radeon HD 5870 to illustrate differences in graphics performance more clearly. Testing was done at a 1920x1200 resolution. Anti-aliasing and anisotropic filtering were set to 8xAA and 16xAF whenever possible via in-game settings to keep the results independent from AMD- and Nvidia-specific driver options. No special software or executable tweaks were used to force anti-aliasing on, meaning that if we were not able to activate a given setting in the game options, then we simply didn't use it. Our 20 games spanned from DirectX 9 to DirectX 11.

Zoom

Nvidia and AMD Graphics Cards
CPUIntel Core i5-750 @ 4 GHz (21 x 190 MHz), Lynnfield design, 1.26875 V core voltage, 45 nm, LGA 1156
MotherboardGigabyte P55A-UD7, PCIe 2.0, 3-Way SLI
ChipsetIntel P55 Express
MemoryOCZ3G2000LV4GK, PC3-16000 Golden Series, 2 x 2 GB DDR3, 2 x 570 MHz 7-7-7-19 Timings
AudioRealtek ALC889
LAN2 x RTL8111D
HDDsSATA 3Gb/s, Western Digital Raptor WD300HLFS
DVDGigabyte GO-D1600C
Power SupplyCooler Master RS-850-EMBA 850 W
Drivers & Configuration
GraphicsAMD Catalyst 10.7, GeForce 258.96
OSWindows 7 Ultimate 32-Bit
DirectX9, 10, and 11
Chipset drivers
Intel 9.1.1

The test results yield two charts: the first shows the utilization of the hardware (CPU and GPU) in percentages and the second focuses on graphics performance using different core counts.

The results are not sorted, but the order remains the same throughout the article. In the first chart,we always list the Core i5 CPU employing one core first (black bar) followed by the GeForce GTX 460 GPU (green bar). Next is the dual-core CPU + GPU followed by the 3 GHz quad-core CPU + GPU. Finally, we show the overclocked 4 GHz quad-core CPU + GPU. In each case, the number of bars corresponds to the number of CPU cores, and the average CPU or GPU utilization is shown as a numerical percentage value.

One more word about CPU performance: the load may be divided between several cores in a number of these benchmarks, but the only relevant thing to notice is the level of utilization. Just because the load is distributed between all four cores does not mean that the game fully supports multi-core CPUs. Ideally, all cores would then be utilized at close to 100 percent. Load distribution and management eats some CPU power. You'll be able to see quite clearly when all four cores are really fully utilized in a game like Grand Theft Auto 4.

If the quad-core 3 GHz CPU is not fully taxed in a game, you should see lower utilization compared to the overclocked 4 GHz CPU, since the processor is even less of a bottleneck. For example, this can clearly be seen in Alien vs. Predator, a game in which the graphics card is the obvious bottleneck. This allows you to get away with having a slower CPU and not negatively impact performance.

In the second chart, you see the frame rates of our different CPU and GPU combinations as differentiated by the number of active CPU cores. This chart for Alien vs. Predator confirms what we just mentioned. The GPU is such a tight bottleneck that it hardly matters whether you have a single-core or an overclocked quad-core CPU. If you want to increase frame rates in this game, you'll have to upgrade the graphics card. The game uses 750 MB of graphics memory, which is almost as much as the GeForce GTX 460 makes available.

According to our tests, a dual-core CPU is sufficient for Avatar, while a single-core CPU wouldn't be able to keep the graphics card fully utilized.

Despite essentially the same load being distributed over more cores when using a quad-core CPU, the frame rates increase significantly. This is not the case when overclocking, which shows little benefit.

We can't really determine where the performance difference between two and four CPU cores is coming from within our data. With AA enabled, the graphics memory is almost fully used at 750 MB. The CPU utilization goes down when overclocking, but there is no frame rate increase.

To get higher frame rates, you'll want to upgrade your graphics card and use at least a quad-core CPU, despite just two of the cores being fully used.

As with Avatar, the Divinity 2 numbers are inconsistent. A single-core CPU will create a huge bottleneck, but at the same time, it shows no more than 43% of CPU utilization. When using a dual-core CPU, some kind of valve is turned on and the game runs much better. Looking at the utilization numbers , we see that two cores are almost overkill. Four cores hardly increase frame rates at all.

The game uses 690 MB of graphics memory with AA activated.

To get the most out of this game, you need a dual-core CPU and a solid graphics card.

Dragon Age scales really well with CPU performance. A single-core CPU will be fully utilized to the extent of holding back the graphics card. A dual-core CPU will be fully utilized as well. But despite this fact, you'll only see slightly better frame rates. With four cores, the load is more evenly spread out, and the graphics card can stretch its legs. Yet there's still a small performance increase when overclocking the CPU. The game seems to be CPU-limited, as indicated by the 90% graphics card utilization. Most likely, it would need even more CPU performance to increase that utilization number to around 100%.

GTA 4 reacts so strongly to increases in graphics memory, CPU performance, and GPU performance that it's easy to constantly wonder if you're fully utilizing your expensive hardware or not. The game’s need for CPU speed is so strong that not even four cores can really satisfy it. Frame rates with a single-core CPU are so ridiculous that the game is practically unplayable. Even with two cores, GPU utilization is just 69 percent. Fewer than four cores are unthinkable for this game in our opinion.

Overclocking the CPU shows that the game just can't get enough CPU performance. To increase frame rates in GTA 4, you need at least four CPU cores, and when overclocking those cores to the max, you can start thinking about upgrading to a better graphics card. In the video settings, you can incrementally set the amount of graphics memory used by the game. This way, 768 MB can actually be enough, but as far as we could make out, only 630 MB are actually used at that setting. GTA 4 can take advantage of more than 1 GB of graphics memory, so if you like the game and play it a lot, you should definitely consider a graphics card with lots of onboard memory.


No comments:

Post a Comment

If you have any Doubt..kindly let me know