MR. Caramel

[MR. Caramel @CaramelFor2024]

@BotMW0 @hasanthehun „Don’t have much use for the devs“… So Battlefield 7 will develop itself or what? If your game is a success, the obvious decision is to keep the people who made it and let them make a sequel

[MR. Caramel @CaramelFor2024]

@CounterStrike @aquaismissing Would be sick. 100% better then a Tie

[MR. Caramel @CaramelFor2024]

@valvesoftware 1000+ bucks without a controller …. You make even the PS5 Pro look like a steal

[MR. Caramel @CaramelFor2024]

@SebAaltonen @KuittinenPetri For LLMs tensor core performance isn’t the real bottleneck, Memory bandwidth is. A 30b Q4 model needs 15GB/token. So if your GPU has 300GB/s it will only be able to do 20t/s no matter if the GPU has 100Tops FP8 or 600 Tops. Also FSR4.1 is coming to Strix Halo next month

[MR. Caramel @CaramelFor2024]

@per_arneng @TweakTown @AMD It will only be used for „small“ AI tasks and replaces the NPU (it’s cheaper to integrate it in the CPU cores then building a whole extra NPU). So it has the same advantages basically:
-Power efficiency
-lower latency
-free up the GPU

[MR. Caramel @CaramelFor2024]

@Brparadox Kryonaut extreme… Why…. Just why?
It’s not made for this

[MR. Caramel @CaramelFor2024]

@thrutheframe @BambulabGlobal Did it change the max fan power?

Before the update when I put it to 100% it was quieter (and probably weaker) then now. It’s not an issue, I just reduced it to 70% now. I was just confused

[MR. Caramel @CaramelFor2024]

@blind_via Seems to need a lot of tuning to function precisely

[MR. Caramel @CaramelFor2024]

@wearetherace They should use the empty space to display battery charge in %. As this currently shows more about if someone is catching up or just draining the battery, then the 3 decimals

[MR. Caramel @CaramelFor2024]

@Oy_Lumo @highdownforce @DavidSB4_ Mixture of normal and alpine way. It flips in the middle basically

[MR. Caramel @CaramelFor2024]

@newerlyz @AimAdapt The video is probably fake and if not it’s a defective unit. There is no logical way how gripping this mouse hard, triggers the induction sensors

[MR. Caramel @CaramelFor2024]

@MrScience193 @gameart_3d @VideoCardz RDNA3 has 1. gen AI Accelerators,which use existing vector units for MMA.Same thing is done in RDNA4 only that it is improved to handle smaller data types. That’s also what NVIDIA does. Actually there are no dedicated ai cores in GPUs. Ever wondered why the ratio is always 32:1…

[MR. Caramel @CaramelFor2024]

@MrScience193 @gameart_3d @VideoCardz RDNA3 has them aswell mate

[MR. Caramel @CaramelFor2024]

@hms1193 Your numbers are totally wrong, I get more FPS at 2880x1800 Native (without FG) then you 1080p with framegen

[MR. Caramel @CaramelFor2024]

@GaryHaubold @NoiseesoiN @DevilsPepper @jukan05 Light task that are high quantity. For example chat surveillance (analyzing millions of chat messages for criminal activity) in real time.

[MR. Caramel @CaramelFor2024]

@NoiseesoiN @DevilsPepper @jukan05 @GaryHaubold Any model that fits into like 256mb is little enough that a modern gpu/npu will interference it so quick that there is 0 reason to have a huge chip and fast and expensive SRAM run it

[MR. Caramel @CaramelFor2024]

@Pirat_Nation Deserved for postponing Switch 2 by 2 years out of greed.

[MR. Caramel @CaramelFor2024]

@HaroldRijn @TechEpiphanyYT Most likely reel. The geometry of the 9060XT and 9070XT would allow one more „mirroring“ so a 9090XT would be 2x the 9070XT

[MR. Caramel @CaramelFor2024]

@Insight_xFF @hms1193 Just as well as Global Foundries prospered?