You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
I think you're right from a technical perspective, but I also suspect cost and capacity could play a major role in the decision here. Intel has had shrinking sales on the desktop (Intel 7 + TSMC N3), so I suspect they had pre-ordered more N3 capacity than they "need" today.
I think you're right from a technical perspective, but I also suspect cost and capacity could play a major role in the decision here. Intel has had shrinking sales on the desktop (Intel 7 + TSMC N3), so I suspect they had pre-ordered more N3 capacity than they "need" today.
That is possible and we can never know specifics of those kind of challenges. However. If the decision was based on loadings and Arrow lake.... when was the decision made? Did Intel have a two versions of the 12X tile (one TSMC one Intel 3)??
Things like this may have happened and led to the delay in Panther lake and the apparent change in layout (the Die photo from June does not look like the pictures shown by Intel at CES (which are not die photos... they are probably AI interpretations)
Panther lake, Like Arrow lake... probably made some very pragmatic decisions on ramp that saved Billion+ in either Capex or cost.
I think you're right from a technical perspective, but I also suspect cost and capacity could play a major role in the decision here. Intel has had shrinking sales on the desktop (Intel 7 + TSMC N3), so I suspect they had pre-ordered more N3 capacity than they "need" today.
That is possible and we can never know specifics of those kind of challenges. However. If the decision was based on loadings and Arrow lake.... when was the decision made? Did Intel have a two versions of the 12X tile (one TSMC one Intel 3)??
Things like this may have happened and led to the delay in Panther lake and the apparent change in layout (the Die photo from June does not look like the pictures shown by Intel at CES (which are not die photos... they are probably AI interpretations)
Panther lake, Like Arrow lake... probably made some very pragmatic decisions on ramp that saved Billion+ in either Capex or cost.
Panther Lake SKU list is a bit messy; Some takeaways:
- Ultra 7 can have 8 or 16 CPU cores, Ultra 5 can have 6, 8, or 12 CPU cores. (= confusing?). No Ultra 3 yet, and Ultra 9 is always 16 cores.
..... Add the letter X when you have 12 iGPU cores..
- They seem to list "real" power instead of "TDP" - with 55, 65, and 80W turbo power limits listed, testing will need to confirm.
- Max Frequency for the iGPU is 2.5 GHz regardless of whether it's Intel 3 or TSMC N3.
- Two "cut down" iGPUs exist -- 10 cores (from 12) and 2 cores (from 4)
- Wide variety of RAM support (good for this year's market); "9600" speed definitely helps boost the iGPU
Intel is laying down the iGPU guantlet in their slides:
"82% faster than AMD HX 370", while consuming less power: 45W vs 53W sustained
and
"10% faster than Nvidia RTX 4050 mobile", while consuming 25% less power (45W v 60W sustained)
That should put Panther Lake neck and neck with the 256-bit bus AMD Strix Halo (AMD Ryzen AI Max 395).
I want to see third party reviews, but if true -- this is very impressive.
Moore's law pointed out that to get the "82% faster than AMD HX 370", they used a the AMD AI 370 with 7500 LPDDR5, versus the Intel Ultra X9 388H, with 9600 LPDDR5. As you know, memory bandwidth has a huge influence of graphics performance.
Moore's law pointed out that to get the "82% faster than AMD HX 370", they used a the AMD AI 370 with 7500 LPDDR5, versus the Intel Ultra X9 388H, with 9600 LPDDR5. As you know, memory bandwidth has a huge influence of graphics performance.
It's fair that AMD's APUs are almost always bandwidth limited, and Intel cheated slightly with 7500, but the HX 370 is rated for 8000 max (6.6% faster). Panther Lake is rated for 9600 max. FWIW HX 470 just announced supports slightly faster 8533.
AMD often does similar - when Zen 4 (and 5) launched they gave reviewers OC 6000 kits while their chips only officially supported 5200 and 5600 speeds respectively.
