Mips, Mops and Flops

[aphexcoil]

Does anyone know where I can get a nice chart or table showing Intel processors starting from the 8088 up to today's processors along with their rated MIPS / FLOPS, etc.?

I've checked Google but my Google-Fu skills weren't strong enough to really find anything that provided a nice chart.

I'm curious what the GFLOP ratings are for the current Sandy / Ivy Bridge processors and how they compare to, say, an 80286 processor.

I know there are other considerations like newer instruction sets that have come along throughout the ages, but I'm curious as to just how much faster a 2600k Sandy Bridge is compared to an 80286 running at 8 MHz.

Also, does anyone know which Intel Processor was the first to feature two cores?

Edit: Did any of you ever own a computer with an 8086 / 8088 processor back in the day? Let's make this a fun thread reminiscing on the old days! How about the old-school 8 inch floppy disks? How much information did they store? How about those cassette drives for computers like the TI-994a. Does anyone know how much data a cassette stored and what the effective baud rate was when they were transferring information?

 

[lsun22]

3.4GHZ/8MHZ. that's how much faster =P.

pentium d is the first chip with two cores.

 

[MrWizard6600]

more specifically, the first dual core Pentium D was the smithfield, which was two prescotts glued together.

and I too am disappointed with my google skills, "moore's law" seems like a good place to start but all any of those charts seem to want to give you is a raw transistor count.

 

[aphexcoil]

more specifically, the first dual core Pentium D was the smithfield, which was two prescotts glued together.

and I too am disappointed with my google skills, "moore's law" seems like a good place to start but all any of those charts seem to want to give you is a raw transistor count.

Exactly. There's a lot of good information but I just wanted a raw "apples to apples" comparison of MIPS. I know you can't really have an apples to apples comparison, but it would be nice to have an approximate comparison.

 

[aphexcoil]

more specifically, the first dual core Pentium D was the smithfield, which was two prescotts glued together.

Interesting. Didn't the older consumer chips support SMP? I remember having an ABIT motherboard once with two CPU slots that could support dual celeron processors.

It's actually kind of sad that Intel removed that support from the consumer level chips. I'm sure enabling SMP wouldn't be that hard. It would kick ass to have two Ivy Bridge's on one motherboard -- especially for video editing, Ray Tracing, etc.

Imagine having 8 real threads to work with! Or 16 threads with hyperthreading

 

[aphexcoil]

This is a fairly good article from Wikipedia:

http://en.wikipedia.org/wiki/List_of_Intel_microprocessors#Intel_4004

It doesn't have a nice chart for FLOPS, but it does mention the speed for a lot of the CPUs listed.

According to the article, the 80286 had the following:

Clock rates:
6 MHz with 0.9 MIPS
8 MHz, 10 MHz with 1.5 MIPS
12.5 MHz with 2.66 MIPS

Also, according to another article, the Ivy Bridge CPU weights in at 20,333 MIPS

That's about 20,000 times faster than the 80286 at 8 MHz. That's far faster than just looking at the clock-speed alone.

Here's something else that is noteworthy. In 1993, the fastest supercomputer had the following specs:

1993 Fujitsu Numerical Wind Tunnel 124.50 GFLOPS National Aerospace Laboratory, Tokyo, Japan

That's probably very close to a top of the line consumer CPU today. It looks like there is a 15-20 year separation between what a supercomputer's performance was compared to when a consumer level CPU will match it.

 

[BinarySynapse]

8086, -888, -186, -286, and -386 had no internal floating point capability. So their FLOP/s rating would be zero. Those systems had to be equipped with a 8087, 80287, or 80387 to do floating point operations. The 80486 (non-SX) was the first Intel processor to include a FPU on die.

As for "enabling" SMP on modern consumer processors, that's not going to happen. Core i(x) processors only have one point-to-point interface to connect to the platform. That interface is connected to the PCH. SMP capable Xeons processors have multiple interfaces that can connect to other processors in addition to the PCH that enables them to communicate with each other.

 

[aphexcoil]

8086, -888, -186, -286, and -386 had no internal floating point capability. So their FLOP/s rating would be zero.

They can still have FLOP ratings as floating point operations can be emulated in microcode, as an operating system function, or in user space code. It would just be significantly slower. They did have floating point capability, they just didn't have specific hardware for those functions.

 

[BinarySynapse]

They can still have FLOP ratings as floating point operations can be emulated in microcode, as an operating system function, or in user space code. It would just be significantly slower. They did have floating point capability, they just didn't have specific hardware for those functions.

That's true. I remember having to use a program called em87 to get some programs to work that required an fpu to be installed. I was really referring to the hardware itself. I guess I should have said "practically zero" since all floating point operations have to be converted to integer operations before executing.

 

[aphexcoil]

That's true. I remember having to use a program called em87 to get some programs to work that required an fpu to be installed. I was really referring to the hardware itself. I guess I should have said "practically zero" since all floating point operations have to be converted to integer operations before executing.

For all practical purposes, you're correct. It's unbelievably slow.

 

[MrWizard6600]

For all practical purposes, you're correct. It's unbelievably slow.

Doing FP in software is fairly common today --many (most?) arm chips dont have an FP unit. I managed to get Linux and the entire QT stack on a 500MHz arm chip that didn't have any floating point and still get perfectly acceptable fp performance out of it.