SiliconFly
406 posts
@diamondrapids 我得马屁一下，读书超多的人我见多了，里面傻逼的比例不比普通人群低。所以关键不是读书多少，有些人就把你说的这些读一百遍，他那脑子也没办法把所有事情串起来你信么？读书多，还能融会贯通，这才是真本事。
"Hey, here's a demo that's still early and unoptimized, coming out end of year hopefully. We're still getting it to work and this is an early alpha." "We shoved it onto two separate GPUs to ensure the base game runs fine and the DLSS works well enough for the demo." "We have 5090s, so why not run it on those. We could have used two GT730s for the lolz, and you would have asked why, but we had 5090s on the desk." Everyone starts screaming 🎶OMG 🎶 🎶 IT REQUIRES 🎶 🎶 TWO 5090s!!1!🎶 🎶 WONT SOMEONE PLEASE THINK OF THE CHILDREN 🎶 I really can't fathom the dumb. You know console games are made on dev kits that are vastly faster than the consoles they run on. It's well documented, especially in early cycle design. To anyone that claims I'm pro NVIDIA or I'm not taking people's complaints seriously, go look at my previous NVIDIA tweets. I don't get passes for events. I find issues with their messaging all the time. This is just one where seemingly most of the TechTube space couldn't be bothered to apply the most basic critical thinking, or didn't ask a basic question or two. Instead, let's jump straight to drama and have it as the cornerstone of the outrage and content. Because clicks, I guess. There are a thousand other things to mention about DLSS5. I gave them a ton of feedback when I spoke to the engineers at the demo in person (most of the time it looks like a repositioning of the light source more than anything else, it was shown on games that aren't that high in graphics fidelity (I'm actually disappointed in modern AAA if this is what we have), half the time it looks simply like the additional tesselation we saw in DX11, and it can ignore a number of doubly-relected shadows). Here's the kicker though. NVIDIA has thousands of employees working on its gaming portfolio. I've seen so many people complain 'why did they work on this when they could have worked on other stuff'. They ARE working on other stuff. You might be able to only think about one feature at a time (make sure you remember to breathe), but these things are all developed in parallel with other features. If anyone has enough money to have teams invested in researching and developing a bucket load of features, it's NVIDIA. Also, you can turn the feature off. Imagine complaining about a TV channel you don't watch, just because it's on the guide and the cable service promoted it and needed two 8K TVs to do so because it was early 3D. I'll say it again. I really can't fathom the dumb. Or rather I can, and I'm just amazingly disappointed on this hardware critique. I'll get backlash from other media just for this post, sure, that much is certain (and thanks for obliging, much appreciated). But perhaps I should just bring @ctnzr on the show for an interview later in the year. He's a really cool dude anyway.
AMD "RDNA 5" to Heavily Boost Shader Performance in Games with New Dual-Issue Pipeline tpu.me/zzfk
Today I analyze #Nvidia DLSS 5 - the latest AI Slop being forced into your favorite games!🔥 I also interview @BryanHeemskerk, Lead Art Director at @Massive_Damage to get his early opinion on DLSS 5... youtu.be/0Tiw7blVRZg
Im a speaker presenting InferenceX and had to ask a top executive for passes fwiw We aren't "analysts" either lol
Rumor: Tesla is going to build its own EUV lithography machines because ASML is “too slow.” to supply to the new chip plant. ASML currently produces <100 EUV machines per year. Apparently that’s not fast enough for Elon Musk. The new plan: • 1,000 EUV machines per week starting 2027 • 10,000 per week by 2028 And since HBM memory is also a bottleneck, Tesla will casually build 10 new memory fabs in the next two years producing 5× the rest of the industry combined. Meanwhile, Micron Technology, SK Hynix, and Samsung Electronics are panicking.
