View Profile
View Forum Posts
View Articles
I finally managed to sit down and get a little time to benchmark the actual client rather than relying on the official FFXIV benchmark for performance numbers. The numbers for the beta client were obtained by running the same 12 minute course, starting at the Adventurer's Guild in Limsa Lominsa, running down through and past Bearded Rock, and then doubling back. As I obviously can't trace my steps exactly and cannot control other variables such as the number of players along the path at any given time, there's a small margin of error involved. CPU usage was tracked at the logical processor level via the native performance monitor built into Windows 7. GPU usage and framerate were obtained via MSI Afterburner. For all tests, the Sapphire 5870 was set to 850MHz core and 1200MHz memory to approximate a reference 5870 as closely as possible.
The test system specs are as follows:
| CPU | Intel i7 860 (2.8GHz stock) |
| Motherboard | Gigabyte GA-P55A-UD7 |
| RAM | Gskill DDR3-1600 7-8-7-24 |
| Video | Sapphire 5870 2GB TOXIC Edition (925MHz/1225MHz stock but set to 850/1200 for the benchmark) |
| Sound | Creative X-Fi Fatality Pro |
| Storage | SandForce 100GB SSD (OWC Mercury Extreme) |
| OS | Windows 7 Professional 64-bit at stock performance settings |
| Video driver | ATI Catalyst 10.7 |
The following image quality settings were used for all tests:
| Display Mode | Full Screen |
| Window Size | 1920x1080 |
| Multisampling | 8xQ MSAA |
| Buffer Size | Resolution |
| Shadow Detail | Highest |
| Ambient Occlusion | Off |
| Depth of Field | Off |
| Texture Quality | High |
| Texture Filtering | Highest |
Test 1 - The Baseline
i7 860 @ 2.8GHz with 2 cores enabled, Hyperthreading disabled, Turbo off
(2 logical processors consisting of 2 physical cores)
Official benchmark 1920x1080 score: 3646
Beta client 1920x1080:
| Average GPU usage | 47.4% |
| Average Framerate | 41.7 |
| Min Framerate | 15.3 |
| Max Framerate | 60.2 |
| CPU1 Avg usage | 83.1% |
| CPU2 Avg usage | 78.7% |
| Total CPU Avg | 80.9% |
Observations: Looking at the graph, you see a very nice symmetrical pattern which makes sense since I ran the path in one direction and then doubled back. Note the dips in framerate where neither CPU nor GPU are pegged and framerate and GPU mimic each other for a great deal of the benchmark.
Test 2
i7 860 @ 4.0GHz with 2 cores enabled, Hyperthreading disabled, Turbo off
(2 logical processors consisting of 2 physical cores)
Official benchmark 1920x1080 score: 4505
Beta client 1920x1080:
| Average GPU usage | 57.2% |
| Average Framerate | 49.3 |
| Min Framerate | 28.6 |
| Max Framerate | 60.2 |
| CPU1 Avg usage | 86.3% |
| CPU2 Avg usage | 68.1% |
| Total CPU Avg | 77.2% |
Observations: Increasing CPU from 2.8GHz to 4.0GHz definitely smoothed the game out and brought the minimum framerate up a lot. Again, we have the nice symmetrical pattern thanks to the path that was followed.
Test 3
i7 860 @ 2.8GHz with 2 cores enabled, Hyperthreading enabled, Turbo off
(4 logical processors consisting of 2 physical cores and 2 virtual cores)
Official benchmark 1920x1080 score: 3739
Beta client 1920x1080:
| Average GPU usage | 47.4% |
| Average Framerate | 41.6 |
| Min Framerate | 17.5 |
| Max Framerate | 60.2 |
| CPU1 Avg usage | 62.3% |
| CPU2 Avg usage | 49.3% |
| CPU3 Avg usage | 25.7% |
| CPU4 Avg usage | 47.7% |
| Total CPU Avg | 46.2% |
Observations: While the system thinks overall system CPU usage dropped, there was little in the way of impact to framerate.
Test 4
i7 860 @ 2.8GHz with 4 cores enabled, Hyperthreading disabled, Turbo off
(4 logical processors consisting of 4 physical cores)
Official benchmark 1920x1080 score: 4613
Beta client 1920x1080:
| Average GPU usage | 49.6% |
| Average Framerate | 41.8 |
| Min Framerate | 17.9 |
| Max Framerate | 60.2 |
| CPU1 Avg usage | 47.7% |
| CPU2 Avg usage | 57.3% |
| CPU3 Avg usage | 40.9% |
| CPU4 Avg usage | 43.2% |
| Total CPU Avg | 47.3% |
Observations: Not much change in the way of framerate off the baseline although the official XIV benchmark would have you believe there's almost a 1000 point difference.
Test 5
i7 860 @ 4.0GHz with 4 cores enabled, Hyperthreading disabled, Turbo off
(4 logical processors consisting of 4 physical cores)
Official benchmark 1920x1080 score: 4782
Beta client 1920x1080:
| Average GPU usage | 63.6% |
| Average Framerate | 54.3 |
| Min Framerate | 33.5 |
| Max Framerate | 60.2 |
| CPU1 Avg usage | 20.4% |
| CPU2 Avg usage | 68.6% |
| CPU3 Avg usage | 51.8% |
| CPU4 Avg usage | 30.8% |
| Total CPU Avg | 42.9% |
Observations: This demonstrates that even at 1920x1080, FFXIV thrives on a faster CPU. FFXIV makes heavy use of two cores and lighter use of a third and fourth core. Framerate spends a great deal more time capped at 60fps when 4 cores are active.
I'm only making an educated guess here based upon what I see in the performance numbers and in the client but I suspect that the code is not currently able to take full advantage of either CPU or GPU, probably due to how it's been written - think lots of conditional debugging stanzas that will likely remain until sometime after launch when they're satisfied that the client has reached some broad level of stability.
There are lots of places that you can go in Limsa Lominsa where neither GPU nor CPU are anywhere close to being maxed out and yet the framerate still drops to 20 and it sits at 20 regardless of what you are looking at - you could be nose-to-wall and still have the same low framerate. If I had to guess, I would say this has to do with some combination of thread synchronization, the code waiting on network data, and performance-limiting debugging code. These issues are likely to get addressed in the days to come. There's definitely a lot of room for performance improvements in the client such that it should perform very well even on a moderate build.
Special thanks to Xariana for setting up the images for this article.
Reply
