This is what happens when people who knows nothing about graphics engineering writes about how it works.
It's like Trump telling people how to play dota. Imagine it and see how stupid it sounds.

Here's a short gist on WHY FP16 doesn't matter on the grand scheme of things, written by people who knows their shit:




To answer your question, if you think RPM will suddenly improve performance for all games - then you'd be sorely mistaken and will be in for a rude awakening.

Modules that has fairly high success rate of being OC'd to 3600mhz (widely available in Msia) -

G.Skill Trident Z/Royal/RGB - 16GTZR(X)*
Corsair Dominator Platinum - CMD16GX4M2B3000C15

*note There is a specific Ryzen module for G.Skill, but there are enough success reports claiming that the regular non Ryzen modules works too.

Those are some mighty lofty claims AMD. Hope you guys can walk the talk.

First things first though, please don't release the cards with a broken driver that takes 6 months to fix ffs.... again.

This post has been edited by SSJBen: Oct 29 2020, 01:36 AM

Infinity cache is not new, dev tools has been supporting it for over a decade now. The xbox 360 had what was called eDRAM and the xbox one/S has eSRAM, both were AMD GPUs. They are the same concept as what Infinity Cache is just that the latter is much larger.

I think by large, AMD should only be catching up to Nvidia with big Navi - not beat them, or so they claim. That's quite an achievement on its own, as they've been no where close to Nvidia for 2 generations now.

Beating nvidia is a tough ordeal, especially when AMD doesn't yet have their own AI upscaler. The more important task for them is to ensure 6000 series of cards "just works" out of the box. No dumb bullshit trying to jump through hoops with drivers and weird registry tweaks just to get the card stable enough to go on youtube. That IMO is much more important than "beating" nvidia.

The fact that you compared DLSS to upscaling techniques used in 4k smart TVs tells me that you're actually rather ignorant about the topic of upscaling as a whole. You're oversimplifying it.

UHD TVs generally use:
- Bicubic
- Nearest Neighbor
- Bilinear

Either of these 3 ancient techniques are used on mainstream UHD TVs, while some of the highest end TVs has more elaborate in-house techniques that rivals Lanczos and Jinc (which are far from cutting edge as they've been around for over a decade now). However, all of these upscaling techniques are pixel based (except for Bicubic and Nearest Neighbor as they are texture based). The upscaler is taking the image as is and processing it for the final resolution output. This means the upscaler does NOT KNOW what images it is receiving before it upscales and errors will often time occur, especially in the chroma range. You'd see issues with variable sharpness (which is annoying) from scene to scene, you'd see aliasing on high geometry scenes, and you'd see dithering artifacts in backgrounds with a lot of noise.

DLSS 1.0 was lackluster because it only used a single perfect frame for a game, meaning it's always the highest quality image possible. That creates an issue because the AI doesn't know what the image looks like if the frame isn't perfect, i.e - lower resolution. DLSS 2.0 fixed this as it is presented with both a high quality and low quality image, then have it in real time upscale the lower quality image to the highest one with the data derived from the former.

With this, DLSS IS NOT pixel based upscaling, it is IMAGE BASED. That's the key difference. The frames are already trained onto the super computer Nvidia leverages and is passed back onto the end-user through the drivers. The info of the image is already there before hand. Since both luma and chroma range are already pre-calculated, errors DO NOT occur.

I am indeed hung up about DLSS because it works and it's like black magic. Native res is not always better, because you're wasting GPU power on resolution where you can't perceive as obvious as you would with higher framerates (which leads to less stutter, lower latency), and better in-game visual effects (that includes the like of ray tracing). It is true that more games need to be supported and I'm sure it will, because we're not hitting native 8k60 anytime soon. We're reaching the limits of silicon process fabrication and there's only so much more transistors you can pack onto a single silicon.

This post has been edited by SSJBen: Oct 30 2020, 11:27 PM

Lol RDNA2 died before it even made an appearance. Nice job.

This post has been edited by SSJBen: Nov 26 2020, 10:27 AM

Intel ain't done lmao. You make it sound like AMD has never beaten Intel before and neither has Intel bounced back ever.

AMD isn't all guns and roses either alright. They sweep a lot of their bullshit under the carpet that comes with their ZEN and RDNA products. Just so to have a facade to trick people like you to say "Intel is done".

Fan boy? How? Been on a 3900x and 5900x, haven't bought anything Intel. I'm as objective as can be.

You just made it sound like intel has no way to climb back up and that's what I found very odd.

On second thought, perhaps I perceived your message wrongly. My apologies.

But Intel's CPUs not looking attractive is down to the media and mindless sheeps that blows their lack of advancement out of proportion. People forget or don't know that as ancient as Intel's skylake arc is, it's backed with a very mature platform. There are so much less problems on their side vs what you see with Zen 3 (and 2), even if the latter performs better for the most part.

It's not reliability or longetivity. AMD has good reliability too.

I'm talking about stability of the platform. AMD has issues with its AGESA based BIOS, so there are a lot of minor quibbles which while not system breaking, it is very annoying when they all add up. There are plenty of people suffering from WHEA errors, not that Intel doesn't have this issue either, but they are a lot more less pronounced and more specific, which is easy to pinpoint the issue and fix, where as AMD is kinda all over the place.

Not a bad showing for the 6900XT, but definitely didn't live up to the promise of beating the 3090. On paper (following MSRP that is), makes no sense to go for the 3090.

But as we know, AMD's launch for RDNA2 has been a shit show.

6600XT really is a stupid card.

It's a PS5 GPU (mostly) that costs more than PS5 and is just as equally unavailable, yet performs worse than the console.

If the 7900XTX can actually compete with the 4090 at that price point, Nvidia does actually look very stupid at this point. Time will tell.

You guys shouldn't dismiss FSR btw. 2.0 is actually pretty good right now, not quite up to DLSS level yet but it's not far behind (and I bet most people wouldn't be able to tell difference without a side-by-side comparison). Besides that, FSR can be implemented quite easily compared to DLSS.

Lol GG to 4080 unless ngreedia lowers the price.

7900XTX not only faster, but much cheaper.

Actually, most won't buy anything.

7900XTX can be had for RM5.2k,

Cheapest 4080 is RM6.2k. RM1k difference for just 25% better RT perf? It's obvious which card looks stupid here. The only saving grace for the 4080 is that it has DLSS3.

52w difference you go say 100w....?

I don't get the whole "I'm worried about power consumption thing".

Is that fair to compare Asus/XFX third party cards to Nvidia's FE card?

Here is a 4080 MSI Gaming X which is an equivalent to the TUF 4080:



Compare that to 7900XTX TUF, ain't close 100w now is it?

Also there's quite a big difference between electric tariffs between EU and MY. You can't use their calculations here. Cukur PMX say no increase in tariffs next year la.