h7per.bsky.social
to-be mathematics student
F84.1 / in the field of applied autism
53 posts
9 followers
4 following
Getting Started
Active Commenter
comment in response to
post
Huh, weird. I was specifically looking for numbers and couldn't find one >600k. I can find the 5.2mio number but not 8.5 for last weekend(only "over 3 million"), but still 5.2 is obviously much better than I had known.
comment in response to
post
I am quite excited! I have been closely following QDEL development over the last roughly 1.5 years, checking google scholar for new papers has become a daily routine.
Hoping to hear that Sharps progress from last year is being handily beaten (by whoever. TCL-Bluestar is also coming up.
comment in response to
post
Then again Elon seems pissed that he is getting hate by the left, so maybe I was wrong and it actually worked. In general terrorizing someone is a two sided sword.
comment in response to
post
-Being rationally outspoken. Essentially converting as many people as possible. I know this isn't always possible and it would be a slow process. Point out as much hypocrisy as possible.
-Violence. I know that is easy to say and I think I wouldn't, but violence is sometimes an answer.
comment in response to
post
Bit late, I know:
-Protesting. The US is extremely weak in this regard. We have here a party in Germany that is less extreme than the republican one and still drew 4 Million people to protests without even being elected. AFAIK the current protests don't even beat that, despite 4x the population.
comment in response to
post
What is the technique here? When I think Voxels, I think voxel-cone-tracing. Are you doing lightprobes casting into a voxel grid?
comment in response to
post
ChatGPT has degraded so hard its not even funny. The thing doesn't even respond to you correcting it anymore, neither 4o mini nor the fullsize model.
"hey XYZ is wrong"
"oh I see, lets try again: use XYZ"
Claude, meanwhile, impresses me. The damn thing answers like it is the Vulkan documentation.
comment in response to
post
Which would make other forms of pressure less effective.
As it were, I also care about maintaining morals and bullying is not compatible with that. I would not be comfortable bending that for something I see as completely ineffective on a rational level.
comment in response to
post
This isn't ostracization. At best this does nothing, at worst this makes him more bitter and withdrawn into his, frankly abhorrent, bubble. I am not saying that this is solvable with "the power of friendship", I am not delusional, but the more bitter he is, the harder he will fight.
comment in response to
post
I maintain, protest should be widely seen. We had 4 Million people go on the street against a far less problematic party in Germany. And the US? Not even a million from what I can find quickly, despite having 4x the population.
comment in response to
post
Calling this a protest is beyond a stretch. This is the lowest form of pointless, hateful behaviour.
You are giving him zero reason to stop what he is doing and all the reason to withdraw even more into his circle-jerk.
Best case, nothing happens, worst case he will get even more extreme.
comment in response to
post
Just to be clear, I don't like this guy, at all.
comment in response to
post
Honestly not a fan of this. Bullying in general doesn't sit right with me, even when the other guy is a bully. I also think this is not effective, you are just making him even more bitter.
I suggest for everyone who is participating in things like this to go to protests instead of slinging shit.
comment in response to
post
www.pcgameshardware.de/Geforce-GTX-...
www.pcgameshardware.de/Geforce-RTX-...
comment in response to
post
I've had this argument a couple of times. I usually suggest going back to Pascal and seeing how the 1080ti ran titles that came out before it even existed and compare it to the 4070ti and 4080(S), inflation adjusted these to flank the 1080ti.
PCGH is great for this, they don't even test with DLSS.
comment in response to
post
Or they just march on and weaken us so much that we have no choice but submit.
comment in response to
post
I literally told you they are not an existential threat today.
This is a problem for the future. If they just keep on up-arming and we are not prepared, they might get through Poland before we have a chance of stopping them. What then? Recapturing isn't an option anymore.
comment in response to
post
If we do then by the time we got production up, they will have captured half of eastern Europe, if not more. Recapturing it will basically be an impossibility. Look at Ukraine.
comment in response to
post
Not to mention purchasing power parity. Russia is well in excess of 400B worth of production in 2024, despite only having a defense budget of 150B.
If we snooze, we loose. Simple as that.
comment in response to
post
Italy having a higher GDP is nice and all, but 1. GDP is not equal to manufacturing output and 2. even if manufacturing output was high, thats not very useful when the production lines for military equipment just don't fucking exist.
comment in response to
post
How many tanks and IFVs? 0
How much artillery? If they do, basically 0 too.
How many cruise missiles? 0
How much anti air? Not aware of any, if it produces any, it will be nearly zero compared to Russia too.
comment in response to
post
Its nice that we can theoretically outproduce them but getting to that point takes time. Unless we are *actually* able to outproduce them, there is no point obsessing about some theoretical numbers.
comment in response to
post
That economic power means jack shit when Russia has higher military manufacturing numbers than the entirety of Europe across basically every sector.
Danish intelligence suggests that within 5 years they could take on the entirety of Europe's NATO.
comment in response to
post
Russia is not an existential invasive threat today, it is one for the future.
I don't think its *that* big of a blow either, to be honest. It hurts, but its not like the KA52 is the only helicopter that can fullfil its role, nor is this capability definitely required.
comment in response to
post
For completeness sake, here is native path tracing without DLSS vs native path tracing with DLSS
comment in response to
post
Since the video seems deepfried
comment in response to
post
Oh certainly, at least when compared to the first game. I am still in a hardcore playthrough of that game and going into Rattay has me dropping into the low 30s. So hearing they've improved that has me very excited to play the second.
I think we should separate CPU and GPU a bit in this regard.
comment in response to
post
I had the thought before that if developers cut the 2-3 highest graphics presets and renamed the highest below into "Ultra", a lot of people would praise the devs for optimization.
I got this feeling particularly with KCD2, don't get me wrong, the game looks good, but the tech is quite conservative.
comment in response to
post
Basically: A DLSS that integrates the less than 1spp into the previous full resolution frame instead of having a denoiser produce noise-free "samples" for it.
comment in response to
post
DLSS3 and FSR4 suffer of this too to some extent, but seemingly in a less defined manner.
comment in response to
post
In my use I only really saw weird striations when I went to Ultra Performance at 1080p. I can't say I'd view this as a major issue.
What I still find problematic is a sort of halo effect around moving objects, where the environment is dragged along. I think this is caused by lower res motion vectors
comment in response to
post
This is not so much about physics in the traditional sense as it is about GPU particle calculation. That is a task that parallelises extremely well but has a high overall cost.
In Mirrors Edge breaking a window will drop you below 10FPS on a 9800X3D
comment in response to
post
I only saw them with Ultra Performance but bump it up to Performance and the issue is gone completely. And that wasn't even the reason I went up in quality, because that specific circumstance just didn't matter to me.
comment in response to
post
Yeah, which is complete nonsense. I don't think he actually knows how temporal accumulators like DLSS work.
The artifacts he showed were obviously misreconstruction/misintegration, but this really only happens on extremely low quality presets and under extremely specific circumstances.
comment in response to
post
Oh no, I didn't read properly x) Thats my bad, not yours.
Anyway, for anyone else, there is what I wrote below, I still think its informative to more casual tech people.
comment in response to
post
When I say "take mouse input after CPU processing" I mean before issueing drawcalls, though I am guessing you could issue some first and others afterwards. So I am being optimistic here too.
comment in response to
post
Though what I have seen games do is take mouse input again after CPU processing, in which case you can basically cut out the CPU processing time from the equation.
Either way, 90ms is just not happening.
comment in response to
post
And that is completely ignoring semi-asynchronicity between GPU and CPU. You could have a game with 100ms CPU processing per frame and 100ms GPU processing per frame and then you would have the same framerate, but 250ms average latency. The rendering calls by the CPU would just be queued up.
comment in response to
post
No, there on average 150ms between input and rendering the frame, if we assume that frametime was one unified block and input was taken at the start.
100ms to 0ms just from when in the last frame interval you did your input, 100ms from rendering.
comment in response to
post
I've had that same discussion but I find these people to be more rare. What is less rare is people thinking games ran so much better a couple of years ago, when measured by contemporary hardware that is just not true.
comment in response to
post
Any word on Reflex 2 in conjunction with Frame Generation? It would seem like a natural progression for me, but there is not one bit on that topic anywhere and it seems like Nvidia just doesn't want to talk about it.
comment in response to
post
(the price figure has been derived by entering the die measurements into a wafer yield calculator and upper and lower bound are governed by 0.05 and 0.1 defects per cm^2 respectively and a wafer price of $15000)
comment in response to
post
That is not to say they don't profit greatly right now, the cards might well be a bit cheaper if there was more competition, but prices that a lot of people these things "should cost" just aren't gonna happen. 1080ti times are frankly over.
comment in response to
post
I guarantee you that this isn't just some Nvidia thing. GPUs have gotten a lot more expensive to produce. The full GB202 die is going to be in the neighborhood of 300-400 dollars per die alone. That's zero memory, cooler, power delivery, PCB etc etc. Also R&D.
comment in response to
post
Even outside of neural uses: Basically any FP16 MatMul operation should benefit, unless there will only be some very limiting abstractions around that.