MediaTek Dimensity 9300 announced with big-core only CPU, boosted GPU with ray-tracing
- e
- e64pro
- ssg
- 07 Nov 2023
Anonymous, 07 Nov 2023I believe ARM big.Little architecture currently max at 12 c... moreoh the 12 core limit is kinda sad,also maybe overclock gpu to 2 ghz or 1.5 ghz since laptops have big batteries compared to phones
- M
- MaiMai Playa
- wrq
- 07 Nov 2023
Anonymous, 07 Nov 2023I believe ARM big.Little architecture currently max at 12 c... moreI don't think so, 4Ghz for turbo/burst speed would be doable but if they did make one they probably going to run max 3.8Ghz most of the time.
It's also preferable to have big.LITTLE architecture or at least clock some of the core higher to inflate single core benchmark.
- ?
- Anonymous
- XUv
- 07 Nov 2023
e64pro, 07 Nov 2023would be better if it didnt have any mid tier cores and onl... moreI believe ARM big.Little architecture currently max at 12 cores.
Yes to 4GHz, for notebook, 4GHz is dooable.
- e
- e64pro
- ssg
- 07 Nov 2023
Anonymous, 07 Nov 2023I hope the pc version have 12 X core instead of 4 xcore + 4 a720would be better if it didnt have any mid tier cores and only prime cores which all are clocked at 4ghz and maybe even 15 cores
- ?
- Anonymous
- XUv
- 07 Nov 2023
e64pro, 07 Nov 2023imagine a pc version of this chip,would be interesting to s... moreI hope the pc version have 12 X core instead of 4 xcore + 4 a720
- e
- e64pro
- ssg
- 07 Nov 2023
imagine a pc version of this chip,would be interesting to see it against Qualcomm x elite and apple m3
- M
- MaiMai Playa
- wrq
- 07 Nov 2023
For the AI performance.... hard to believe ngl, even an rtx with xformers would need a couple of second for stable diffusion, but if they are talking about 128x128 resolution 1 step then it makes sense I guess...
I've never run SD on phone but my rtx can generate a 1024x1024 sd1.5 in about 10 seconds, now they claim they can do it less than a second?
And for LLM it's even worse, LLMs are notoriously hard to run with people literally buys ToTL rtx 3090 just to run a quant model, but I can see it's done on a phone like red magics with 24gb ram though, albeit RIDICULOUSLY SLOW!
for the perspective, up to this point my comment is 171 tokens long.
I have seen people saying you can get 2-3 tokens per second with s23U or iphone (forgot which) pro max running mistral-7b (and media tech claim this can go up to 33b) with zero context, that means for it to generate this comment up to my token count (with best case scenario) would need at least 60 seconds!
now imagine a model almost 5x bigger!
worse yet, if we fill the context window with only 512 tokens that 2-3t/s will goes down to only 0.5t/s!
As of now, unfortunately, AI in phones is still a scam, look up google's tensor which literally said to be bUiLT fOR Ai but they offload all of the AI task to google server smh... this doesn't even mention the privacy implication with sending google your photos...
- ?
- Anonymous
- q3R
- 07 Nov 2023
IpsDisplay, 07 Nov 2023I saw the video with closed caption English I am mega impre... moreThere was a time before big.LITTLE architecture though...so technically all of those CPU's would have been big only cores.
- M
- MRSTEK
- PWc
- 07 Nov 2023
ashrobb, 07 Nov 2023Watch the latest video in Chinese Geekerwan channel. It bea... moreDon't trust their prototypes, wait for production model reviews...
- M
- MRSTEK
- IW%
- 07 Nov 2023
Never trust MTK's reference designs or prototypes. They've been proved to be way superior than production models. D9000/8100/9200 prototype reviews were far from real-life experience.
Numerous Xiaomi/BBK Dimensity users in China, especially Xiaomi Dimensity 9000 users and Vivo Dimensity 9200 users, have witnessed lackluster experience after fhe spectacular prototype reviews...
- C
- Curious
- 2Tu
- 07 Nov 2023
IpsDisplay, 07 Nov 2023I saw the video with closed caption English I am mega impre... moreYeah.
But I am waiting for a power efficient SoC, which can run any app so fast still conserving energy, having normal battery watt/hour which can run whole day
- ?
- Anonymous
- pKL
- 07 Nov 2023
Anonymous, 07 Nov 2023They always speak about 40%+ more performance and power eff... moreTrue especially when you read 30-50% improvement. Either screen size or apps themselves are eating more power. Plus Google and Chrome are remarkably quiet about their power usage and the volume of video ads on webpages must be eating into power.
- ?
- Anonymous
- 8v@
- 07 Nov 2023
Anonymous, 07 Nov 2023X100 Pro ➡️ D9300 X100 Pro+ ➡️ 8g3 X7 ➡️ D9300 X7 Pr... moreInferior is probably not the CPU performance but camera processing and AI stuff.
- C
- Capt. Cake
- CGH
- 07 Nov 2023
Anonymous, 06 Nov 2023I'm still thinking 🤔🤔🤔 If x100 & x100 pro using 9... moreVivo always tune it not to surpass that of the pro+
- I
- IpsDisplay
- 4ra
- 07 Nov 2023
Curious, 07 Nov 2023Yes but in chinese channel of geekerwan . It has same chann... moreI saw the video with closed caption English I am mega impressed but the 9300!!
This is a first gen first ever big core only processor
Those are some incredible results for a experimental approach like this
As for other channels I watch goldenreviewer and a bunch others in Chinese lol
- ?
- Anonymous
- KRW
- 07 Nov 2023
Finally Mediatek able to catch up ... This might not be the best but it still among the best .. glad we able to see some alternative to Kirin as 3rd force (just ignored exynos)
- N
- No Regrets
- 39y
- 07 Nov 2023
Not Again, 06 Nov 2023Vivo X's have split the Soc between Snapdragon and Dim... moreQualcomm Snapdragons are getting even more costlier. This may be where Mediatek turns the corner especially in foldables as they all seemed to be Snapdragons
https://www.gizmochina.com/2023/06/05/snapdragon-8-gen-2-apple-a16-bionic-costs/
- ?
- Anonymous
- 0p}
- 07 Nov 2023
X100 Pro ➡️ D9300
X100 Pro+ ➡️ 8g3
X7 ➡️ D9300
X7 Pro ➡️ 8g3
14T Pro ➡️ D9300+
14 Pro ➡️ 8g3
What a coincidence it is inside inferior phones compared to 8g3.
- ?
- Anonymous
- vgM
- 07 Nov 2023
Anonymous, 07 Nov 2023They always speak about 40%+ more performance and power eff... moreLol another stupid comments. This user doesnt know how efficiency works