Qualcomm has Snapdragon 8cx beating Core i5-8250U in PCMark 10

[soresu]

I think It Is a matter of price than anything else.

For example im looking for an arm board with better gpu perf than the rpi 4. They are just too expensive. To the point that im better off getting a 2200g combo.

Enviado desde mi Redmi 6A mediante Tapatalk

Probably the HiKey 970 board (HiSilicon Kirin 970 SoC) is the best you can get without some connections to Qualcomm for their latest dev board/platform.

As you say though, they tend to be pricey - even with the age of the Kirin 970 SoC.

If you want higher GPU more than CPU, there is a smaller Jetson Nano board I think.

 

[DrMrLordX]

I think It Is a matter of price than anything else.

For example im looking for an arm board with better gpu perf than the rpi 4. They are just too expensive. To the point that im better off getting a 2200g combo.

Rockchip is releasing an A76-based SoC in 2020 that will likely go into a Pine product. That's what you want. Or you can break down and get an SQ1-based Win10 machine. Or just get an Android phone with an 855+ or 865 in it and start using stuff like UserLAnd (assuming what you want is Linux).

 

[DrMrLordX]

@Thala

I tried the "freezer trick" by running 3DPMRedux while the phone was in my freezer. It definitely caused the battery to under-report its charge for awhile until it warmed up, but otherwise there were no problems. The chip was definitely thermally-limited in longer runs of 3DPMRedux: without the freezer, peak performance in selection 5 was around 184M/sec, but in the freezer, it was able to hit over 199 M/sec. It still seems as though the SQ1 is a fundamentally faster CPU, unless the Android subsystem is holding back my 855+ somehow. According to the CPU monitoring app I installed, the CPU stays busy doing . . . something even when the phone is not doing much of anything, so it is entirely possible that the CPU has more power on tap that it can't commit to a single application thanks to Android.

 

[wintercharm]

@soresu

Woops I had forgotten about the distinction between SVE and SVE2. So you don't predict consumer support for SVE2 to come until ARM v9? A bit unfortunate, but I guess that's understandable.

I wonder if we'll see SVE2 deployed in ARM server cores before consumer devices?

IIRC Apple's A13 already has some sort of custom SVE-like implementation of advanced vector instructions called AMX. The specifics haven't been teased out yet, but:

Apple puts their performance at up to 1 Tera Operations (TOPs) of throughput, claiming an up-to 6x increase over the regular vector pipelines. This AMX instruction set is seemingly a superset of the ARM ISA that is running on the CPU cores.

Source: Anandtech's iPhone 11 review.

 

[soresu]

IIRC Apple's A13 already has some sort of custom SVE-like implementation of advanced vector instructions called AMX. The specifics haven't been teased out yet, but:

From what I understand, AMX is concentrated on ML/AI code acceleration.

 

[soresu]

IIRC Apple's A13 already has some sort of custom SVE-like implementation of advanced vector instructions called AMX. The specifics haven't been teased out yet, but:

Source: Anandtech's iPhone 11 review.

"The Lightning cores feature machine learning accelerators called AMX blocks. Apple claims the AMX blocks are six times faster at matrix multiplication than the Apple A12's Vortex cores. The AMX blocks are capable of up to one trillion 8-bit operations per second "

From the A13 Wiki page, sourced from Anandtech's own article. Link.

My guess is that Apple were frustrated by lacking ML focused features in the ARM v8-A ISA which are now being addressed in v8.6-A and Matterhorn.