First of all, this is a very cool bench and I thank the creator for making it. It's an easy way to show these limitations to people and I appreciate it; especially the 1%, .1%, and minimums. Awesome!
I honestly wish W1z would include something exactly like this in his suite. You're doing great work and it helps simplify things for a lot of people, I think!
I appreciate ya'll posting the 1080p SW/HW Lumen results for 9070xt/7900xtx. I was curious if my theory would hold up about both being able to keep ~60fps mins in most any rational sitch...and AFAIK it does.
Obviously that requires overclocking on the 7900xtx, and UV on the 9070 xt, but still...I think it holds up? Should be able to keep 1080p (or 1440p 'quality' up-scaling) and 60fps mins AFAIK; in software if not hw.
If only those XTX owners could get FSR4 and/or AMD would improve 'performance' up-scaling...That part will stay pretty great imo (especially if you're a native 4k native user and/or can use the extra ram/compute).
Not to say it isn't still great at 4k native for raster...but YKWIM. It could be a nice catch-all if that were to happen; worse-case scenario of having upscale from 1080pRT and the compute/ram should help it.
Even over a 9070 xt (or even better) potentially, at least in some (perhaps weird to people that don't run a 4k display) situations.
Just curious
AVA, did you UV at all to increase the clock? I'm kind of curious what it would take to keep 60mins in a sitch like where HW Lumen is used (Where you have 58.5).
I think that's similar to many things also limited by RT (say
Spider-man 2 max RT or
Wukong at 1080p)...should be very similar. I know you saw my post in the other thread about ~100 in Time Spy or 20k in Unigine.
Not perfect allegories, but close approximations. This would probably work much better for closer to definitive I think.
FWIW, I think core/ram equalization is something like ~3136/20000, and I don't *think* you'd need to overclock the ram to keep 60fps? I could be wrong. Shouldn't take much to get ~2.5%.
Trying to figure it out bc I think it helps a lot of people when you can say "adjust to roughly [x clock]", especially if it's just a UV on the core (which again, you might need bw...IDK your clock you have right now).
I think a lot of people are making the mistake of trying to run the core too hard and don't understand the cache only gives bandwidth (it's main limitation) at ~1/2 the rate of the memory; so it helps to adjust both. Look at Roman's video, for instance, where he has his set up pretty perfectly (something like 3369/21488), but I know all can't run that high, and frankly, I don't think they *need* to and could save power.
Considering 1440p quality up-scaling obviously performs a *little* better than 1080p native, and already would keep 60 where 1080p is slightly lower (as Wukong or your bench shows), it'd be a nice catch-all to know that approximate level of required perf for both 1080p and 1440p users to keep 60mins. And, ofc, if people can save a little power by lowering power limit to keep that level stable.
If you don't have the time, I totally get it. And again, perhaps I am wrong and it's not possible on all cards (1440p quality up-scaling should still keep 60 mins), I was just trying to figure out average power in worse-case 1080p60 scenario, as I think a lot of cards in the future will scale off approx whatever that required speed is. Ofc, they may also scale off quality up-scaling too (in which case it should already be fine).
Thanks again to all of you.
