Slashdot Mirror

But, I would be reluctant to call a ChromeOS device as a "PC" because:
- They need to have a network connection to access user data
- Local file systems (ie USB drives) are absolutely painful to access and work with (the paradigm is to use GDrive storage and anything else is work)
- There simply isn't enough memory/drive space available for anything other Extensions which are measured in the low tens of MBytes
- Applications are limited to Javascript (although I'm hoping Webassembly will be an option in the near future) with browser built in debug tools with a somewhat convoluted load/test process. A full featured IDE for application development is nothing more than a dream at this point

My first PC (an Atari 800)

• Stored user data on a cassette tape drive (the floppy drive was too expensive for me to afford as a kid).
• You think USB drives are painful? How about having to play a tape for 5 minutes to load a program? And if your schoolmate is fooling around with the cord and accidentally unplugs it halfway through, you have to start over again.
• Had 32 kB of RAM.
• Applications were limited to BASIC and assembly. No IDE, no browser. You typed BASIC code line by line, or you switched it to assembly mode and typed in raw CPU instructions. Single-threaded I should add.

It worked fine for writing reports and playing games.

A PC is a personal computer. A computer you can use for personal computing tasks. As opposed to a mainframe which was shared by multiple people. Or a workstation which was used for tasks at work. Or an embedded system which was made to control a specific object. That's it.

A Chromebook is a PC. A tablet is a PC. A phone is a PC. Heck, a smartwatch is a PC if it's got a decent way for you to input data to it. The phone in my pocket is many times faster than the fastest Cray supercomputer from around the time I owned my first PC. 160 MFLOPS for the Cray vs 718 MFLOPS for a single thread on a Qualcomm Snapdragon 800. The definition of "PC" you've crafted means PCs didn't exist until the late-1990s.

ARM will never be able to compete with x86 in terms of computing power and x86 can't compete with ARM in terms of efficiency and low power.

Be careful with words like "never", I remember very well when ARM was running circles around 80x86 in terms of computing power: back in 1987, ARM's selling point was speed rather than low power.

AFAICT: The Wikipedia article you link to doesn't mention x86 processors at all...

I used to run a software PC Emulator on Archimedes(1) in 1987, ARM (around 4 to 8 Mips (2)) was at that time emulating 8086 at the speed of an IBM PC/XT or AT (both below 1 Mips (3)).

While calculating a screen-sized Mandelbrot fractal at the time took minutes (up to half an hour) on IBM PC, the Archimedes did it in seconds.

(1) The 80186 co-processor card mentioned at the end of the linked article was unfortunately never released, the emulation was 100% in software.
(2) Acorn Archimedes speed
(3) IBM PC/XT and AT speed

ARM will never be able to compete with x86 in terms of computing power and x86 can't compete with ARM in terms of efficiency and low power.

Be careful with words like "never", I remember very well when ARM was running circles around 80x86 in terms of computing power: back in 1987, ARM's selling point was speed rather than low power.

AFAICT: The Wikipedia article you link to doesn't mention x86 processors at all...

I used to run a software PC Emulator on Archimedes(1) in 1987, ARM (around 4 to 8 Mips (2)) was at that time emulating 8086 at the speed of an IBM PC/XT or AT (both below 1 Mips (3)).

While calculating a screen-sized Mandelbrot fractal at the time took minutes (up to half an hour) on IBM PC, the Archimedes did it in seconds.

(1) The 80186 co-processor card mentioned at the end of the linked article was unfortunately never released, the emulation was 100% in software.
(2) Acorn Archimedes speed
(3) IBM PC/XT and AT speed

Still basing the "quality" of the CPU on clock speed I see.

You should do some basic research before posting, it's well known that ARM11 doesn't compete well on a per-mhz basis with even older chips. Don't take my word for it tho:

Whetstone
Celeron 733mhz: 598 MWIPS, 185 MFLOP, 162 MFLOP, 116 MFLOP
ARM 11 700mhz: 270.5 MWIPS, 97.8 MFLOP, 100.8 MFLOP, 85.7 MFLOP

Dhrystone
Celeron 450mhz: 720 VAX MIPS
ARM 11 700mhz: 847 VAX MIPS

Still basing the "quality" of the CPU on clock speed I see.

You should do some basic research before posting, it's well known that ARM11 doesn't compete well on a per-mhz basis with even older chips. Don't take my word for it tho:

Whetstone
Celeron 733mhz: 598 MWIPS, 185 MFLOP, 162 MFLOP, 116 MFLOP
ARM 11 700mhz: 270.5 MWIPS, 97.8 MFLOP, 100.8 MFLOP, 85.7 MFLOP

Dhrystone
Celeron 450mhz: 720 VAX MIPS
ARM 11 700mhz: 847 VAX MIPS

OK, the idea behind ARM is that it is "fast enough" for desktop and notebook PCs. Well, if that's the case, then a P4 is also "fast enough" and you should consider not buying anything newer.

Why am I saying that? Let's look at one benchmark that *is* multi-core ready and that Nvidia kindly ran on the upcoming Kal-El quad-core systems: Linpack.

Now I know Linpack is not a perfect benchmark, but it does do a decent job of showing off number-crunching power and it is multi-core capable and there are results from a wide range of architectures.

Here's a result from a 1.7 Ghz P4 system (see: http://www.roylongbottom.org.uk/linpack%20results.htm)

CPU Mhz Opt (MFlops) Non-Opt (MFlops)
Pentium 4 1700 382.00 131.59

I think (but I'm not sure) that Opt means optimized (such as using SSE) and non-Opt is a minimal x86 implementation with no optimizations.

Now, here are Nvidia's results for its not-yet-on-the-market Kal-El Quad Core ARM at 1.0 Ghz:

Multi-threaded Linpack: 309 Mflops

See: http://www.xbitlabs.com/news/mobile/display/20110921142759_Nvidia_Unwraps_Performance_Benchmarks_of_Tegra_3_Kal_El.html

I'm going to assume that Nvidia will go out of its way to make sure the code is optimized for benchmarks that it posts as part of a marketing push.

So a QUAD CORE Arm architecture is still lagging behind a P4, and while the P4 has a clock speed advantage, it's a lot smaller than is justified by the difference in performance considering the Nvidia chip has 4 cores compared to a single-core P4.

Now, I'm not saying that Kal-El won't be awesome for use on tablets and smaller devices, but on a desktop or even a notebook, don't go around expecting miraculous performance.

It takes less than you think.

Let's approximate: there's about 200 line segments in those wireframes, projecting a 3d point on a plane takes (can't be bothered to remember the math) about 20 multiplications, so about 4000 flops per frame - so with 25 fps it's 100kflops.

And if we're talking not actual real-time rendering, but just previewing the model, then 8-12 fps would be enough, making it 30-50kflops.

And here's historic Whetstone results - check PDP-11/45, for example.

Pentiums? I'm not sure how reliable benchmarks are these days, but here's a table of dhrystones.

Pentium 75 achieves 87.1 VAX MIPS (Dhry2); Core 2 Duo 2.4 Ghz, 6248. 32 times the clockspeed, 70 times the performance. And well short of Moore's law.