Arc B390 vs. Radeon 8060S: Two very different GPUs for the same 14-inch screen size

[Redaktion]

How close does Intel's newest Panther Lake Arc B390 iGPU come to the Strix Halo Radeon 8060S?

https://www.notebookcheck.net/Arc-B390-vs-Radeon-8060S-Two-very-different-GPUs-for-the-same-14-inch-screen-size.1213164.0.html

[LH]

The 8060S is able to stay close in efficiency by utilizing several times the die area for the GPU. That's a big part of the reason why it is so rare (and expensive).

[Good question]

It'll be interesting to see how AMD's eventual successor to the Radeon 890M can compare against the super-efficient Panther Lake iGPU series.

Good question, because if Intel is using TSMC' 3N node for its iGPU chiplet and AMD for its APU (which contains the 890M iGPU) is using TSMC 4N, then Intel's power efficiency advantage should only be ~30% (4N to 3N -> 1 node power efficiency advantage), but the B390 is over 60% more power efficient (at a 25W power limit) and this is a 2 nodes advantage.

CP2077: 2560x1440 Ultra Preset (FSR off):
8060S is 71% faster, but power consumption was twice, did you confirm?

Test power efficiency for CP2077 Overdrive preset (yes, the 390B is going to be too weak for this preset). Maybe it could run Overdrive at 720p?

LH, yes, Strix Halo it a big, 256-bit wide, APU. While Panther Lake is 128-bit wide. This's almost half the bandwidth (9600 MT/s (PL) vs 8000 MT/s (SH)).

[On battery?]

would consume around 56 W and 110 W on the Asus and HP, respectively.

This is running connected to power outlet. Running on battery, on balanced profiles - I would presume it being closer to half this for both, so around ~28w and ~55w.

It would be interesting to see if you could test performance when on battery to confirm this. As many are interested in gaming perf. while on battery.

Both kind of meh, tbh. Halo is too power hungry and expensive. PTL won't be in any real volume until closer to end of year, so essentially out of stock paper launch for several months. Though it is interesting seeing what gains of future laptops will look like.

Seems if you need something rn, the best deals are either kraken point, LNL, or Arrow Lake-H/HX.

[Terror Byte]

Halo also has 16 full Zen 5 cores, unlike Panther Lake with only 4P cores, so it's idiotic to attribute all that extra power to just the iGPU. In fact it's silly to compare these two at al,l they are targeted at completely different segments. Halo is not for thin or light and low power and is on 4nm.

[opckieran]

It'll be interesting to see how AMD's eventual successor to the Radeon 890M can compare against the super-efficient Panther Lake iGPU series.

Good question, because if Intel is using TSMC' 3N node for its iGPU chiplet and AMD for its APU (which contains the 890M iGPU) is using TSMC 4N, then Intel's power efficiency advantage should only be ~30% (4N to 3N -> 1 node power efficiency advantage), but the B390 is over 60% more power efficient (at a 25W power limit) and this is a 2 nodes advantage.

CP2077: 2560x1440 Ultra Preset (FSR off):
8060S is 71% faster, but power consumption was twice, did you confirm?

Test power efficiency for CP2077 Overdrive preset (yes, the 390B is going to be too weak for this preset). Maybe it could run Overdrive at 720p?

LH, yes, Strix Halo it a big, 256-bit wide, APU. While Panther Lake is 128-bit wide. This's almost half the bandwidth (9600 MT/s (PL) vs 8000 MT/s (SH)).

Manufacturing nodes are pretty much pointless to bemoan at this point. Could X product be more efficient at Y node? Of course. But it's basically a pointless hypothetical unless such a product is actually launched, at which point its competitors would likely have shifted nodes too. And in all honesty, even a "2 node advantage" (across competing fabs makes a direct comparison harder too) is only going to net 15-20% improved perf/watt at today's node sizes. This isn't like the giant leaps in perf/watt we used to get 20 years ago when going from 130nm to 90nm to 65nm, for example.

Basically, it's all about what products are actually delivered. The fact that AMD COULD sell [newer product] at [better node] in some amount of time doesn't change the fact that Intel is selling Panther Lake at its node, today.

[GeorgeS]

I recall reading that Nova Lake is ALL on Intels own 18a node.

While these are very different APU's, IMHO the Intel offering scores the win for power efficiency and likely run/game time on battery power and the AMD offering win's for raw performance.

[COR]

"The benchmarks below compare the graphics performances of the 14-inch Asus ExpertBook Ultra to the 14-inch equipped with the Core Ultra X7 358H and Ryzen AI Max 395+, respectively."

HP ZBook Ultra G1a carries Ryzen AI Max+ PRO 395.

[Good question]

And in all honesty, even a "2 node advantage" (across competing fabs makes a direct comparison harder too) is only going to net 15-20% improved perf/watt at today's node sizes.

But how did Intel' B390, limited to 28W, vs 890M, manage a 61% power efficiency improvement? (and, limited to 20W, vs 880M, manage a 138% energy efficiency improvement?)
(source: notebookcheck.net/Intel-Panther-Lake-Arc-B390-performance-and-efficiency-analysis-Intel-s-new-iGPU-trades-blows-with-the-Nvidia-GeForce-RTX-4050.1212582.0.html)

(True, TSMC states the +15% are "at same power" and 30% is at same performance, maybe this helped?)

PS: Imagine a 5090 at TSMC 2N consuming only 575W/1.3² = 340W (and, of course, for LLMs: Using also 3GB GDDR7 chips, offering 48GB of VRAM (not interested in 32GB of VRAM per GPU)).

[Verification Failed]

Halo also has 16 full Zen 5 cores, unlike Panther Lake with only 4P cores
...
In fact it's silly to compare these two at al,l they are targeted at completely different segments.
...
and is on 4nm

I'd say it's more like 16 vs 12 cores because unlike meteor lake, the E-cores used here in PTL are actually good this time and more comparable to Zen 5c (minus the lack of avx-512 support).

People are mostly comparing them because these are the best igpu's here today (outside of apple and console).

I'm also unsure if there's any point looking at fab processes. It's difficult to compare nodes today when they're so highly specialized. So I don't even bother trying.. to me 18A looks like generally a positive step in the right direction, after all, PTL came out alright. Even if 18A still isn't equal to TSMC's very best and leading edge nodes yet.

But how did Intel' B390, limited to 28W, vs 890M, manage a 61% power efficiency improvement?

Likely a combination of several factors. Some of which from using better node, backside power delivery, more modern GPU architecture improvements, the interface used to connect the chiplet tiles using less power than whatever AMD uses, better memory compression tech, using a smaller bus in general reduces power needed, etc. No one single silver bullet.

[Verification Failed]

**Err, sorry, not narrower bus. That would be a reason why it's more efficient than 8060S not 890m/880m