i’ve ran around in the beta a bit, just checking out the zones and trying classes. don’t wanna try the content too much because i want it to be new to me when it comes out.
but damn some of the zones are pretty rough for a gtx 1080 at 1440p 4x msaa. especially ardenweald is very demanding, seeing anywhere from 48 to well over 100 depending on where i am in it, gpu load always pegged at 99%. dropping down to cmaa helps quite a bit at least.
if only there was some graphics card out there that could help with this. maybe something that starts with a 3?
i tried playing at 720p, it was absolutely beautiful of course. seems a stock 9900k can do about 110 fps on average in ardenweald with 10 except for view distance and environment on 7.
Yep, on the path it’s normal while on the sides the FPS was bit wobbly.
GTX 380, R9 390X… there is a couple. 3080 is a good option, although before it gets into stock AMD will announce and likely release their cards as well, where 16GB alike-3080 card will be available at competitive pricing.
Hm… this can be a bug just like Rustbolt had it few times. If you see such FPS drop try to determine what is causing it by rotating the camera around to somewhat pin-point the view - do a screenshot and post it on the forums. For massive ones also a bug report I guess. If others can reproduce it then it will be even better - and running with Nvidia Nsight or Radeon GPU profiler to profile that to see if it’s just “complex” or if something goes YOLO on the rendering like Rustbolt did.
i want to believe, but been disappointed by amd so many times in a row now. give me another hawaii.
ardenweald just seems really demanding in general. seem to be getting 130-140 avg in maldraxxus and like 70-80 avg in ardenweald, with like 120 in revendreth and 90-100 in bastion. makes some sense, visually there’s way more going on in ardenweald than the rest, especially maldraxxus is just a barren wasteland.
i can’t find any below 60 fps areas now while running around, weirdly. i wonder if it’s related to the night time lighting (which doesn’t seem related to time of day?), since it’s the day time now.
they always are. fury, vega, vii. it seems like they should have a good chance this time around with the 8nm samsung node and how ampere poorly ampere performs at 1440p relative to 4k (i play at both so they’re both relevant to me), but i feel like getting my hopes up will just result in another disappointment. especially when it’s not being shown for another month, and probably releasing in 6 or 8 weeks. that’s assuming you can even buy one when they come out. also worry about the polish of the RT implementation. having freesync over hdmi for my TV would be nice at least.
oh of course, it’s NA servers. now it makes sense.
I wouldn’t say it’s poor performance. It’s just that they expect high resolution gaming to rise, not to mention support for ML upscaling like DLSS but not limited to. Current GPU design is reaching it limits and without that fake resolution it won’t cope. There is a reason why Hopper, the next arch is rumored to be MCM and similar work is done on AMD side.
And if they went for TSMC we would not have such prices
I’m of the opinion that (a part of) the target market for RTX 3xxx is those who want to get better experiences with a bare minimum of 1440p or regular forays into higher refresh rate 4K… using these cards for 1080p may (intentionally) give disappointment compared to the extra spend on top of an RTX 2XXX…
yeah i didn’t mean it’s bad, it’s just worse than at 4k relative to the 2080 ti.
don’t understand gpu architecture that well, but adoredtv’s video on the evolution of nvidia’s architectures was pretty nice i thought. ampere seems to be more of a compute focused architecture, which would make sense because it performs really well in high resolution gaming as well as compute.
my understanding is that that results in less of a performance increase at lower resolutions because the amount of SM’s isn’t as much higher compared to turing as the number of cuda cores is so you get bottlenecks elsewhere and the power of the cores can’t be utilised as effectively.
chiplets on gpus would make sense. they’re getting really big now. tu102 is 754mm² and ga102 is 628mm². must be expensive to produce.
true. probably wouldn’t give better performance either, 3080’s quite an appropriate leap. so mostly power draw wouldn’t be so ridiculous maybe. but then they can be undervolted quite effectively it seems.
for the 3080 and 3090, yes. absolutely. what’s interesting is that that isn’t necessarily true of the 3070, and certainly not of the 3060. i think a lot of prospective 3060 buyers are playing at 1080p.
curious to see how that works out, seeing as the 3080 is quite underwhelming at 1080p, so how will that work out when the architecture is scaled down? imagine a 1080p/1440p card that loses a decent chunk of performance at 1080p. kind of a weird product.
