AMD Has A Bit More RDNA3 Graphics Driver Code Ready For Linux 5.20

[bple2137]

Any particular issues with screen capture performance? Seems to perform great here with GNOME Wayland, e.g. with OBS.

I haven't tried RDNA2 myself, last time I used AMD it was Polaris and it was pointless to even try using GPU video acceleration. I guess it's better with RDNA2, but when I saw some records, it was a bit lossy, even on Windows. Is the performance good with high GPU load as well?

[Vlad42]

What? That would be a reasonable price for a midrange, just because Nvidia and AMD milking the bad supply situation doesn't mean its unfair to charge normal prices.

Let's look at the mid-range that you are thinking of

• RX480 used a 232mm^2 die on 14nm - MSRP $239 for the 8 GB GRRD5 model
• GTX1050 used a 132 mm^2 14nm die - MSRP $109 for the standard 2 GB GRRD5 model
• GTX1060 used a 200 mm^2 16nm die - MSRP $249 for the standard 6 GB GRRD5 model

The DG2-512, is 406 mm^2, built on TSMC's N6 - a much larger die on a much more expensive manufacturing process. In addition, there is no way Intel would be foolish enough to give it less than 8 GB of GDDR6 ram, which is more expensive than GDDR5. So the price must be substantially higher than $250 in 2016 dollars for both Intel and the board partners (Asus, MSI, etc.) to break even.

This does not even take into account the fact:

1. Shipping costs tripled during the pandemic (they are coming down but have not hit pre-pandemic levels yet)
2. Tariffs were imposed in an ill conceived trade war
3. The cost of raw materials increased due to covid shut downs
4. Foundries such as TSMC increased their price
5. Inflation has occurred since 2016 ($200 in 2016 is equivalent to $250 in 2022)
6. New functionality has been added to the cards, thus increasing die area (ray tracing & ai acceleration for upscalers like DLSS)

Alchemist has simply been delayed too long to be competitive. If it had launch when it was supposed to, in 2020-2021, then it would have been fine and Intel likely would have charged about $450 - $500 for it. However, they cannot simply halve the price unless the OEMs and distributors can still make a profit and Intel is just writing off the inventory (someone would have to lose money to bring the price down that low).

Now 1), 3), and 6) should work themselves out in time. In particular for 6), we have seen a normal generational increase in traditional GPU hardware and the equivalent of 2+ generations worth of ray tracing related hardware. Eventually we will have an appropriate balance between the raw ray tracing performance and the amount of traditional rasterization performance so that we are no longer seeing disproportional increases to the amount of ray tracing hardware each generation. This should help future GPUs not disproportionally increase transistors count & die area in successive generations (die area may even go back to being flat or decreasing gen over gen).

Regardless by your logic, under no circumstances can the price of any product ever increase as it would be more than the 'normal' price it was before and thus 'unfair', whatever that means . The price for a new midrange GPU has gone up due to all of the above reasons.

[Anux]

Let's look at the mid-range that you are thinking of

Exactly, 480 and 1060 is what I would've called midrange in 2016, they were priced around 200 €/$. Hard to compare MSRPs these days, I only compare street prices. One would expect to get atleast double the performance in that segment after 6 years.

And than take a look at 6500 XT its hardly any faster than a RX 480 but still costs the same. Whatever excuses they gonne make for its increased price, they could have used the old 14 nm and just push out good old RX 480 with 233 mm² die area for dirt cheap if somehow the new process is that cost intensive that you have to charge more with half the die space. But it would still not be good enough to have the same performance 6 years later for the same price. The 6600 XT should be the new middle class at 200 € but its at 400€. In no way this could be explained with increased shiping or wafer cost.

Also new features can't be an argument for the price increase because every generation had new features and all stayed pretty much in the same price range till 2019/18.

Regardless by your logic, under no circumstances can the price of any product ever increase as it would be more than the 'normal' price it was before and thus 'unfair', whatever that means . The price for a new midrange GPU has gone up due to all of the above reasons.

Unfair was your wording, I said it's not unfair. And yes releasing new cards after 6 years with the same price and performance from 6 years ago is only going to work if you have no competition (or sell everything that you produce to miners and thirsty gamers for much to much money).
I know of no law in america that prevents you from asking any price you want. Look at printers or gaming consoles they regularly get sold under manufracturing cost. If Intel had the same cost per die area (and the current prices were justified) they would have to price their 400 mm² chip like a 3070 Ti but with those frame pacing issues and bad performance on many games no one would buy them.

BTW the used market gets flooded with old mining cards like crazy. You can get a RX 480 for under 100 € and don't need a system with PCIe 4 and RBAR support.

[MrCooper]

I haven't tried RDNA2 myself, last time I used AMD it was Polaris and it was pointless to even try using GPU video acceleration.

I was interested in "screen capture performance" specifically, separately from GPU video acceleration.

(I'm mostly testing on a Picasso laptop with Vega graphics, not RDNA2. I haven't tested with high GPU load specifically)