One year after Apple's A7, Nvidia announces first 64-bit ARM CPU for Android
Chipmaker Nvidia on Tuesday detailed the new "Denver" variant of its Tegra K1 mobile processor, a high-performance in-order design that represents the first foray into 64-bit processing for Android devices --?nearly a full year after Apple shocked the mobile world with its own 64-bit A7 processor.
Denver combines Nvidia's popular Kepler GPU with a new 64-bit dual-core CPU, the company's first in-house CPU design. Like Apple's A7, Denver is compatible with the ARMv8 architecture.
The chip also packs a 4-way, 128-kilobyte L1 instruction cache alongside a 4-way, 64-kilobyte L1 data cache and a 16-way, 2-megabyte L2 cache. Nvidia says each Denver core is capable of processing up to seven operations per clock cycle, compared with a reported six instructions per clock for the A7 and just three per clock for the 32-bit Tegra K1.
With Denver, Nvidia is touting a new architectural addition known as Dynamic Code Optimization. Using DCO, the CPU will "translate" oft-used ARM code into microcode and cache those instructions in a dedicated 128-megabyte translation buffer -- the company said this can "effectively double" the performance of the base silicon.
There is no word on which device will be the first to ship with Denver, though Nvidia does promise full pin compatibility with 32-bit Tegra K1 variants for easier integration. The first 64-bit version of Android is currently in testing and is slated for release this fall.
Chipmakers have been scrambling to catch up with Apple's 64-bit A7, which the company unveiled alongside the iPhone 5s in September of last year. The chip's surprise introduction was said to have left industry insiders "slack-jawed, and stunned, and unprepared."
"Apple kicked everybody in the balls with this," a Qualcomm employee said at the time. "It's being downplayed, but it set off panic in the industry."
Denver combines Nvidia's popular Kepler GPU with a new 64-bit dual-core CPU, the company's first in-house CPU design. Like Apple's A7, Denver is compatible with the ARMv8 architecture.
The chip also packs a 4-way, 128-kilobyte L1 instruction cache alongside a 4-way, 64-kilobyte L1 data cache and a 16-way, 2-megabyte L2 cache. Nvidia says each Denver core is capable of processing up to seven operations per clock cycle, compared with a reported six instructions per clock for the A7 and just three per clock for the 32-bit Tegra K1.
With Denver, Nvidia is touting a new architectural addition known as Dynamic Code Optimization. Using DCO, the CPU will "translate" oft-used ARM code into microcode and cache those instructions in a dedicated 128-megabyte translation buffer -- the company said this can "effectively double" the performance of the base silicon.
There is no word on which device will be the first to ship with Denver, though Nvidia does promise full pin compatibility with 32-bit Tegra K1 variants for easier integration. The first 64-bit version of Android is currently in testing and is slated for release this fall.
Chipmakers have been scrambling to catch up with Apple's 64-bit A7, which the company unveiled alongside the iPhone 5s in September of last year. The chip's surprise introduction was said to have left industry insiders "slack-jawed, and stunned, and unprepared."
"Apple kicked everybody in the balls with this," a Qualcomm employee said at the time. "It's being downplayed, but it set off panic in the industry."
Comments
And Nvidia's 64-bit chip will be a year behind Apple's A8 processor.
... All two of you must be very excited at this announcement.
It is interesting, A7 releases -same week-"it's just a gimmick"-the next week-"we will all rush in 64 bit chips in 2014"-now half butted 64 android, Samsung devices are all going to come.
Here's the big problem for Nvidia:
They built a powerful processor that isn't suitable for phones (tablets only due to power consumption). Apple (1st) and Samsung (2nd) have the high-end tablet market locked down. That leaves a very small market of potential customers that could actually use this processor.
Nvidia should have been thinking of who they could sell a processor to instead of making a "beat" (we'll have to wait and see on this one) that has a very limited market potential.
What do meteorologists have to do with it?
Looking forward to Anandtech's review of the A8 once it's out in the world and how it compares to this Nvidia chip (once it's actually in a shipping product that is). We can be pretty sure the A8 will be out in September. Will the Nvidia even be shipping by then (let alone in an actual shipping product)? What about 64-bit Android and the developer tools? Apple has a nice lead and their owning the whole platform (hardware, software, dev tools) is a big advantage. Although some complain about it, their willingness to drop support for older products also allows them to move forward more quickly.
Here's the big problem for Nvidia:
They built a powerful processor that isn't suitable for phones (tablets only due to power consumption). Apple (1st) and Samsung (2nd) have the high-end tablet market locked down. That leaves a very small market of potential customers that could actually use this processor.
Nvidia should have been thinking of who they could sell a processor to instead of making a "beat" (we'll have to wait and see on this one) that has a very limited market potential.
I don't think phones are the main goal for Nvidia's 64BIT ARM chip but their server division, they announced back in 2012 that they were working on a 64BIT ARM Opteron.
http://www.anandtech.com/show/6418/amd-will-build-64bit-arm-based-opteron-cpus-for-servers-production-in-2014
I don't think phones are the main goal for Nvidia's 64BIT ARM chip but their server division, they announced back in 2012 that they were working on a 64BIT ARM Opteron.
http://www.anandtech.com/show/6418/amd-will-build-64bit-arm-based-opteron-cpus-for-servers-production-in-2014
AMD announced they were working on it, not Nvidia??
I just don't understand the necessity to belittle everything that isn't Apple.
Just trying to nip the whole Nvidia is just copying thing in the butt.
Fast Iron... means nothing.
Quote:
So... doing the math on a dual core 7stage pipeline... You need a compiler/language/IDE that can parallelize and optimize 14
ARM instructions in parallel.
My guess... you'll see about 8 NOPs in that instruction stream.
Quote:
128Megabytes? 'effectively double? I'm trying to figure out if the entire ARM instruction set is 128MB?
Sounds like a great way to speed up benchmarks, but real world context switching code.... not so much.
Quote:
Backwards compatibility with 386 processing is what makes Intel chips so slow to evolve. Why in the world would you want to install a 32bit system onto a 64bit optimized chip?
Quote:
So this chip and this OS are basically being developed in vacuums and it will be up to the 'integrators' to make them work together. Who is invested in Performance? (Integration is a 'time to market' problem.)
I say all this in the vein that CPUs and OSes and Compilers have cycled back to the 70's and 80's when and have learned from their mistakes as well.
1) You don't design long term backwards compatibility into your chips. You design forward compatibility into your compilers.
2) Optimizing your pipeline is more an exercise in compiler technology
3) Designing OSes and SoC separately slow down performance innovations.
We are talking about Denver, a mobile chip. If they are just getting around to announcing it in mid-August do you really think it will ship in a week and be in mobile devices in 2 weeks? Stop reading into comments. There is no belittling, just pointing the common timeline for chip makers to get their products into a quantity of shipping products. This is the same for Intel, which is why we know about Broadwell and Skylake but won't see them released immediately and won't see them in shipping Macs a good time after they are officially released by Intel. Without a TARDIS this is how component sales work.
/sarcasm
AMD announced they were working on it, not Nvidia??
ooops, Nvidia as well announced their 64bit chips a while back http://www.forbes.com/sites/patrickmoorhead/2013/10/10/nvidias-mobile-custom-64-bit-arm-cpu-its-sooner-than-you-think/
I am still hoping to see more address space for app developers in A8 based iOS devices with 2GB of RAM. Currently we only get 31 bits of address space (2 GB) on 64 bit A7 devices due to a limitation of iOS. With more address space, apps can work with larger files more efficiently. Even though the RAM of the iPhone has increased from 128MB to 1GB, the address space has remained at 2GB since the original iPhone. If the address space remains at 2GB when the RAM increases to 2GB then effectively the iPhone will lose its virtual address space ability entirely.
sigh... I had to live with 64K (not 640) PSECTS and DSECT, and ISAM files, and processed millions of records a day on a sub MIPS system. you're working with flash drives and ghz cycles. I know we expect speediness, but it seems like a small set of problems at this time. Time to do some computer engineering;-)
Obligatory - "You were Lucky"
We are talking about Denver, a mobile chip. If they are just getting around to announcing it in mid-August do you really think it will ship in a week and be in mobile devices in 2 weeks? Stop reading into comments. There is no belittling, just pointing the common timeline for chip makers to get their products into a quantity of shipping products. This is the same for Intel, which is why we know about Broadwell and Skylake but won't see them released immediately and won't see them in shipping Macs a good time after they are officially released by Intel. Without a TARDIS this is how component sales work.
Yeah, I screwed up sorry, got my chips crossed. Denver was announced back in 2011, and 64bit to be released in 2014, so they are on schedule, they also released the K1 32bit extremely quick.