News:

Willkommen im Notebookcheck.com Forum! Hier können Sie über alle unsere Artikel und allgemein über notebookrelevante Dinge diskutieren. Viel Spass!

Main Menu

Post reply

Other options
Verification:
Please leave this box empty:
Shortcuts: ALT+S post or ALT+P preview

Topic summary

Posted by :(
 - Today at 16:08:10
Quote from: RDNA 3.5 by 2027? nah on Today at 10:06:32As for the rest 0.66x, a higher MT/s memory will be required

Yes, likely probably. It will use LPDDR5X. They could also use a last level cache similar to halo's MALL. That too could help alleviate this problem as it a acts similarly to a infinity cache or 3d v-cache. They already using bLLC (Intel's answer to this) in the upcoming high end nova lake desktop parts.

They don't even need more bandwidth than halo. As we seen from panther lake, intel's memory controllers are just more efficient. They can make do with less. So even maybe if it only did 256 GB/s it would act almost equivalent to halo @ 400 GB/s. Also I don't think halo's memory controller even does the full 256 GB/s, independent testing shows it's actually less. Not giving away sauces tho. :p Google chips and cheese strix halo analysis, if interested in nerd write-up's.

It's going to be very expensive tho. This SoC is using 2nm TSMC. So I'm expecting $3000 laptops. xD .. on the bright side maybe panther lake will be affordable by then.
Posted by RDNA 3.5 by 2027? nah
 - Today at 10:06:32
Thanks for the clarification.
Doubling the memory bus width (from the (mainstream) 128-bit Panther Lake), is what would be required to double the gaming performance. So, it would indeed need to be a 256-bit APU (like Strix Halo). As for the rest 0.66x, a higher MT/s memory will be required (probably LPDDR6(X)):

Panther Lake: 153.6 GB/s = 128-bit * 9600 MT/s / 1000 / 8.
Nova Lake AX: 409.6 GB/s = 256-bit * 12800 MT/s / 1000 / 8.
2.66 = 409.6 GB/s / 153.6 GB/s.

Compared to Strix Halo's 256.0 GB/s, Nova Lake AX' 409.6 GB/s could make it 60% faster than Strix Halo and still be at an equal 256-bit, just like Strix Halo (this is good as it's important price- and chip-size-wise).

A 256-bit APU would still be in a different price class vs Panther Lake' mainstream 128-bit, of course, but this is maybe why an "AX" is added to its name.
Posted by :(
 - Yesterday at 11:28:26
Just for clarification, in the video the 2.66x calculation is based off what currently the arc B390 performs at in games like BF6 and monster hunter wilds.

It is also stated that Nova Lake AX will use 256 bit bus and that by 2029, Titan Lake will arrive with Nvidia igpu tile, incase anyone was wondering.
Posted by RDNA 3.5 by 2027? nah
 - February 14, 2026, 21:51:57
Quote from: :( on February 14, 2026, 13:51:51
Quote from: opckieran on February 14, 2026, 08:17:16The news just keeps getting worse for AMD...

According to MLID, next year nova lake ax (Intel's halo competitor) will come with 32 xe3p cores and offer 2.6x performance improvement at minimum. Which is roughly rtx 5070 ti / 5080 mobile performance.

Strix Halo is RTX 4060 Laptop performance. If Nova Lake AX is supposed to be 5080 Laptop of performance? This is, according to the table in the vid, 2.4x = 161.7/67.4 the performance of Strix Halo's 8060S iGPU. For this, Nova Lake AX would require a similar increase of 2.4x in bandwidth.

Let's see how it could be done:
Strix Halo: 256 GB/s = 256-bit * 8000 MT/s / 1000 / 8.
Nova Lake AX: 614.4 GB/s = 512-bit * 9600 MT/s / 1000 / 8.
Nice, this does exactly fit the 2.4x = 614.4/256!
But it would be an APU twice in size over Strix Halo and not be a Strix Halo competitor, but something in an entirely different class.
Posted by RDNA 3.5 by 2027? nah
 - February 14, 2026, 21:35:47
Quote from: opckieran on February 14, 2026, 16:43:42Zen 6 APU with, oh, yeah, that's right, RDNA 3.5... in 2027...
Yeah, nah, I'm not even sure whether I should get an APU-obly RDNA 3.5 laptop now or instead a used or refurbished NVIDIA gaming laptop which at least has DLSS / ML upscaling (yes, it would be a whole different weight class, too), let alone a RDNA 3.5 by 2027.

@In some cases it can be
This is a whole different topic (6GB vs 8GB VRAM and how much Watt and the cooling is given to the 4050 Laptop vs 4060 Laptop), but on average it's something like this: 3dmark.com/search:
4050 (notebook): "Average score: 8289"
4060 (notebook): "Average score: 10506"
1.267 times = 10506/8289.

Quote from: In some cases it can be on February 14, 2026, 14:34:00But this issue is fixed in rtx 5000 series where 5050-5060-5070 all have same 8GB, so no real point spending more really..until you get to 5070 ti.
Absolutely correct. If there wasn't the AI hype, they'd use 3GB GDDR7 memory chips by now (instead of the current 2GB ones), on at least the highest performing 128-bit GPU chips, where it really would make sense to use 3GB memory chips (8 GB VRAM would then become 12 GB VRAM (4 GDDR7 chips, each 3 GB), 12 GB VRAM would become 18 GB VRAM, etc. (3 GB chips are already being used in the 5090 Laptop, which, as a 256-bit chip, would normally only have 16 GB VRAM, but it has 24 GB VRAM)).
Posted by the new king?
 - February 14, 2026, 20:17:14
"Handhelds and ultralights have a new king!"

For that price, it will be a king wo the kingdom 🤣
Posted by opckieran
 - February 14, 2026, 16:43:42
Quote from: :( on February 14, 2026, 13:51:51
Quote from: opckieran on February 14, 2026, 08:17:16The news just keeps getting worse for AMD...

According to MLID, next year nova lake ax (Intel's halo competitor) will come with 32 xe3p cores and offer 2.6x performance improvement at minimum. Which is roughly rtx 5070 ti / 5080 mobile performance.



That's okay, maybe they'll have Gorgon Halo with +3% perf over Strix Halo 😅 and then after that, maybe a Zen 6 APU with, oh, yeah, that's right, RDNA 3.5... in 2027...
Posted by In some cases it can be
 - February 14, 2026, 14:34:00
Quote from: Not almost as fast on February 14, 2026, 10:13:26Not sure I'd call it "Almost as fast", because then one could also say that the RTX 4050 Laptop is almost as fast as the RTX 4060 Laptop, but it's a tier above (20% faster). I'd define a 10% difference as "almost as fast".

Depends on the FPS you're comparing from 20% faster at 100 or 200 FPS is like 120 FPS or 240 FPS respectively. That's a whole 20 or 40 fps more. But 20% faster of 30 fps is only 36 FPS. Which I wouldn't really consider a completely different experience or one paying significantly more for.

Regarding rtx 4050 vs 4060, assuming tdps or power limits are roughly equivalent, people did say these cards were similar and still do. The only reason why people recommended
4060 over 4050 wasn't because of the 20% faster raster but because of the castrated vram amount in severely vram bottlenecked games. In games which go over 6GB vram buffer, it absolutely tanks FPS, like 1% lows down to 5 fps levels. But this issue is fixed in rtx 5000 series where 5050-5060-5070 all have same 8GB, so no real point spending more really..until you get to 5070 ti.

As such I'm personally against using percentages to determine arbitrary tiers in general, unless it's like 2x or at least 1.5x which is a big enough difference to be felt at no matter what fps you're comparing from.
Posted by :(
 - February 14, 2026, 13:51:51
Quote from: opckieran on February 14, 2026, 08:17:16The news just keeps getting worse for AMD...

According to MLID, next year nova lake ax (Intel's halo competitor) will come with 32 xe3p cores and offer 2.6x performance improvement at minimum. Which is roughly rtx 5070 ti / 5080 mobile performance.
Posted by Not almost as fast
 - February 14, 2026, 10:13:26
The Arc B390 has 20% more cores (12 vs 10) and is 21% faster (call it basically 20% -- nice linear scaling, btw). Not sure I'd call it "Almost as fast", because then one could also say that the RTX 4050 Laptop is almost as fast as the RTX 4060 Laptop, but it's a tier above (20% faster). I'd define a 10% difference as "almost as fast". Find a different term. Still good performance tho, thanks of the linear scaling, which also could mean that a, say, 16 Xe cores Arc iGPU could be 33% faster than the Arc B390, if it had the bandwidth, that is (LPDDR5X-9600 -> LPDDR5X-12800 would be required).
Posted by opckieran
 - February 14, 2026, 08:17:16
Handhelds and ultralights have a new king! Truly epic efficiency here. Performance per watt is truly off the charts for x86. The news just keeps getting worse for AMD...
Posted by Redaktion
 - February 13, 2026, 16:14:04
We had the chance to test the performance of the new Intel Core Ultra 5 338H including the Arc B370 GPU with 10 Xe cores in some benchmarks. The results are very promising and the performance difference to the Arc B390 is small.

https://www.notebookcheck.net/Intel-Core-Ultra-5-338H-Arc-B370-Benchmarks-Almost-as-fast-as-the-Acr-B390.1225433.0.html