"Joe" comments
Joe
9 days ago
Reply:
L3 Cache 128MB (shared) -> 256MB (shared)
...read more
Joe
9 days ago
Reply:
Successor - AMD Ryzen 7 9700X -> AMD Ryzen 7 7700X
...read more
Joe
27 days ago
Reply:
Memory Bus -> 6x 16 Bit -> 3x 32 Bit
Architecture (E-Core) 4x 0 GHz –> 4x 2.64 GHz
L2 cache -> 16 MB
GPU frequency -> 1620 MHz
FLOPS -> 2903 Gigaflops
...read more
Apple A20 Pro
(11)
Joe
2 months ago
Reply:
GPU name -> Adreno 629
Architecture -> Adreno 600
GPU frequency -> 840 MHz
Pipelines -> 1
Shading units -> 256
Total shaders -> 256
FLOPS -> 430.1 Gigaflops
Vulkan version -> 1.3
OpenCL...
...read more
Joe
2 months ago
Reply:
The numbers is from a dev board which consumes up to 30W.
In a smartphone form factor (
...read more
Xiaomi Xring O3
(43)
Joe
2 months ago
Reply:
For cache its Geekerwan video, plus C1 Ultra only supports up to 3MB L2 cache.
For memory bandwidth its calculated from 96 bit and 5333MHz clockspeed.
...read more
Xiaomi Xring O3
(43)
Joe
2 months ago
Reply:
WIFI8 is only a minor improvement over WIFI7, and both are kind of overkill on a mobile phone.
...read more
Xiaomi Xring O3
(43)
Joe
3 months ago
Reply:
GPU name -> Adreno 812
Architecture -> Adreno 800
Pipelines -> 2
Shading units -> 128
Total shaders -> 256
Vulkan version -> 1.3
OpenCL version -> 3.0
...read more
Joe
5 months ago
Reply:
Also
Shading units 128 -> 448
Total shaders 1024 -> 1792
FLOPS 4014.1 -> 7024.6 Gigaflops
...read more
Samsung Exynos 2600
(41)
Joe
5 months ago
Reply:
Memory Bus 2x 32 Bit -> 4x 16 Bit
Thanks to a higher resolution die shots, its 2x (2x16) memory channels rather than 2x 32 bit memory channels.
...read more
Samsung Exynos 2600
(41)
Joe
5 months ago
Reply:
DirectX version 12 -> None
Only DXD series has DX support.
...read more
Google Tensor G5
(31)
Joe
5 months ago
Reply:
L2 cache -> 512 KB
L3 cache 4 MB -> 3MB
...read more
Joe
5 months ago
Reply:
Successor Qualcomm Snapdragon 6 Gen 4 -> Qualcomm Snapdragon 695
...read more
Joe
5 months ago
Reply:
Vulkan and OpenCL supported API was provided by Qualcomm.
Since Vulkan support is 1.3 that means Adreno 710 or 810.
...read more
Joe
5 months ago
Reply:
Vulkan version -> 1.3
OpenCL version -> 3.0
...read more
Joe
5 months ago
Reply:
Glad to help.
Vulkan version ->1.3
OpenCL version -> 3.0
...read more
Joe
6 months ago
Reply:
Cores turns off under low power use cases.....
...read more
Samsung Exynos 2500
(33)
Joe
6 months ago
Reply:
L1 cache -> 128 KB
L2 cache -> 3 MB
L3 cache -> 16 MB
SLC -> 16 MB
...read more
Samsung Exynos 2500
(33)
Joe
6 months ago
Reply:
1x 2.75 GHz – Taishan Big -> Taishan v124 Big
4x 2.27 GHz – Taishan Mid -> Taishan v124 Mid
4x 1.72 GHz – Taishan Little -> Taishan v124 Little
L2 cache -> 2MB
L3 cache -> 12MB
SLC ->...
...read more
Hisilicon Kirin 9030
(11)
Joe
6 months ago
Reply:
GPU clock isn't known yet, it should be a small overclock over the 4G2 955MHz.
...read more
Joe
6 months ago
Reply:
Successor Samsung Exynos 2500 -> Samsung Exynos 2400
...read more
Samsung Exynos 2200
(11)
Joe
6 months ago
Reply:
L1 cache -> 192KB
L2 cache -> 3MB
L3 cache -> 16MB (shared)
SLC cache -> 16MB (shared)
...read more
Samsung Exynos 2600
(41)
Joe
6 months ago
Reply:
It seems that you updated the GPU here incorrectly.
Kirin 8020 GPU ->2CU -> 512ALU
Kirin 9020 GPU -> 4CU -> 1024ALU
Even though they're both Maleoon 920, the config is different.
...read more
Hisilicon Kirin 9020
(16)
Joe
6 months ago
Reply:
Yes this is what I said, the Pro is 1+4+4 while the non pro is 1+3+4
You might want to add the non pro version (Kirin 9030) which has some minor difference on the clockspeed and GPU as well.
...read more
Hisilicon Kirin 9030
(11)
Joe
6 months ago
Reply:
Thanks, this is remaining.
FLOPS 1720.3 -> 860.2 Gigaflops
...read more
Joe
6 months ago
Reply:
Whats wrong with several GPU specs disappearing from some SoCs?
...read more
Joe
6 months ago
Reply:
Memory frequency -> 5333 MHz
Max bandwidth -> 85.3 GB/s
...read more
Samsung Exynos 2600
(41)
Joe
6 months ago
Reply:
Pipelines 4 -> 2
Total shaders 1024 -> 512
FLOPS 1720.3 -> 860.2 Gigaflops
...read more
Joe
6 months ago
Reply:
Cortex-X4 L1 cache is always 128KB.
L2 cache source https://x.com/Kurnalsalts/status/2012128650473447579
...read more
Joe
6 months ago
Reply:
The GPU clock on wiki was taken from this site which was wrong, its 1620 MHz as per Geekerwan.
...read more
Apple M5
(4)
Joe
6 months ago
Reply:
Ariel and Magellan are GPU codenames, they are not architecture.
Both X540 and X940 are Xclipse 2nd generation architecture.
...read more
Samsung Exynos 1580
(23)
Joe
7 months ago
Reply:
Cortex A78 can't have more than 4MB L3 cache, also cache is very expensive.
...read more
Samsung Exynos 1380
(11)
Joe
7 months ago
Reply:
Shading units 24 -> 12
FLOPS 370.6 -> 247.1 Gigaflops
...read more
Joe
7 months ago
Reply:
Battery is not the issue, can easily deliver the power.
Heat is the main limitation.
...read more
Joe
7 months ago
Reply:
Yes both M4 max and M5 max have 32x16 bit memory channels.
Memory Controllers can't be more than the physical channels themselves, if there are 32 memory controllers, that means there are at...
...read more
Apple M4 Ipad
(5)
Joe
7 months ago
Reply:
https://en.wikipedia.org/wiki/Apple_silicon#Comparison_of_M-series_processors
https://en.wikipedia.org/wiki/Apple_M4
...read more
Apple M4 Ipad
(5)
Joe
7 months ago
Reply:
https://x.com/Kurnalsalts/status/2027777240772317622/photo/1
Another dieshot that shows 8-channels......
...read more
Apple M4 Ipad
(5)
Joe
8 months ago
Reply:
The benchmark that you are reading is for the CPU not the GPU.
...read more
Joe
10 months ago
Reply:
iGPU -> Radeon Graphics iGPU (Ryzen 7000)
...read more
Joe
10 months ago
Reply:
Can be easily calculated.
Assuming 64 bit bus per memory channel which 99% of modern desktop/laptop CPUs use.
0.008*NumberOfChannels*MaxMemorySpeed.
0.008*2*3200 = 51.2 GB/s.
...read more
Joe
10 months ago
Reply:
What you said is correct, for most people 5700X is the smarter choice, but for people who are chasing more performance, the 5800X can be overclocked slightly higher.
...read more
Joe
10 months ago
Reply:
So you're saying that Qualcomm themselves don't know what their GPU supports?
...read more
Joe
10 months ago
Reply:
Its Qualcomm themselves.
https://www.qualcomm.com/smartphones/products/8-series/snapdragon-8-gen-5-mobile-platform
...read more
Joe
10 months ago
Reply:
Still not fixed, even though the die shot clearly shows 8 channels.....
...read more
Apple M4 Ipad
(5)
Joe
10 months ago
Reply:
L3 cache is still wrong....
...read more
Joe
11 months ago
Reply:
Lol.
https://x.com/aschilling/status/1802764280078737836/photo/1
...read more
Apple M4 Ipad
(5)
Joe
11 months ago
Reply:
It was a typo on Wiki, plus Wiki is not a reliable source for these kind of specs, I literally spent months fixing specs there and there is still a lot more to do.
I fixed it there.
...read more
Apple A19 Pro
(30)
Joe
1 year ago
Reply:
GPU name -> Apple M5 GPU
Architecture -> Apple GPU
Pipelines -> 10
Shading units -> 128
Total shaders -> 1280
...read more
Apple M5 Ipad
(6)
Joe
1 year ago
Reply:
+60% GPU performance in 2 years isn't good enough for you?
...read more
Joe
1 year ago
Reply:
For starter, GPU cache increased from a total of 12MB to 18MB, not to mention probably architectural improvements here and there as well.
...read more
Joe
1 year ago
Reply:
ARM design relies on small dedicated L2 cache and big L3 cache whereas the Snapdragon has big shared L2 cache and no L3 cache.
...read more
Joe
1 year ago
Reply:
3D Mark Solar Bay Extreme is basically a Ray Tracing test, a useless feature for a smartphone SoC.
...read more
Joe
1 year ago
Reply:
You didn't fix the L2 & L3 cache.
...read more
Joe
1 year ago
Reply:
Geekerwan review showed the 8 Elite 5 GPU is slightly more efficient and tops out at 12W versus 15W for the D9500 GPU.
...read more
Joe
1 year ago
Reply:
I went back and double checked, it wasn't a typo.
ARM GPU datasheet v7 said that G57 supports OpenCL 3.0.
ARM GPU datasheet v8 (latest) says that G57 supports OpenCL 2.1.
...read more
Joe
1 year ago
Reply:
Sorry for the typo, it supports 2.1 as per the ARM GPU datasheet.
G31/G51/G71/G72 -> 2.0
G52/G57/G76/G77 -> 2.1
G78/G310/G510/G710/G715/G720/G725/G1 -> 3.0
...read more
Joe
1 year ago
Reply:
Download and upload speeds are correct, its the model number for the Modems that are wrong.
Exynos 2400 uses Shannon 5153 modem.
Exynos 2500 uses Shannon 5163 modem.
Exynos modem lineup like...
...read more
Joe
1 year ago
Reply:
Instruction set -> ARMv9.0-A
...read more
Joe
1 year ago
Reply:
8sG3 is overclocked 8G2 -> False
8sG3 is a downclocked 8G3 -> False
...read more
Joe
1 year ago
Reply:
7+ Gen2 is a heavily downclocked 8+ Gen 1, they downclocked the CPU and GPU in order to keep the performance lower than their previous flagship.
If you compare the 8+G1 to the 8sG3 and keep it...
...read more
Joe
1 year ago
Reply:
Instruction set -> ARMv9.3-A
GPU Architecture -> 5th gen
...read more
Joe
1 year ago
Reply:
GPU frequency 1750 -> 1716 MHz
FLOPS 5376 -> 5271.552 Gigaflops
...read more
Joe
1 year ago
Reply:
LPDDR6 specs were published by JEDEC last July, here is the important info:-
Speed 10667-14400 MT/s.
Density (per chip) 512MB-8GB.
It should start hitting the market in Mid 2026.
...read more
Joe
1 year ago
Reply:
GPU frequency 1000 -> 1100 MHz
FLOPS 1536 -> 1689.6 Gigaflops
...read more
Google Tensor G5
(31)
Joe
1 year ago
Reply:
NP, you also need to change the Gbit to GB for all SoCs, its Gigabyte not Gigabit for memory bandwidth.
...read more
Samsung Exynos 2500
(33)
Joe
1 year ago
Reply:
UFS is backward compatible, the E2400 supports UFS4.0, if the phone manufacturer (Samsung Mobile) decided to use UFS3.1 instead its their issue not the SoC itself.
...read more
Samsung Exynos 2400
(20)
Joe
1 year ago
Reply:
Depends.
Sometimes the information is publicly available, either disclosed by the Manufacturer or one of the mobile manufacturers.
Looking at die shots if someone made it...
...read more
Joe
1 year ago
Reply:
Pipelines -> 6
Shading units -> 128
Total shaders -> 768
...read more
Google Tensor G5
(31)
Joe
1 year ago
Reply:
Username checks out, just lol @ your comment.
...read more
Joe
1 year ago
Reply:
Pipelines 16 -> 8
Shading units 64 -> 128
To keep both consistent across all Samsung RDNA GPUs.
...read more
Samsung Exynos 2500
(33)
Joe
1 year ago
Reply:
TFLOPs are a theoretical value for floating performance and are not really comparable across different architecture.
There are many other parts of the GPU that contributes to overall...
...read more
Joe
1 year ago
Reply:
Snapdragon 4 Gen 2 is made using Samsung 4LPX process which is a rebranded Samsung 5nm process, its 5nm not 4nm.
...read more
Joe
1 year ago
Reply:
1) Adreno 722 is 1/3 the shaders of a flagship which means its relatively a small GPU.
2) Snapdragon 888 is a 5-year old SoC, and the only reason its faster is due to the higher clocks thanks to a...
...read more
Joe
1 year ago
Reply:
Don't compare apples to oranges. Flip phones don't have much thermal headroom due to design and throttle. The 8G3 in the Flip6 scored ~2000/6000 on average.
...read more
Samsung Exynos 2500
(33)
Joe
1 year ago
Reply:
You're the one who should learn CPU architecture before commenting.
Adding more cores improves multicore efficiency at the cost of die space (cost).
...read more
Samsung Exynos 2500
(33)
Joe
1 year ago
Reply:
And now it supports 256GB.
https://www.msi.com/news/detail/MSI-Release-the-Latest-AMD-AGESA-Combo-PI-1-2-0-3e-BIOS--Supporting-all-64GBx4-DRAM-Chips-and-New-CPU-146587
...read more
Joe
1 year ago
Reply:
Manufacturers usually don't update their websites post launch, I already gave you an example of the AMD Ryzen 9 7950X supporting 192GB RAM via BIOS update (Support came via AGESA update by AMD...
...read more
Xiaomi Xring O1
(18)