Lord Sojar

Yeah just for clarity, someone asked about balancing various under-used weapons and said something about "dual pistol mesmer" as an example, I figured that's what he calls off-hand pistol and answered using the same verbiage asked. I've found travelling abroad, as well as across the fan base everyone uses different terms for part of the game, people still ask me all the time about our quest system for example, it's easier to just respond with the terms they use so they understand.

I'm well aware mesmers can't dual-wield pistols, we've only been working on it for 6 years now If there was double-pistols, my Mesmer would whoop you in PvP with them though, just sayin.

As for some of the other questions, I'm sorry we can't provide the exact details folks want to know on every single question, but on many of them we simply haven't decided yet what we're going to do, or we're currently working on it but may massively change direction due to the outcome of play tests and don't want to get locked into one plan/expectation. If we say literally anything, we're held to it as the law forever, so we have to try and be very careful not to promise anything. As an example, someone asked about commander book and I mentioned we'd love to have it get more functionality, a day later I read on the forums we're in the middle of making commander book better, which we aren't working on yet. It's really tough, sort of a no-win situation but we do our best and will keep delivering content every month that hopefully, eventually makes most folks happy with what they get.

Off to play to my dual-pistol warrior!

Lord Sojar, on 15 November 2012 - 04:10 AM, said:

Actually, if i recall correctly, the V4 has a Marvell controller (think plextor M5P, Crucial M4, Intel 510, Corsair perf pro) with a custom Indilix firmware that just makes it scream. OCZ decided to use a very reliable controller and Indilix did a great job on the firmware.

So, i grabbed some performance info from the stress test. These two systems followed the exact same path (at the same time, so same dynamic events ect.) eastwards through the Charr starter zone, leading into the swamps area in the far east portion of the map (the gtx670 shows a relatively steep decrease in FPS when entering the swamp). I'll provide a more detailed analysis of the different sections, but this is an idea of how the stress test was performing in the Charr starter area (which is more demanding than the others in my experience).

Both systems were running complete max, including supersampling, at 1920x1080.

Min/Max/AVG:

GTX670: 7/107/48.4
HD5870: 6/56/20.7




Right! So now that i'm slightly recovered from a long week of work and two late nights with the stress test as well, i will actually provide some slightly more useful info, and give a little more insight into some of the weirder behaviour (more notably in the gtx670 system, which is a more interesting graph anyway).

First off, the obvious:  The two graphs are not entirely representative of what each component can do. This wasn’t intended to be a ‘graphics card comparison’ graph, or any form of truly apples to apples comparison. What it is meant to show is how a ‘mid range’ and ‘lower high end’ system currently handle the game with every bell and whistle turned up. I can (and will) talk a little bit about what i may think is holding the HD5870 back though (a hint: i don't think it's the CPU).

The mid range system chokes in the Charr starter area. Although there is definitely a CPU bottleneck at work in GW2 with the Phenom II X4 940 in many areas, a lot of the performance in this particular benchmark, due to supersampling, is down to the HD5870. Comparing the two data sets, we can see that the gtx670 is roughly twice as fast as the HD5870 throughout most of the run, this matches pretty well to what you would expect the difference to be in a GPU bound environment. But there are several points where the gtx670 massively extends this lead.

Now there is one particularly interesting spike in the graph, which is the period of time around the 1000 second mark, where the GTX670/i5-2500K system sees a massive framerate increase. This was when entering a small outpost towards the east of the Charr lands (where a skill point is located). I would have expected the HD5870 system to follow this performance increase, but it remained stubbornly at the 20-30fps region. I would be very interested to investigate this anomaly more, but time was limited. My hunch is that the 1GB of vRAM available to the HD5870 simply prevents it from pushing out a supersampled image out at any decent pace. Data from a personal story instance, where the gtx670 ran into the high 80's giggling, whilst the HD5870 sat at the 20-25 region whilst the CPU was semi-idling further pointed to this, a HD5870 should be able to keep 50% pace with a gtx670 if vRAM is not an issue and the CPU allows (which it should have).

There are some spikes in the graph where the HD5870 record a single point FPS of above 30fps, but these were two single data point situations, so i think i'll dismiss those for now. I will investigate vRAM usage next stress test so i can see if there is a pattern (unfortunately, i never really got around to checking out this data until i had stopped playing).

The swamp area (~1050s onwards) of the Charr land is extremely demanding, due to the large number of reflections, and even a gtx670 manages to stoop below 30fps on occasion. The HD5870, meanwhile, only drops a small amount of FPS, which again indicates that it had been operating under a bottleneck, possibly vRAM, for much of the benchmark.

But for now, the simple conclusion is that to enable supersampling and have it run well in all areas, you’ll want a truly high end card (and even then, it’s not going to be perfect) with at least 2GB vRAM.

P.S. the large FPS drops for the gtx670 system in the 200-300s timespan are me being hit by catapults (taking it on the chin like a bawss) and being surrounded by lovely particle effects and smoke.