While our test hardware has not changed since Which Web Browser Is Best Under Windows 8?, our benchmark suite underwent substantial changes. First, the startup times are now taken using a stopwatch, and the test pages are hosted from our local Web server. Using a stopwatch includes the time it takes for the actual applications (Web browsers) to open, as opposed to just timing how long it takes the browsers to load their tabs. By hosting the test pages from the local Web server, we cut out the instability and variation that comes with testing live pages.
As you may gave guessed, since we ditched our startup timers, we also ditched the page load timers that they're based on. We now use EEMBC's BrowsingBench to gauge page load times. This new test addresses several of the shortcomings that plagued our old page load timers; for instance, it tests both desktop and mobile pages, multiple pages on the same site, foreign sites, and it runs multiple iterations per test run.
Moving on to JavaScript, we finally retired SunSpider from the test suite. This benchmark hasn't been updated in several years, and there are far better JavaScript benchmarks available today (such as Kraken, which we're also retiring in favor of an even better test: Rightware Browsermark). Octane will not appear in the WBGP because the V8 portion still skews the results far in Chrome's favor. RIABench and Peacekeeper remain in the JS portion of the test suite. In order to minimize the effect of Dromaeo DOM in the final scoring, that test is now counted equally with the three JS tests in the final JavaScript/DOM composite score.
The HTML5 section also received several cuts, including GUIMark 2, Asteroids, and Smashcat. Only Impact remains. Joining it is the first consumer preview of Principled Technologies' WebXPRT, an HTML5 applications benchmark that frames the performance testing around common office Web apps, such as an image editor, stock tracker, and notes.
The final massive change to the benchmark lineup is in WebGL, both WebGL Solar System and Mozilla's WebGL FishIE Tank were replaced. Airtight Interactive's WebGL Demo and Scirra's WebGL Performance Test now make up that portion of the test suite.
Test Setup And Benchmark Suite
| Test System Specs | |
|---|---|
| Operating System 1 | Microsoft Windows 7 Ultimate SP1 (64-bit) |
| Operating System 2 | Microsoft Windows 8 (64-bit) |
| Processor | Intel Core i5-2500K @ 3.3 GHz (quad-core) |
| Motherboard | Gigabyte GA-Z68XP-UD3 (F10 BIOS) |
| Memory | 8 GB Crucial DDR3 @ 1333 MT/s (2 x 4 GB) |
| Graphics | Asus GeForce GTX 560 Ti 1 GB GDDR5 (PCIe 2.0 x16) |
| Storage | Seagate Barracuda 7200.12 500 GB SATA 3Gb/s, 7200 RPM, 16 MB Cache |
| Optical | Asus DRW-24B1ST/BLK/B/AS |
| Power Supply | Corsair TX750W (750 W max) |
| Case | Zalman MS-1000 HS2 |
| CPU Cooler | Scythe Mugen 2 Revision B |
| Monitor | AOC E2752Vh 27-inch LED (1920x1080) |
| Keyboard | Logitech Wireless Keyboard K320 |
| Mouse | Logitech Wireless Trackball M570 |
| Local Web Server Specs | |
| Operating System | Ubuntu 12.04.1 LTS Server Edition "Precise Pangolin" (32-bit) |
| Processor | Intel Pentium 4 @ 2.41 GHz |
| Motherboard | Biostar P4M80-M4 |
| Memory | 768 MB DDR @ 333 MT/s |
| Storage | Western Digital Caviar SE WD1600AAJD, 160 GB EIDE, 7200 RPM |
| Extra Packages | Apache2, MySQL Client, MySQL Server, PHP5, PHP-GD, PHP5-MySQL, PHPMyAdmin, SSH, Node.js, NPM |
| Network Specs | |
| ISP Service | Cox Preferred (18 Mb/s down, 2 Mb/s up) |
| Modem | Arris Touchstone Telephony Modem TM502G |
| Router | Linksys WRT54G2 V1 |
| Benchmark Suite | |
| Startup Time | Cold Start Time (Google SERP, Cached) |
| Hot Start Time (Google SERP, Cached) | |
| Cold Start Time (Eight Tabs, Cached) | |
| Hot Start Time (Eight Tabs, Cached) | |
| Page Load Time | EEMBC BrowsingBench |
| JavaScript | RIABench JavaScript (Eight Tests) |
| Futuremark Peacekeeper v2.0 | |
| Rightware Browsermark v2.0 | |
| DOM | Mozilla Dromaeo DOM (Core) |
| HTML5 | Principled Technologies WebXPRT CP1 |
| Impact HTML5 Benchmark | |
| Hardware Acceleration | Facebook JSGameBench v0.4.1 |
| HTML HWA | WebVizBench |
| Psychedelic Browsing | |
| WebGL | Airtight Interactive WebGL Demo |
| Scirra WebGL Performance Test | |
| Memory Efficiency | Memory Usage (Single Tab) |
| Memory Usage (40 Tabs) | |
| Memory Management (-39 Tabs) | |
| Memory Management (-39 Tabs, Two Additional Minutes) | |
| Reliability | Proper Page Loads |
| Security | Browserscope Security |
| Standards Conformance | HTML5Test.com |
| The CSS3 Test | |
| Ecmascript Language test262 | |
While applicable links are included in the table above, we also have a public delicious account dedicated to Web Browser Grand Prix benchmark links.
Detailed methodologies are explained on the individual benchmark pages.
- Possibly The Last "Top Four"
- Chrome, Firefox, Internet Explorer, Opera
- Test Setup And Benchmark Suite
- Startup Time
- Page Load Time
- JavaScript And DOM Performance
- HTML5 Performance
- Hardware Acceleration Performance
- Memory Efficiency
- Page Load Reliability And Security
- Standards Conformance
- Windows WBGP Winner's Circle


i personally am still on ff10, memory is really the reason i use fire fox, along with chrome. i would love to see how older versions stack up to newer ones.
2. " Opera jumps on-board the Chromium bandwagon" is false. Opera is using Webkit rendering engine for rendering. We dont know what the Javascript engine and JIT compiler is going to be.
3. On page4, in "the average wait time" , you are taking the Geometric Mean. I think that using values less than 1 in a geometric mean is skewing the results. Can you take a Arithmetic Mean, and check please ?
4. Memory usage reporting of IE9 and IE10 is completely bogus. For some workloads, in TaskManager, IE10 is seen using about 300MB memory, while it is actually using around 3GB RAM !
5. I am completely mystified why quite a few browsers do better in JS benchmarks on Windows8 . I can understand getting better scores in WebGL or HW acceleration tests (because of potential better Dx) , but JavaScript execution does not use Dx. Any ideas ?
i personally am still on ff10, memory is really the reason i use fire fox, along with chrome. i would love to see how older versions stack up to newer ones.
1. Yup
2. The press release mentioned "Chromium", so I'm assuming WebKit/V8.
3. Yes, you're right! The last timers went by milliseconds, so that wasn't an issue - the replacement charts should appear soon. Good catch!
4. It seems pretty reasonable to me, basically mirrors Chrome in this regard.
5. Nope
#2, incorrect, they have said they're going with Chromium's V8.
@alidan
Mozilla saw the err of their ways and got after memory. The most recent version of Firefox should beat version 10 in both memory and performance. The last benchmarks I saw had it beating all the other browsers in memory usage as well. (In this article you can see a snapshot of this in the "40 tabs" graphic. Like many techies, this is more what my browsing looks like.)
#2 : I must have forgotten the release. Thanks for the correction.
#4 : I am not saying that IE10 uses excessive memory. I am saying that the total memory used by all the ieexplorer processes in the task manager is incorrect. For some image heavy pages, total memory usage of IE as reported by task manager is about 300MB, but total system memory usage gets around 3GB!. So if you close IE10, system memory goes from 90% full to 50% full.
The composite hardware acceleration scores is most likely the main reason why IE9/10 is so far behind Firefox and Chrome on performane. Yet, from what I can tell, this composite score is heavily influenced by the WebGL scores, which is exclusive to Chrome and Firefox.
In that respect, MS has at some point stated that they do not even wish to support WebGL, as it represents a significant security risk, as it gives the browser close access to the computer hardware.
Long story short, your methods of calculating performance scores heavily favors Chrome and Firefox as they are the only ones to implement support for WebGL.
Additionally, I wish you would make it more clear how you arrive at your composite scores and of course the final Performance Index. How do you add numbers that are so varied in nature, without some method of normalizing the numbers?
I heard that Futuremark Peacekeeper is unreliable, it used to miscalculate it's own benchmarks and it's still a black box so it may still be buggy. Do you know anything specific about it?
There is a reason why he uses the default install settings, and he explained this before in the former WBGP. The number of enthusiasts are far smaller than the 1-bit users, thus the default. Also, most enthusiasts actually know what is to be expected from the mods/tweaks, but normal users hardly know anything about browsers (if they even know the term: internet browser).
Because if it switches to webkit, all of it's test results will be identical to Chrome's, or it will hardly differ.
@alidan: FF17 ESR would be more safe than FF10, and even better with memory.
Edit: I hate how it keeps deleting the links.
Same for my Android phone.