For the price of an Atom you can get the best quad ARM they're selling :P
Originally Posted by RonJohn
Hopefully we can see some new 64 bit ARM chips in somewhat higher clock speeds (2GHz or so) in a quad-core package. Such a chip can easily beat the x86 arch in speed and be much more power efficient so these can be made into small formfactor desktop PC's or even media centers. With low power we may not even need much in the way of cooling.
I for one would LOVE an ARM desktop and look forward to seeing this "old" architecture become the new wave of the future!
What I would love to see is how many cores and at what speed an ARM processor could pull off within the same power envelope as an x86 chip.
Originally Posted by DeepDayze
I would like to see something like this too
Originally Posted by deanjo
Here is a very rough comparison of Core Count/GHz.
An Intel Core i7-980X Extreme Edition Gulftown (32 nm) 3.333 GHz @ 6 Cores 130 W TDP
I could not find much info on TDP for a 1.5GHz Dual Core ARM CPU (Qualcomm SCorpion was my example) however I found a few sites stating ARM CPU's rarely go over 2W at fullload, so I will take a maximum of 4 W for 1.5Ghz Dual Core (This includes the GPU as well :P).
For Intel CPU, Watt/GHz = 130 / (6 * 3.333) = 6.5 W/GHz = 0.154 GHz/W
For ARM CPU, Max Watt/GHz = 4 / (1.5 * 2) = 1.3 W/GHz = 0.769 GHz/W
ARM = ~ 5x GHz/W of Intel x86.
So for 130 Watts, Intel has total: 20 GHz, ARM has 99.97 GHz
However, nVidias new Quad Core ARM SoC (4 Core) will preform roughly the same as a Intel CoreDuo (2 Core), both in the region of 2-2.5 GHz.
However, the Core Duo architecture was improved on a fair bit in terms of Performance/GHz, and then the first i3/i5/i7 were a a good step up from that, and then Sandy Bridge, and now Ivy Bridge soon are steps above that again in terms of performance/Ghz . So a Quad Core ARM is level to a Core Duo, meaning it could not get anywhere near a new Intel Series chip, and the question is whether at the same TDP, it could outpreform it or not?
GHz per Watt doesn't make much sense.
Remember AMD's quantispeed arch/rating?
And here you are comparing RISC vs. CISC design. This is too much difference for a GHz per W comparison.
Well it does, because I was showing (roughly) the comparison in terms of GHz available to each architecture by TDP, but I also mentioned the differences in the architecture performances; The Quad Core ARM comparing to and old Core Duo Intel in performance (Like 3 architectures old if you do CoreDuo -> Core 2 Duo -> First i series (Nehalem etc...) -> Sandy Bridge. Anyhow someone had asked about it, so I gave one answer.
Originally Posted by Adarion
First of all ARM is as strong as Atom: Atom=2.4dmips/mhz ArmA8=2dmips/mhz ArmA9=2.5dmips/mhz. All previous CPUs have half the procPower of a SandyBridge/mhz. Arm28nm goes up to 3GHz and has Neon acceleration for GPUs (1core=48gflops/GHz). Second, Arm has A7core which is 5 times smaller than A9 and 20 times smaller than Atom, this will lead in 20-40core CPUs till end 2012, plus Arm has V8-64bit. Third, Cisc does not mean complex instructions and Risc does not mean simple ones. The first letter C or R refers to the relations of the instructions. Cisc=Complex relations wile Risc=family relations. This means that Risc can run many instruction by one unit wile Cisc can't. Risc run different instructions by one unit wile Cisc needs different units: float to fpu and integer to alu and else to else. Risc run them with the same way (like FMAC) wile Cisc have different ways. Cisc will never be Risc and Cisc does not decode anything in Risc mode, that's a common error. Risc will always be 40 times better in performance per watt and per dollar. http://en.wikipedia.org/wiki/Loongson This CPU in L3C part has 1instructionTflop or 12streamTflops per 20W plus emulation and 3D instructions.
They're not going to beat any modern amd64 based chip for some time. A 2ghz ARM is nowhere near comparable to a SandyBridge or even a Bulldozer in terms of raw power. Aside from just the difference in interconnects (say, built in triple channel DDR3-1866 and PCI-X 2.1), the branch prediction, SIMD units, and so on are just not there in the ARM chips. Gigahertz just don't matter that much (try comparing a 2.2ghz i7 to a 3.4ghz Bulldozer, for instance).
Originally Posted by DeepDayze
Power consumption will be lower, and performance per Watt may even still be way better, but raw performance will not be.
Tags for this Thread