Widescreen Gaming Forum

Introduction

It's a wide, wide processing world out there. A whole host of processing options are now available on the market. They run the gamut of extremes between high efficiency and high power. More and more are coming with built-in graphics, and those graphic options are now actually usable.

Modern processors contain from 2-8 processing cores, with speeds ranging from 3.1GHz to 4.2GHz. This wide range of options creates a lot of price posts, with mainstream prices ranging from $120 to $350.

This mix of cores and speed produces a wide range of performance ratings in CPU-intensive tasks, but how does that translate to gaming which is more GPU dependent? Is a high-end processor even needed for gaming?

Should you save money on a CPU purchase, and instead put that towards a better (or possibly 2nd) GPU? Or should that money go to better equipment such as an SSD, higher quality input devices, or additional monitors for Surround/Eyefinity?

These are the questions we hope to answer here.

CPU Power

To compare the processing capabilities of each chip, I tested them with GeekBench 3, and CineBench 11. We tested each in 32-bit and 64-bit modes. The data presented here shows the results for both single core and multi core performance.

Looking at the single core GeekBench results, you can see performance improvements between generations and/or series. For example, there is a noticeable improvement between the AMD FX-8150 and the FX-8350. However, the results between the FX-4350, FX-6350 and FX-8350 are similar.

We see similar single core results with the AMD APUs. There is a measurable difference between the A10-5800K and the A10-6800K, while the performance between the A10-6700 and the A10-6800K are similar.

Conversely there is significant single core performance increase between the Core i3, Core i5 and Core i7.

Comparing the single core and multi-core numbers allows us to understand how efficiently the chips multiply their performance across the whole die. The main comparison is to look at the multi-core performance between the chips, which allows us to compare overall processing performance.

Finally we look at the performance versus price. Bigger chips give us better performance, but are they worth the price premium? The APUs show a slow stead performance improvement, with little change in price.

The AMD FX chips show a rapid performance improvement, with relatively small price increases. The FX-6350 provides ~50% performance improvement over the A10-5800K and A10-6700, with only a $20 increase in price. The FX-8350 almost doubles the performance of these chips, with a ~50% increase in price.

The Core i3-2130 prices right along with the AMD APUs, but the performance is much lower. The Core i5-2380p comes in at just under the FX-6350, but prices at 50% higher. The Core i7-2600k performs just under the FX-8350, but costs 70% more.

From a price/performance perspective, the AMD FX chips provide the best value. But the question still remains on how CPU performance actually translated into improved gaming performance.

Benchmarking

This project was weeks in the testing. I tested each of the 10 processors, across three games, each of those three GPUs, in both 1080p HD and Eyefinity 3x 1080p HD. This results in a ton of tests - 180 to be exact (10 x 3 x 3 x 2).

Each CPU was tested at stock clock speeds, with 8GB of RAM in Dual-Channel configuration. The GPUs were also run at stock clock speed.

PLEASE NOTE: When I started this testing, I tested on the Radeon HD 7790, Radeon HD 7870 GHz and Radeon HD 7970 GHz. In my final days of testing, AMD re-branded and relaunched these cards as the Radeon R7 260X, R9 270X and the R9 280X. This article we published three weeks ago confirms that the new cards perform identically to old cards.

Based on this finding, we are using the new branding and pricing for the article. This allows us to provide information to users on current products, and allows us to use current market pricing in our analysis.

Metro Last Light

For Metro Last Light I used the be built-in benchmarking tool. The test was run at "Very High" quality, with AF 16X texture filtering, Low motion blur and Normal tessellation.

The standard settings on the benchmark run for three loops. I used FRAPS to capture the performance data of the third and last loop.



Looking at 1080p HD widescreen performance, the different CPUs provide absolutely no performance improvement for the R7 260X. There is less than 1fps difference between the lowest and highest processors. This is well within the margin of error.

For the R9 270X, the AMD FX processors provide a frame or two improvement over the AMD APUs, but performance levels off there through the Core processors.

There is marked improvement through the performance of the R9 280X. The AMD FX processors provide ~10fps improvement over the AMD APUs, which is a roundly 25% performance increase. For the FX-4350 and FX-6350, there is no price increase required. The FX-8350 gives an extra frame or two, but comes at a 50% higher price tag.

The Core i5-2380p performs similar to the FX-8350, at a similar price point. However, the Core i7-2600k offers no performance in crease, but costs significantly more than the other chips.



For Eyefinity, all chips perform within 1fps of each other, for each respective GPU. There is no value provided in purchasing significantly higher priced chips.

Excellent review mate, it is somewhat surprising to see the results of this all, i too would have thought that investing in a great all mighty powerful CPU would have boosted FPS leaps and bounds, seems those days are gone and the focus is mainly on GPU.

its interesting to see that the APU's would equal some of the more powerful Intel CPU's too when it comes to gaming, ive been asked to help a friend build a rig soon and these results will certainly be taken into consideration! Typically I just invested £500 on a new Intel i7-4820k+motherboard recently, seems totally overkill now, tho im sure it will give me years of life, or so I hope!

Thanks!

_________________
| Gigabyte X79S-UP5-WiFi Skt2011 | i7 4820k @4.4GHz Watercooled (Ivy Bridge-E) | Corsair Vengence 32GB @2133MHz 11-11-11-27 2T Quad-Channel (8x4GB) | AMD R9-Fury-X 4GB @1140/505MHz + EKWB Watercooled | 256GB Crucial MX200 SSD | 8x 1TB Samsung F3 SATA-II | Corsair 850W TX PSU | 3x Dell S2209W 22"wide TFT's (5760x1080res) | Logitech Z-5400 5.1ch speakers | Win7 Ultimate x64 |

Typos on page 2:
" There is little to know benefit"
"Only when paired with a high end GPU like the Radeon R9 280X do bigger more expensive GPUs make a difference."

Congrats on a very amibitious article, but let's not jump to hasty conclusions...

GPU-limited results do not necessarily mean "going for the bigger CPU does not make sense" ! To be blunt, Ultra graphics and unplayable framerates simply cannot give much insight into CPU performance.

IRL no one is satisfied with 25 fps in Eyefinity; you'd go for lighter graphical settings, thus revealing potential differences between CPUs and chipsets.
The same goes for single widescreen and you can see that when the GPU is big enough to "catch a breath", CPU variability does exist... In my view, the article should start here - looking for the differences, not making them vanish under improbable settings...

Edit: btw, physical cores ≠ maximum threads. The i7 2600K is a 4-core CPU.

scavvenjahh raises an excellent point. In a lot of cases, Metro in particular, the system is obviously GPU limited. Of course a more powerful CPU wouldn't help in that case. Playable settings, while horribly annoying from a benchmarking standpoint, would be a lot more useful as a realistic metric.

Also, this article misses the huge difference in minimum framerate in Metro, particularly with a powerful GPU. By focusing on the average, that may only go up 1-2 fps, it misses that the minimum frame rate may go up by 5-15, or 25%-75%. If a game runs at 20fps, even briefly, it will feel slugging and frustrating to play during those demanding segments. Going from 20 fps to 25, 30, 35 yields a tangible difference, and greatly improves the perceived gameplay. Spending more on a CPU that raises that minimum to 30+ will assure smoother gameplay during the most demanding segments, and thus would be well worth it.

As such, looking at the minimum as well as average frame rates, I wouldn't go any lower than an AMD FX-6350 or a 4-core i5 on the Intel side. And if you're pairing it with a powerful GPU such as the AMD R9 280X, then stepping up to the more expensive AMD FX-8350 or Intel i5 3570K / 4670K is entirely justified to ensure smoother gaming throughout. Otherwise, your $25 savings on the CPU could inflict a 17% performance drop on your $300+ GPU during the most demanding parts - i.e. the parts where you'll need that performance most.

Thanks for finding those. I've fixed them.


Not all instances were GPU limited. In the widescreen results, we see that the CPU does have impact in some instances. And, the framerates in many of the single screen tests were playable.


I agree on the 25fps in EF. I'm doing some testing in CFX now to help get rid of the CPU limits. My goal with the article was to shed light not on the minor differences or nuances between the chips or chipsets. Personally I don't worry about such details. I guess I'm not as much of a hardware enthusiast as I once was. My goal was two-fold. One was to show that high-end CPUs don't given real-world gaming benefits to correlate with their price tags. I fell victim for two upgrade cycles thinking I needed a big Core i7 CPU. The other was to try and show the price/performance matrix between CPU and GPU combinations.


Bad copy-paste on my part. Fixed.


I agree that Metro is really stressful on a system. That's one reason I included DiRT Showdown, as it isn't as demanding.


I agree that minimum framerates are important. That is why I included them in the tables. Adding them to the graph would have made it too confusing. It may have worked had I done one graph for each GPU (three lines on each graph - mix, max and avg).

The minimum data probably should have gotten more of a spotlight. In general the minimums on the EF side had little variability. There were some improvements in minimums in the widescreen testing. The APUs tended to have significantly lower minimums, especially when paired with an R9 280X. However, the minimums with the two lower GPUs still tended to be steady across each processor. For Metro and DiRT, the mid range FX-6350 had similar minimums to the higher priced Intel CPUs. Unigine Valley was the one instance where the CPU really showed marked minimum improvements was the Intel CPUs matched with the R9 280X.


I agree that the FX-6350 or i5 are real minimums for gaming performance. And, I agree that with the higher GPU, you can see the benefit of the higher end CPU. Really, this is the kind of conversation I was hoping to get out of the article. And I was hoping to help people put their dollars to best use. Hopefully everyone feels we've done both.