The Performance & Power Improvement Of Steam Deck OLED's 6nm APU

[DrYak]

I think it'd be interesting to see how the 6nm and 7nm APUs compare at lower TDP settings too,

My bet would be that FPS performance wise both should perform virtually the same (baring some minor discrepancies as the RAM is clocked higher on OLEDs, and the OLED features additional refresh rates), but lower power consumptions (as long as reasonnable OLED brightness settings).Basically, my prediciton is that is should still looks the same, but you should be able get even more hours out of the battery.

Let's see if real world number support this bet.

[Anux]

Does anyone find it strange that while the OLED has a lower power consumption during tests, between tests it pulls sometimes 2-5W more then the LCD. It is this significant, that the OLED actually uses (slightly) more Joule per run on Superposition and Gravitymark.

Especially in idle OLED should be the clear winner, there must be some driver problems on the new platform or the brightness was just much higher or a white screen between benchmarks?

[catpig]

Just a small note: "6nm" is not a shrink compared to "7nm". These are not measurements but marketing terms, which is why I put them in quotes. You can find some details here: https://en.wikipedia.org/wiki/7_nm_p...cess_offerings

[drakonas777]

Just a small note: "6nm" is not a shrink compared to "7nm". These are not measurements but marketing terms, which is why I put them in quotes. You can find some details here: https://en.wikipedia.org/wiki/7_nm_process#7_nm_process_nodes_and_process_offeri ngs

N6 has a higher average density, ergo it's an effective shrink. Perhaps a "die shrink" would be more accurate term for those who are anal about details.

[cj.wijtmans]

OLED burn in would have been less of a problem if we could buy the panels new and replace them. My samsung s7 edge has a bit of a burn in but still going strong. Needs refurbishing though but no battery nor screen to buy new to do it.

[Anux]

Just a small note: "6nm" is not a shrink compared to "7nm".

How do you explain the die size 163 mm² vs 131 mm² if not from a shrink (both have same amount of transistors)? On 6nm they use EUV which should lead to density improvements vs quad-patterning.

[Weasel]

LCDs aren't impervious either, the progression of time affects all, and consumer goods are not built with longevity in mind. Take a look at your stored devices that are 15 to 20 years old, like him you might find that they did not fare so well even with no usage. You will also find that your devices that are currently 10 years old may suffer the same fates in a few years too.

I have a 17 years old 27" LCD monitor that's been used daily (averaging ~16 hours per day) that is still in pristine condition and works as on day 1.

My friend had to replace his OLED after 2 years.

I simply don't care about theory. This will remain a fact for me.

[catpig]

How do you explain the die size 163 mm² vs 131 mm² if not from a shrink (both have same amount of transistors)? On 6nm they use EUV which should lead to density improvements vs quad-patterning.

N6 has a higher average density, ergo it's an effective shrink. Perhaps a "die shrink" would be more accurate term for those who are anal about details.

See the links - "6nm" N6 and "7nm" N7+ have virtually identical transistor density, whilst "7nm" N7 and N7P have quite a bit less. Don't ask me why TSMC's marketing chose to label them in this fashion, I'm sure they had their reasons. In the end the "7nm" and "6nm" processes are usually grouped together anyway, I suppose (and same with "5nm" and "4nm").

As for the reduction in area - maybe I should've been clearer, I didn't say there's no improvements between them. There's many ways to improve production process - and chip design - other than making things smaller.
However, the end result in terms of perf and power is, as the benchmarks have shown, only marginally different. Whether 131mm² at "6nm" is cheaper than 163mm² at "7nm" we'll probably never know, and since the OLED deck has many other changes the retail price of the overall device can't help either. I'm assuming you quoted the chip sizes correctly.

Edit: I forgot, our host DID write "die shrink" Although it's noteworthy that the shrink on this chip has been quite a lot greater than what would be expected solely by the change in process, going by the numbers cited on wiki. But then, of course, there's a lot more to average density than production process alone.

[Anux]

My friend had to replace his OLED after 2 years.

That sounds more like a manufacturing defect than dying of old age. With LCDs you have brightness loss and often a yellow tint, most people wont recognize it because it happens gradually and they have no reference. Also LCDs that displayed the same picture over very long periods will show burn in.

With oleds you have the problem of the first few generations, that is to be expected. LG gives 5 years burn in warranty on their newer TVs.
It will take some time till we know for sure how burn in behaves on newer panels.

What I do with my oled monitor is:
activating all those pixel shift and refresh features
always use lowest brightness
have a short display off/standby time
use dark themes

As long as you don't display the same picture for days it should be fine. I played already > 100 h Baldurs Gate 3 on it (has UI elements always at same position) and don't notice the slightest burn in.

[Anux]

See the links - "6nm" N6 and "7nm" N7+ have virtually identical transistor density

N7+ is their first EUV version, N6 is the second EUV version of N7. Just ignore the naming and compare Steam Deck with process X vs Steam Deck with process Y.

Also don't give too much about any gate/fin pitch distance values because they clearly have nothing to do with real world density. Just look at the values of intels 10 nm process and then compare sram size.

There's many ways to improve production process - and chip design - other than making things smaller.

But don't we agree that they only made it smaller here?

However, the end result in terms of perf and power is, as the benchmarks have shown, only marginally different.

I would say it's the same with some error tolerance. But if they are running the same frequencies with the same voltages it is to be expected.

Whether 131mm² at "6nm" is cheaper than 163mm² at "7nm" we'll probably never know

Well why else would they go through the trouble of changing the chip? Its harder to cool and it's surface area smaller, so a redesign of the cooler was likely needed. That would all be wasted money compared to using the old design.
PlayStation does it all the time with their slim versions.

I'm assuming you quoted the chip sizes correctly.

https://www.techpowerup.com/gpu-spec...deck-gpu.c3897 https://www.techpowerup.com/cpu-spec...cpu-oled.c3398