Chrispy_Nvidia is on 8nm Samsung this time around. They may have an impressive design with Ampere, but Samsung's 8nm is not good - at least it's nowhere near as mature or as efficient as TSMC's 7nm EUV.

Can you please elaborate on why you think this is so?

Super XPAll GPU prices are out of whack. All low, mid, med, high and enthusiast prices. They are ALL overpriced!

I agree and I've seen videos by folks like Tom who say that Nvidia will charge even more. In fact, I hope they do because even while there are total knobs who will pay anything for the top card/performance most profit ironically comes from the much higher numbers from lower end sales. The Radeon 500 series cards have sold well because more people could buy them. I'd really like to see AMD release most of the 6000 stack within two months of all the models and see the low end cards not go over $150 USD. It's just a waiting game at this point though. The cheapest 5500XT is $180! That's just too much. I will be buying a 6950XTX or whatever it's called when it comes out but I'll be going from below average income to well above it.

JAB CreationsI agree and I've seen videos by folks like Tom who say that Nvidia will charge even more. In fact, I hope they do because even while there are total knobs who will pay anything for the top card/performance most profit ironically comes from the much higher numbers from lower end sales. The Radeon 500 series cards have sold well because more people could buy them. I'd really like to see AMD release most of the 6000 stack within two months of all the models and see the low end cards not go over $150 USD. It's just a waiting game at this point though. The cheapest 5500XT is $180! That's just too much. I will be buying a 6950XTX or whatever it's called when it comes out but I'll be going from below average income to well above it.

I'm OK with that so long as AMD doesn't also price hike. If they do, the GPU economy is going to tank IMO.

paragonCan you please elaborate on why you think this is so?

Hints all over the place really; Samsung 8nm isn't constrained because it's not a particularly desirable node, it simply doesn't hit the efficiency or clockspeeds that customers want for desktop/server/laptop power budgets. Most of the customers of Samsung's 8nm are in the smartphone/embedded market. It's a cheap process - probably because Samsung can't afford to sell it for more - it's simply not good enough to compete with either of TSMC's 7nm nodes, and is priced against GloFo's 12LP+ FinFET node. Let's face it, Samsung as selling their manufacturing process to customers for as much as they possibly can. If it was competitive with TSMC's 7nm it would be priced like TSMC's 7nm.

More fuel to the fire is that the most popular product built on Samsung's 8nm is its own Exynos phone processors but Samsung's flagship Exynos 990 is inferior to the Qualcomm Snapdragons built on TSMC. It's not an apples-to-apples comparison because we're talking about two different ARM designs here as well as two different process nodes, but ARM cores are ARM cores so in like-for-like testing, the chip built on the best manufacturing process ought to clock higher, for longer and win. In every tested case, that victory does indeed go to TSMC's 7nm Snapdragon 865 chips, even when the test is between two otherwise identical Samsung S20 Ultra phones and running the same Samsung OS.

Tom from MLID is a pretty solid source. Here's a specific example that includes a lot of educated guesswork and industry-sourced info that builds a picture of Samsung's 8nm:


This video is a month old and some of the info from it (such as the 21Gb/s GDDR6, for example) was only confirmed by the rest of the press in this last week.