Have I gone totally insane yet?

[Ampera]

WARNING: THIS IS AN IMAGE, AND IT'S LARGE. DO NOT CLICK IF YOU ARE UPSET OVER SPILLED BANDWIDTH.

http://prntscr.com/e23w5v

Windows Advanced Server 2000 on an i7-4790k (Using a hypervisor) and 8GB of RAM (I am most entertained by 8GB of RAM on a 32-bit system)

Speaking of which, I don't get why people were so upset over the supposed 32-bit 4GB limit with Intel processors, especially since Intel had 64GB addressing since the Pentium Pro.
I find it weird, why was 64-bit so amazing?

The cool thing about this setup is that -2gb of RAM and 4 threads, I could do this on real almost period correct 32 bit server hardware.

[Halcyon]

WARNING: THIS IS AN IMAGE, AND IT'S LARGE. DO NOT CLICK IF YOU ARE UPSET OVER SPILLED BANDWIDTH.

http://prntscr.com/e23w5v

Windows Advanced Server 2000 on an i7-4790k (Using a hypervisor) and 8GB of RAM (I am most entertained by 8GB of RAM on a 32-bit system)

Speaking of which, I don't get why people were so upset over the supposed 32-bit 4GB limit with Intel processors, especially since Intel had 64GB addressing since the Pentium Pro.
I find it weird, why was 64-bit so amazing?

The cool thing about this setup is that -2gb of RAM and 4 threads, I could do this on real almost period correct 32 bit server hardware.

That's not unusual. It's known as PAE (Physical Address Extension) which was supported from the time of the Pentium Pro. It allowed 32-bit machines to "break" the 4GB RAM limit and utilise page table entries 64-bits wide. These days, the modern Windows OS's mandate PAE support.

From Windows 2000 you could use more memory than "possible" with 32-bit hardware. See Memory limits on 32-bit editions of Microsoft Windows, with PAE support. It wasn't magic, it just allowed ways of overcoming the 32-bit memory barrier. Back in the day, many PC users also used methods to overcome CHS (Cylinder-Head-Sector) limits relating to hard disks.

Was it ideal? No, it was largely a band-aid fix but it (mostly) worked.

Enjoy Windows 2000 though, it'll scream along and is solid as a rock. I still run it on a machine at home tucked away in the server rack. It just does some legacy stuff (including hosting a TFTP server for my Commodore 64).

[rollatorwieltje]

Speaking of which, I don't get why people were so upset over the supposed 32-bit 4GB limit with Intel processors, especially since Intel had 64GB addressing since the Pentium Pro.
I find it weird, why was 64-bit so amazing?

A 64 bit OS allows a single application to use more than 2 or 3GB of memory. A 32 bit OS can use more than 4GB easily, but it gives an application 32 bit memory pointers, and part of that range is already claimed. Afaik the limit on a 64 bit OS is typically in the order of several terrabytes.

[T3sl4co1l]

IIRC, the problem was, both drivers and applications were rarely made to support PAE, so it went almost entirely unused, outside of servers (where the hardware and driver quality is generally better, and so was worth enabling).

Even so, a flat 4 gigs is a flat 4 gigs, so to use that much in an application, you have to resort to the ancient Intel tradition of segmented memory (or since we're talking protected mode, technically, "selector"), which so it seems, no one has ever liked, at any time, in history.

Which is funny because they had the identical problem in the late 70s (migrating from 8 to 16 bit CPUs, like 8080 and Z80 with flat 64k addressing, to 8086's 1MB), and this was their solution (segmentation).

Tim

[Monkeh]

Speaking of which, I don't get why people were so upset over the supposed 32-bit 4GB limit with Intel processors, especially since Intel had 64GB addressing since the Pentium Pro.
I find it weird, why was 64-bit so amazing?

Because 64GiB is small, PAE is slow and hacky, and it does not grant any single piece of code access to more than a 32-bit address space.

[HwAoRrDk]

Am I remembering incorrectly, or didn't NT 4 Server's 'enterprise' (or 'datacentre' or whatever it was called) version also support PAE? That is, it wasn't something introduced with Windows 2000.

I recall there were special versions of SQL Server and Exchange for if you wanted to take advantage of that large memory support.

[TiN]

Try to edit/process any 4K footage or 40mpixel image (common from today DSLR's) on any 32bit environment and then same on 64bit OS+apps. Experience is....eye-opening  .

[rrinker]

 I think Server 2000 was firs to have the /PAE switch in boot.ini.
It wasn't so much a special version of SQL as it was just making sure the /PAE switch was enabled in teh OS and then a similar setting was enabled in SQL. The base version I don't think supported this, but the higher version was for more features than just the extra memory - stuff like clustering.

 

[David Hess]

IIRC, the problem was, both drivers and applications were rarely made to support PAE, so it went almost entirely unused, outside of servers (where the hardware and driver quality is generally better, and so was worth enabling).

I think Microsoft was more of a problem here.  The Linux and BSD guys figured it out.

Some hardware could only handle 32 bit addresses and drivers were a problem but the later could have been fixed and both problems could have been solved with a minor performance hit using double buffering but Microsoft instead used operating system PAE support of more then 32 bits of memory for market segmentation.  I do not remember about non-server 2000 but early desktop XP could address more than 32 bits of address space.  The only thing which prevented this was a a lack of systems which supported enough memory.  Then service pack 1 or 2 implemented a hard limit when that became a real possibility; patches became available to get around it.

Even so, a flat 4 gigs is a flat 4 gigs, so to use that much in an application, you have to resort to the ancient Intel tradition of segmented memory (or since we're talking protected mode, technically, "selector"), which so it seems, no one has ever liked, at any time, in history.

Even if the applications did not take advantage of PAE to access more than 2 or 3 GB depending on the configuration, Windows could use PAE to provide each application its full 2 or 3 GB slice of memory.  There is no substitute for a flat address space though when performance and simplicity matter.

Which is funny because they had the identical problem in the late 70s (migrating from 8 to 16 bit CPUs, like 8080 and Z80 with flat 64k addressing, to 8086's 1MB), and this was their solution (segmentation).

The Apple ][ and later Apple systems and some other 6502 systems also used bank switching to access more than 64k; the limits of a 16 bit address space was recognized early as a problem when memory mapped I/O and possibly multiple optional ROMs had to be included.  Later CP/M systems did as well with large banks of RAM although I never used one; the biggest CP/M system I had experience with had 63k of available memory with a boot ROM taking the other 1k.

Implementing virtual memory on the broken 68000 is a whole discussion in itself but at least one company managed to do it with heroic measures.