Dimensity 9600 Pro on N2P with M5 level Single Core - Digital Chat Station
Posted by DerpSenpai@reddit | hardware | View on Reddit | 37 comments
Geekbench 6:
ST : 4200-4300 (+22% vs C1 Ultra on Dimensity 9500)
MT- 12000-12500 (+25% vs Dimensity 9500)
Original text:
https://m.weibo.cn/u/6048569942?is_hot=1&jumpfrom=weibocom
TSMC N2p Dimensity 9600 Pro (provisional name), dual ultra-large core high frequency near 5GHz. The current early design specifications for ES samples are GB6 single-core 4200-4300±, multi-core 12000-12500±
Hanselltc@reddit
We will see how much power these cores draw when running near M5 big core performance.
osemec@reddit
That's higher single core score than any desktop AMD/Intel cpu
DerpSenpai@reddit (OP)
That was implied as it has M5 level single core, the highest ST in a PC. Next Gen AMD and Intel most likely only reach M4 level
Polar_Banny@reddit
Might* or might not…
Middle_Regret9290@reddit
these numbers are looking pretty solid for early samples. the jump to n2p process is doing work here - 22% single core improvement over 9500 is nothing to sneeze at. dual ultra cores at nearly 5ghz is wild, mediatek really pushing the envelope this time.
mt scores around 12k-12.5k puts it in serious competition territory. wonder how thermals will handle in actual devices though, those frequencies gonna be spicy.
DerpSenpai@reddit (OP)
Yeah it seems that they are going the route of Apple. 2 Super Cores + efficiency, not sure if they are using 3 C2 Premium or just 6x C2 Pro with 2 sets of frequencies
Kryohi@reddit
> this can be used in Chromebooks at the same price as Windows Intel Wildcat CPUs
x to doubt
More silicon on a more expensive node.
DerpSenpai@reddit (OP)
The Lenovo Chromebook Plus 14" had a Dimensity 9400
Vince789@reddit
Yes more expensive node, but die size should be within say 5-15%
MediaTek just announced their Kompanio Ultra 910 late last year which is essentially a D9500
So far there's only two "premium" Chromebook with it:
$800 - Acer Chromebook Plus Spin 514
$830 - Lenovo Chromebook Plus
Wildcat Lake is replacing Raptor Lake-U, not Processor N100 like some earlier leakers thought
Those prices are close to mid-range Raptor Lake-U laptops, there will probably be cheaper Kompanio Ultra Chromebook later that are close to budget Raptor Lake-U laptops
CalmSpinach2140@reddit
Yeah it’s definitely not going in Chromebooks
Vince789@reddit
DCS said the D9600P is 2 Canyon + 3 Gelas-b + 3* Gelas
We already know Gelas is the C1-Pro
Canyon is presumably the C2-Ultra
Artoriuz@reddit
The leaked GB6 scores of the Exynos 2700 indicate it has a 1+1+4+4 core clusters, so I think there's a good chance Samsung is using the same 2 super cores strategy. I believe at some point Google did this too, and Qualcomm has been doing it as well.
Front_Expression_367@reddit
Google did it from the first Tensor through Tensor G2 actually, but forgone it in the next generation in favor of 2 more middle cores. Also Xring O1 from Xiaomi used 2x X925 cores and it achieved great results. And of course Apple has been using 2 super cores since forever.
Vince789@reddit
It's interesting that MediaTek won't be using the C2-Premium since they were the first to use it
I wonder if the E2700's 1+1 are both C2-Ultra or C2-Ultra+C2-Premium
Also Google's G6 is supposedly only 1x C1-Ultra, their G7 is supposedly 2+4 custom cores
Polar_Banny@reddit
Interesting, in somehow ARM allows its customers to print out SoCs with custom cores layout, am I dreaming?
Vince789@reddit
Yep, Arm provides the Core IP, which their customers can use to design their own compute clusters, with whatever core layout they want (that Arm's DSU supports)
However, customers can also pay Arm to design the whole compute cluster too (called Arm CSS)
Polar_Banny@reddit
Are you sure about Tensor G7, do you have any links to the source?
Vince789@reddit
Sorry, the OneDrive source was taken down by Google, who sued that ex-Tensor employee
But note that Google are terrible at hitting their own roadmap, like the G5 was year late
Artoriuz@reddit
I'm hoping it's C2 Ultras for the Exynos 2700, but that's nothing more than wishful thinking on my part.
GenZia@reddit
Let's see how the move to N2P translates to the desktop world.
12 cores per CCD on Zen 6 pushing 7+ GHz, alongside a massive 48MB SRAM, is exciting stuff... even for an old man like me!
Of course, these are just rumors.
MrFailo@reddit
there is no 7ghz on n2p (6.3-6.4), only on n2x that got delayed
GenZia@reddit
As I said, these are just rumors.
"Kepler_L2" is seemingly convinced that all major “flavors” of Zen 6, i.e., Venice (EPYC), Olympic Ridge (AM5), and Gator Range (high-end mobile), are on N2P and not N2X, and he is one of the better leakers out there.
Besides, there is no real data as to how far a consumer desktop-class chip fabbed on N2P can be pushed, theoretically or otherwise, in terms of frequency. At the very least, I have not personally come across any such data points.
But moving on, Zen 6 apparently has a single-digit IPC uplift over Zen 5. If that's actually the case, AMD may have to push the cores harder to make up for it… not that they have much incentive to do so with Intel mostly sitting on their hands as of late (outside the mobile market).
Point is, AMD may, in fact, have hit (or even broken) the 7 GHz barrier on the “lesser” N2P, albeit at the inevitable cost of power efficiency.
But again, just leaks and (wild) rumors so... fingers crossed.
Geddagod@reddit
Unless N2X offers double the perf uplift over N2P that N3X got over N3, then Zen 6 on N2P would have hit 6.6-6.7GHz (if Zen 6 on N2X hit 7GHz).
beneficiarioinss@reddit
Mediatek needs to ditch the 4 level cache system just like Xring did. And keeps the system level cache for the GPU because Xring also showed that they help the GPU a lot
Arm cores can match Qualcomm but the latency of those caches hurt far more than helps. El
DerpSenpai@reddit (OP)
ARM needs to ditch L3 in general, private L2 only make sense for Servers
bazooka_penguin@reddit
Was there a single phone that was able to hit 4000 consistently? 3500 seems to be where most phones with the Dimensity 9500 land.
DerpSenpai@reddit (OP)
Yeah never seen higher than 3600
jacktherippah123@reddit
This makes me excited for Apple M6 lol.
DerpSenpai@reddit (OP)
Doubtful that the M6 is that big of a leap though. Most likely 4600 IMO. 7% faster
New arquitectural gains means you are using a LOT more area
Specially with using N2, they might benefit from doing a smaller update on going to the new node because of costs and then push it to the max on a newer chip
CalmSpinach2140@reddit
? Apples M5 increased ~10% from M4 to M5 and that was from N3E to N3P. It’s going to larger than 7%.
So Mediatek increases 22% but Apple only 7% is a very funny thing to suggest
DerpSenpai@reddit (OP)
ARM was on A18 Pro level
SniperChicken39@reddit
If only they partners with Nvidia, so we can get Rubin SMs on the next one, and we can have a decent GPU not whatever garbage arm holdings shits out this morning
DerpSenpai@reddit (OP)
Their partnership is only for Nvidia CPUs on desktop chips AFAIK.
ARM GPUs are not bad at all, just need Windows and Linux drivers. Their Android performance is best in class, beating Apple and Qualcomm in anything. Even with ray tracing
SniperChicken39@reddit
Yeah but there's no BcN block on the chip so it's gunna run like shit for anything modern made for PC games or programs.
alvenestthol@reddit
It's Mediatek's fault for turning the BCn blocks off in the design to save chip area, if you look at development boards like the Radxa Rock 5B you'll find all BC enabled
By the way, it's called BCn because it's actually a set of formats named BC1, BC2, BC3, etc.
Oxezz@reddit
I have read somewhere that the Mali architecture has BCn decoding but MediaTek chooses to ship their chips with these features disabled to save on licensing cost. Still a bummer if this chip still doesn't come with it, seems to be kinda a big thing for emulation.
SniperChicken39@reddit
It almost certainly will not include it