AMD on the right track as Linux in 64-bitness

In your letter, you conclude that AMD needs to take off its "budget" reputation. This, in my opinion, is a complete failure in the analysis.

The battle you are describing, has been fought before. It was fought between Intel and Motorola, between their respective x86 ane 680x0 architectures. Motorola having a vastly more powerful cpu, both in instruction set and performance. Lost the battle, primarily because it was focusing on the high-end market. Primarily concerned with that market, the user market was considered secondary and by the time they discovered that they're analysis was wrong. That the "cheap" and "budget" Intel cpu's were takin over the markets, because of their wide spread market share... they were too far behind. Even today, you can see the remnant of this between Apple and Microsoft.

AMD is on the right track, as is Linux ... Windows won it's user base on the fact, that it was "free" with every bought computer and still is. Intel is able to sell the cpu's at outrageous prices, because so many users believe that having the "true thing" is better than having a "copy", as this will ensure their software will be 100% backwards compatible, something that was greatly discussed during the K6 era and gave birth to concepts, like "Intel Inside".

By making sure that their CPU is priced right. AMD ensures that they get a market share, and focusing more on compatibility than Intel. They also ensure, that companies don't have to upgrade their software at the same time as they upgrade their hardware. An extremely important issue, for every major company that is running licenses for 100's or 1000's of employees. To scale this, imagine an average user having to face the issue of needing only to upgrade his core hardware (motherboard, cpu) or having to upgrade everything he has ... merely to get a state of the art machine. The former choice, ensures that more users will take that step, which in turn will create a market for new 64-bit software, which otherwise would only be awailable for the high-end market. In turn, ensuring that AMD may indeed have the chance of taking over the market, in precisely the same way as Intel did itself, so many years ago.

Regarding the following paragraph...

"Now let's get onto some better arguments for 64-bit. One is that more than 4GB of physical memory is needed to support many active, large processes and memory map many, large files - without paging the system to death. This is true, but it is not a good argument for 64-bit addressing. The technique that Intel and Microsoft call PAE (Physical Address Extension) allows 64 GB of physical memory but each process can address only 4GB. For most sites in 2003, 64GB is enough to be getting on with." As mentioned previously on your site. "Any claim that "4GB is enough" or that address windowing extensions are a viable solution are just plain nuts. Do people really think programmers will re- adopt early 1990's bank-swapping technology?

Where are the Opteron workstation motherboards? http://www.theinquirer.net/?article=10475

I believe that PAE was a temporary patch for x86 32bit to allow 36bit memory access. PAE is slow and awkward, AMD-64 is much more elegant, simple and faster. That is for address resolution. 64bit does not equal double 32bit performance (what a pity). Back in the days of the 286, 386, 486, 586 (Pentium) it was always hard to justify the next processor architecture was worth the move until several years down the track. I wonder how a 3GHz .13um version of a enhanced i486 would perform if it were possible. Over time, the Instruction, memory and operating modes where altered to extend capabilities and plan for the future. When 99.9% of OS and Apps were 16bit with 24bit mem, who wanted a 32bit CPU/OS/Apps (Server/Science used 64bit ALU, 64/80bit FPU). Now 32bit Int + 36bit mem will be not enough for the main stream market of tomorrow (3yrs).

I would like to see the clock speed increase considerably to be within 30% of Intel P4/5 based systems to really hit Intel where it hurts. With >80% of common benchmarks having AMD results being >20% above Intel's best.

AMD still lack on cache throughput and raw speed when running certain tasks. They lack the resources (deep pocketed cash reserves) to keep a technology advantage ahead of Intel. IBM may be able to help in this regard. I think many AMD systems are let done by other components such as motherboard, BIOS, drivers.

Catch 22 times 2.

AMD need greater market share to get more developers to optimise for it. AMD needs more developers to optimise for it to gain greater market share. AMD make more profit to then spend on R&D and capital expenditure to remain competitive. AMD needs to remain competitive to sustain profit.

AMD and the Opteron platform have lots of potential. They need to keep the channel partners, retailers, corporate, government and retail users happy with consistently good reliability, regularity, product, support, performance, all at a good price.

I'd love to see a 90nm Opteron with high bandwidth cache shared with double the core which can behave as single CPU, dual CPU or multithread across both. More SEE2 execution and more MHz would be nice. I'm using my old AMD Duron 850MHz for Folding@Home (distributed computing project) and I'm looking for my next upgrade.

That's enough for now. Thanks for reading. With technology, I seem to waste more time then I save.

Tim Chambers

Questions, would 64bit computing not help increase performance in the Multimedia environment... 3D Gaming type stuff, real time high polygon rendering(Solidworks, 3DSMAX...), graphics (photoshop, paintshop...) and Audio/Video compression (MPEG, MP3...). To my knowledge I would say that the gaming industry is increasing considerably and being that this industry is growing shouldn't 64bit computing be in more demand.

Being that many people rip audio and video, wouldn't having 64bit computing help in compression processes. I know it won't do anything for word precessing since that hasn't been a challenge for PC's since the PII/AMDK6-2. I don't like waiting for several hours to rip a DVD and compress it to MPEG 4... it renders my only PC useless for that time. How about people that use graphics applications like photoshop, should it not help in the application of filters?

Anyway I do agree that for the average person who uses there PC today there is no need for 64bit, but since I've been in the industry of Tech support for many years I've noticed a trend that happens to all Computer users. This trend is that people usually start off with low end PC then when they find there niche they start to build up on that, I don't believe that the industry should force people to buy %10 to %20 increments of performance between generations when they could buy %50+ performance increases... and I do mean in the areas where you will actually notice performance... word processors work in real time for users but audio/video compression (which is what most users do that is demanding from there system) is not, when it takes longer to convert an audio file or video file than the files play time then that is too slow. BTW I'm talking about high quality settings here... no degradation in quality.

BTW one last trend I've noticed is that most people try some things a few times but when it takes too long they give up so that means that most people give up on trying to do something high level with there PC's because it takes too long and they usually never try it again even if they get a newer faster PC. My self when I get a new PC I always try to do things that I did before and see if I notice a difference, most people are not that patient.

Are you saying that 64bit on the desktop isn't need on the desktop until 2010 like intel said? Are you saying that no home user needs the speed and power of 64bit processors? After I read this, I think I should have stayed with a commodore64 (6502 processor) or a osbourne (cpm and Z80 processor). I must have been silly to upgrade all these years. I truly don't understand why I don't need 64bit right now. I do some serious graphics(simulation and ray tracing) and would love to have the 64bits in a PC. Am I wrong to want this?

bobby hill

Sorry to multiple-mail you, but I just read Nick Maclaren's piece on 64-bit computing and wanted to thank him (and you) for answering some questions that have been bothering me ever since I was first told about Alpha back in 1991. The DEC sales people were full of reason why 64 bits would be far better, but few of them seemed to make much sense. Nick's article has finally closed that issue for me.

Believe me, there are not many "news" sites that also remedy gaping holes in my basic understanding of how computers work! The Inquirer strikes again...

Tom

I just wanted to tell you in response to your article “Why 64-bits are good, and why they're not,” that you are wrong. 32bit addressing has a 4GB cap. However, Intel implemented 36bit in the Xeon and later in the P4 Northwood, therefore allowing the processors to address a maximum of 64GB of RAM.

The author may have extensive computing platform experience, but he is doing a rather bad job of explaining the cost/benefits of 64 versus 32 bits.

The compiler writer and OS library writer are responsible for wringing the most out of the machine and therefore it takes time for the OS and programs to get faster on a 64-bit arch.

One big reason you get a big boost out of 64-bit computing is the size of the data transfer. You may not need 64-bit integers very often, but it is the 64-bit transfers that make things run faster. You use memset and memcpy alot in C programming. Those libraries will be written to use the biggest transfer blocks available. Transferring 8000 bytes 4 bytes at a time will take twice as long as 8 bytes at a time. That is also why SSE2 will have such an impact on the Athlon64. SSE2 can handle 128-bit transfers.

When you compare an optimized 32-bit library on either sun or sgi machine versus 64-bit you will see many changes in the machine instructions to improve performance. The libraries also grow faster over time. It took years for the server OSes to get to where they are now.

This bodes well for linux, and bad for Windows. The 64-bit windows is not likely to use have optimized the OS for 64 bits, while unixes and linux have had 64-bit optimizations for years.

Jeff Hill

p.s. Keep up the good work!

This article is confusing, muddled, and poorly written. The author jumps to and fro without any consistency or coherency making the article hard to follow.

The criticisms of 64 bit he produces are pure drivel too.

I have users who will jump at 64 bit PCs. Engineers will love the increased memory and high bandwidth for modelling software. Artists will love the ability to natively work with 64 bit colour values. Video editors will love the large amounts of memory - hold an entire DVD in memory? No problem.

The author suggests that Intel's PAE is an acceptable substitute for large memory access. If it takes microseconds to bank-switch then that is too slow. Remember that direct access takes nanoseconds - a thousand times faster.

Roll on Windows XP-64

Quentin Stephens

"On the other hand, if you are interested primarily in ordinary, single user workstations, what does 64-bit addressing give you today? The answer is precious little. The needs of workstations have nothing to do with the matter, and the move to 64-bit is being driven by server requirements. µ"

for users of scientific applications (molecular modeling, physics simulation), large 3-d modeling applications and CAD, the 4GB limit of 32-bit does come into play. Even non-linear video digital video editing can benefit from the increase in address space. If the author meant 'single user PC' or 'typical desktop PC' - that's one thing, but by saying that 64 bit means nothing to Workstations, I'm afraid he's opened up a can of worms.

Jon Franz