- 이전 포스팅에서 텐서G3 성능 문제를 다뤘는데 구글 텐서 설계 주체는 가설에 영향을 크게 주는 요소인데 너무 대충다룬 느낌이었음. ( 구글 텐서G3 CPU 성능 이슈 분석. (Tensor G3, S5P9865) ) 그래서 이를 명확히 하고자 추가 정보를 체크해봄. - 확인 가능한 정보. (공식 표기는 텐서, 텐서G2, 텐서G3지만 여기서는 텐서1,2,3로 표기하겠음.) 삼성 텐서1,2,3 시스템 레벨 캐시 아키텍처, 마이크로 아키텍처(CPU?), RTL 디자인. 텐서3 인터커넥트, 메모리 서브 시스템 아키텍처. CPU를 포함한 주요 부분 프론트 엔드를 작업한 것으로 추정. 구글 pre/post silicon validation 구글 텐서 GPU RTL 디자인. 구글 텐서1,2,3 CPU, 시스템 레벨..
Seems like Google is progressively getting more involved in the design process, which makes sense if the rumours of going TSMC for the Pixel 10 are true
That might be true, but the Tensor G3 has unexpectedly worse CPU efficiency than it’s predecessors according to GoldenReviewers testing, despite the better processing node (SF4/4LPP second gen 4Nm node), supposedly same amount of cache (according to XDA) and with cores that are two generations newer. The SPEC results (will add them later) of the big X3 core are worse than the ones of the X1 core inside the SD888, which also clock at a similar clock speed.
My current theory is that there is something flawed about the Tensor G3’s CPU design and the problem doesn’t stem from the processing node, which most people seem to conclude to.
For CPU design, there’s no change from OG to G3. Samsung S.LSI has done all the frontend design and Google has done all the backend design
My current theory is that there is something flawed about the Tensor G3’s CPU design and the problem doesn’t stem from the processing node, which most people seem to conclude to
Why didnt Samsung get feedback from Google in the CPU design development process? I’m guessing Google wanted a specific function in the CPU and Samsung delivered it with a flaw.
It seems there is a chronic problem at Samsung with soc/fab development my guess is that people underestimate how godly hard it is to build a soc/fab company from scratch.
It is not easy to make the brains of a computer. Ram/SSD/display is child’s play compared to an soc.
If I was the chairman at Samsung, I would get talent from across the world for the Fab and exynos division.
Sigh…they should just get an American to develop an exynos chip and branch it off.
That might be true, but the Tensor G3 has unexpectedly worse CPU efficiency than it’s predecessors according to GoldenReviewers testing, despite the better processing node (SF4/4LPP second gen 4Nm node), supposedly same amount of cache (according to XDA) and with cores that are two generations newer. The SPEC results (will add them later) of the big X3 core are worse than the ones of the X1 core inside the SD888, which also clock at a similar clock speed.
My current theory is that there is something flawed about the Tensor G3’s CPU design and the problem doesn’t stem from the processing node, which most people seem to conclude to.
For CPU design, there’s no change from OG to G3. Samsung S.LSI has done all the frontend design and Google has done all the backend design
The main reason GoldenReviewers’ testing shows worse efficiency is because the subscore for 462.libquantum is extremely low, only 25.05, when it should be >150
It’s possibly a design flaw in hardware
But it could also be a bug in the scheduler/DVFS/something else in the BSP
Why didnt Samsung get feedback from Google in the CPU design development process? I’m guessing Google wanted a specific function in the CPU and Samsung delivered it with a flaw.
It seems there is a chronic problem at Samsung with soc/fab development my guess is that people underestimate how godly hard it is to build a soc/fab company from scratch. It is not easy to make the brains of a computer. Ram/SSD/display is child’s play compared to an soc.
If I was the chairman at Samsung, I would get talent from across the world for the Fab and exynos division.
Sigh…they should just get an American to develop an exynos chip and branch it off.