PowerVR

dont want to disappoint you ailuros but I had a cheap chinese tablet (Onda V972) running Android 4.1 with Allwinner A31 quad core A7 at 1.2GHz, so very similar to the 6589, and the CPU is very very slow. My Nexus 7 is way faster in every task. After one week, I just stopped to use it...
This A7 core is a big performance regression. if you have to choose, go for A9, especially the rockchip rk3188 at 1.8Ghz on 28nm process really flies.
 
dont want to disappoint you ailuros but I had a cheap chinese tablet (Onda V972) running Android 4.1 with Allwinner A31 quad core A7 at 1.2GHz, so very similar to the 6589, and the CPU is very very slow. My Nexus 7 is way faster in every task. After one week, I just stopped to use it...
This A7 core is a big performance regression. if you have to choose, go for A9, especially the rockchip rk3188 at 1.8Ghz on 28nm process really flies.

Point taken. Note that I'm not looking for a tablet for one and we're talking about a price around half of the initial MRSP of the Nexus7. A friend was lucky and caught a white box GT I9300 smartphone which even has the startscreen of a Galaxy S3 at around 1/4th the street price of the real thing; that's what I was looking for but it's sold out and while you can still get similar offerings the price is times higher which defies the purpose.
 
dont want to disappoint you ailuros but I had a cheap chinese tablet (Onda V972) running Android 4.1 with Allwinner A31 quad core A7 at 1.2GHz, so very similar to the 6589, and the CPU is very very slow. My Nexus 7 is way faster in every task. After one week, I just stopped to use it...
This A7 core is a big performance regression. if you have to choose, go for A9, especially the rockchip rk3188 at 1.8Ghz on 28nm process really flies.
Absolute nonsense.

If I lock my S4 to only the A7 cores, everything is still perfectly smooth and running very fast. The A7's are by no way such a regression, I see no difference to a similarly clocked A9.
 
Absolute nonsense.

If I lock my S4 to only the A7 cores, everything is still perfectly smooth and running very fast. The A7's are by no way such a regression, I see no difference to a similarly clocked A9.

Well even if he'd have a point, the particular price/performance ratio is still hard to beat I am talking about.
 
Absolute nonsense.

If I lock my S4 to only the A7 cores, everything is still perfectly smooth and running very fast. The A7's are by no way such a regression, I see no difference to a similarly clocked A9.
maybe non sens to you but I stand still.
It's my hands on experience. ARM specifies A7 as 1.9DMIPS/MHz/core and A9 is 2.5DMIPS/Mhz/core
If you read this anand article:
http://www.anandtech.com/show/4991/...dualcore-more-power-efficient-highend-devices
A7 is a much inferior core compared to A9 ( no OoO and partial dual issue) so technically you shouldn't be surprised to see A7 at 1Ghz ~50% slower than A9 at 1.3Ghz. And when you use the 2 devices side by side, it's obvious, the A9 feels so much snappy in every task...
 
He said specifically same frequencies between the cores; besides you'd probably should also take under consideration what is exactly going on on the sw side of things. A Google reference device is obviously in a completely different ballpark than a chinese OEM white box budget item.

.......so technically you shouldn't be surprised to see A7 at 1Ghz ~50% slower than A9 at 1.3Ghz

Even if you compare a 1GHz A9 against a 1.3GHz A9 the difference is already at 30% :LOL: I didn't say 1 but 1.2GHz, so the difference is way smaller than 50% in realtime, while the price difference is times larger than that. Price/performance ratio, nothing more nothng less.
 
It's my hands on experience.
And mine isn't? I have both a Quad A9 device and a Quad A7/A15 device. I see no difference in performance between similarly clocked A9's and A7's. ARM themselves state 90% of A9 performance for the A7's.

Don't confuse crappy software with the actual hardware's performance and capability.
 
Results today also confirmed series 6XT is announcing later this year. Some key points are it focuses on further power improvements and improvements related to GPGPU. Sounds more that just glueing existing announced "cores" together.

IMG also confirmed that Samsung supports opencl in the exynos5 equipped galaxy S4. Don't know it is exposed or whether the inference was it was being used by the os for some video/camera computation.
 
Series6XT stands for multi-core? So far we "know" that each core can scale up to 8 clusters and above that it goes to MP (that's at least how I've understood it so far).

Else crystall ball speculation: they'll announce the first 8 cluster Rogue variant later this year (G6830?) which can also theoretically scale to multi core for 2015 and beyond.

Trick question: with a Hydra thingy or without? ;)

***edit: by the way it it just me or are the MIPS margins probably crapalisiously low? If IMG manages in the longrun to get their CPU IP to somewhat higher penetration rates, I can see pretty much anyone to benefit from it except ARM.
 
***edit: by the way it it just me or are the MIPS margins probably crapalisiously low? If IMG manages in the longrun to get their CPU IP to somewhat higher penetration rates, I can see pretty much anyone to benefit from it except ARM.

Last quarter average royalty rate for MIPS was 5c. That's been coming down signficantly over the last 18 months. Still higher than ARM.
 
As oppose to those very expensive 8 bit pics ?

No, I'm just pointing out that it doesn't necessarily mean much about the licensing fee for, say, an Exynos Octa. This was a comment about ARM itself, not in relation to MIPS.
 
No, I'm just pointing out that it doesn't necessarily mean much about the licensing fee for, say, an Exynos Octa. This was a comment about ARM itself, not in relation to MIPS.

Its an average, by definition it doesn't tell you ANYTHING about high end royalties, nor low end royalties.

ARM is use in tiny controllers, and high-end multi-core stuff

MIPS is used in tiny 8-bit controllers right up to Caviums newly announced 48-core MIPS64 bit networking controller.
 
Its an average, by definition it doesn't tell you ANYTHING about high end royalties, nor low end royalties.

ARM is use in tiny controllers, and high-end multi-core stuff

MIPS is used in tiny 8-bit controllers right up to Caviums newly announced 48-core MIPS64 bit networking controller.

How can be MIPS be used in 8-bit controllers when they only offer 32-bit and 64-bit cores?

http://www.mips.com/products/product-materials/processor/processor-cores/

MIPS average royalty cost is higher than ARM's precisely because they don't go as small and cheap in practice; ARM7 and Cortex-Mx are viable alternatives to 8-bit MCUs in a lot of places. If you look at MIPS on the other hand, its place in the MCU world is limited to stuff like PIC32 where it's positioned as a tier above the PIC24s a Cortex-M4 will tend to compete with or the PIC16/18s an ARM7 or Cortex-M0 will sometimes sort of compete with.
 
How can be MIPS be used in 8-bit controllers when they only offer 32-bit and 64-bit cores?

http://www.mips.com/products/product-materials/processor/processor-cores/

MIPS average royalty cost is higher than ARM's precisely because they don't go as small and cheap in practice; ARM7 and Cortex-Mx are viable alternatives to 8-bit MCUs in a lot of places. If you look at MIPS on the other hand, its place in the MCU world is limited to stuff like PIC32 where it's positioned as a tier above the PIC24s a Cortex-M4 will tend to compete with or the PIC16/18s an ARM7 or Cortex-M0 will sometimes sort of compete with.

Yeah bad by me. I've never programmed a PIC, recently seen a lot of references to microchip on IMG PR stuff and assumed they had the whole range. I didn't realise MIPS was 32 or 64bit only. Given that, the fact that ARM's ASP is not far of MIPS's is quite surprising, given how low ARM can go in the performance, and hence price, envelope.
 
Last edited by a moderator:
Back
Top