Vince Longman

[Vince Longman @VinceLongman]

@Reptalicant WTF. Should've kept the P-core L2 the same. Dropped 2x M-cores and reduced the M-core L2

[Vince Longman @VinceLongman]

@IanCutress @awscloud Arm needs to hurry up and release the Neoverse V4 & V5 lol

[Vince Longman @VinceLongman]

@ExoticSpice101 Woah, those M-cores are only about half as big as the P-cores. They're decently smaller than Qualcomm's P-core & Arm's C1-Premium

[Vince Longman @VinceLongman]

@gr_roman1770 @ExoticSpice101 The C1-Pro is still closer to an E-Core

C1-Ultra + C1-Premium would be more interesting, essentially copying Apple's Super+Perf in their M5 Pro/Max

[Vince Longman @VinceLongman]

@VideoCardz Every two years for RTX Spark is very disappointing

[Vince Longman @VinceLongman]

@VideoCardz I wonder if the N1 is a new die, or just the GB10's weird S-die cutdown

[Vince Longman @VinceLongman]

@lafaiel Qualcomm really messed up by not getting into the server market sooner. Their P-cores are small with the highest perf/mm2, plus are close in efficiency to Apple/Arm. Making them perfect for the server market

[Vince Longman @VinceLongman]

@MichaelSmicqfw @TheGalox_ Ahaha, same for the Tensor G5, which uses an external Samsung modem https://t.co/FaUBROW2dy

[Reptalica @Reptalicant]

Tensor G5

CPU Cluster 16.855mm²
X4 : 2.25mm², 1.55mm² no L2$
A725 : 0.98mm², 0.83mm² no L2$
2x A520 : 0.67mm², 0.585mm² no L2

GPU : 18.3mm², 1 SPU 6.6mm²

NPU 11.625mm², 1 tile 0.68mm²
MEM PHYs : 1.264mm² x 4
DSP : 4.77mm² (Q8 core 1mm²)
Always-on Subsystem : 5.22mm²

[Vince Longman @VinceLongman]

@MichaelSmicqfw @TheGalox_ Incorrect, Google's Tensor G5 (121mm2) is about 22% larger than Apple's A19 Pro (99mm2). Thus probably 25-30% more expensive. That's how amazing the Apple Silicon team are

[Vince Longman @VinceLongman]

@sayandeep0307 @Reptalicant Nope, more cores won't help sub 5W when there's like only 2 cores active lol

[Vince Longman @VinceLongman]

@lafaiel @saurax04 It's #2 the X2 Plus has worse ST than their lower power 8E g5 in phones. Qualcomm just loves their own foot, must be the Radeon heritage 😂😂

[Vince Longman @VinceLongman]

@techanalye1 Die sizes??

[Vince Longman @VinceLongman]

@lafaiel A19P's P-core 12.11 at 7.4W = 1.64/W

M5P's M-core: 10.3 at 4-5W = 2.06-2.58/W, still a big 26-57% uplift

M5P's M-core: 10.3 at 2.5-3W = 3.43-4.12/W, insane 209-252% uplift

[Vince Longman @VinceLongman]

@lafaiel jht5132 also said the M-core scores 508 in Cinebench 2026 at 5W

I think 4-5W is more accurate, standard methodology: system power minus idle

2.5-3W is only M-core itself, but that's non-standard methodology

Still 10.3 at 4-5W is still a huge lead in perf/W vs the industry

[Vince Longman @VinceLongman]

@jukan05 For the X3 Elite?? Or server chip??

[Vince Longman @VinceLongman]

@Cartidise Do they have Add me? Its genuinely very very useful

[Vince Longman @VinceLongman]

@lafaiel Interesting 10C setup, 1+4+1+4? I assume 1x Ultra, 1x Premium, 4x Pro (mid) & 4x Pro (little)??

[Vince Longman @VinceLongman]

@UniverseIce @faridofanani96 S27 Ultra without the S Pen? So just an S27 Ultra but with a slightly larger battery? That doesn't really sound appealing

The S27 Pro should be a 6.3" S27 Ultra Mini. Competing with the iPhone 18 Pro, Pixel 11 Pro & Xiaomi 18 Pro

[Vince Longman @VinceLongman]

@yo80526 @leclercsletters He's harvesting

[Vince Longman @VinceLongman]

@toniievych @MCH2024 @geekbench Lol that quote is regarding comparing ARM systems with & without SME. Since Arm's SME is quite new, hence only recently supported by GB6 & some CPUs

It wasn't about ARM systems vs x86 systems. SME support actually levelled the playing field as
GB6 already supported AVX/etc