[opensuse] What's the point with 64 bit
it seems that almost everybody I've talked to has advised me not to go with the 64 bit version of OpenSuse because there really is no speed advantage (if that's not good advice I'd like to hear otherwise). So then, if there is no speed advantage, what's the point in even having a 64 bit processor right now? -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
On Monday 14 May 2007 20:23, Pueblo Native wrote:
it seems that almost everybody I've talked to has advised me not to go with the 64 bit version of OpenSuse because there really is no speed advantage (if that's not good advice I'd like to hear otherwise). So then, if there is no speed advantage, what's the point in even having a 64 bit processor right now?
Addressing large virtual address spaces and / or installing large amounts of physical RAM. If you don't need one or both of these things, it's just overhead without payback. Randall Schulz -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
otherwise). So then, if there is no speed advantage, what's the point in even having a 64 bit processor right now?
Addressing large virtual address spaces and / or installing large amounts of physical RAM.
If you don't need one or both of these things, it's just overhead without payback. Although the analogy is going to be a little contrived, its something like
On Monday 14 May 2007 22:28, Randall R Schulz wrote: the concept of cylinders in an internal combustion engine... there were cars made back in the 30s-50s with 10, 12, and 16 cylinders... but due to harmonics, balance, and other issues (8) seems to be the best (optimum) number of cylinders. von Neumann processors are going to be similar... my gut feeling is that 32 bit width is going to be optimum and that 64 bit is the beginning of the end of no returns. I mean if PCs really ever do need to have more than 4gig of real storage/virtual storage then.. .maybe. But also remember that the processor's instruction set complexity (and performance) play a role (are impacted). I really think that multiple cores is going to be more practical (32 bit) than a wider bus. I would much rather have a quad processor (32 bit) right now than a 64 bit bus... which for the most part I do not need. -- Kind regards, M Harris <>< -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
M Harris wrote:
On Monday 14 May 2007 22:28, Randall R Schulz wrote:
otherwise). So then, if there is no speed advantage, what's the point in even having a 64 bit processor right now?
Addressing large virtual address spaces and / or installing large amounts of physical RAM.
If you don't need one or both of these things, it's just overhead without payback.
Although the analogy is going to be a little contrived, its something like the concept of cylinders in an internal combustion engine... there were cars made back in the 30s-50s with 10, 12, and 16 cylinders... but due to harmonics, balance, and other issues (8) seems to be the best (optimum) number of cylinders. von Neumann processors are going to be similar... my gut feeling is that 32 bit width is going to be optimum and that 64 bit is the beginning of the end of no returns. I mean if PCs really ever do need to have more than 4gig of real storage/virtual storage then.. .maybe.
Need, I don't think so, but remember, computers or government, what is spent rises as a rate that quickly eats up any surplus and then some. -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
On Tuesday 15 May 2007 05:57, Pueblo Native wrote:
M Harris wrote:
On Monday 14 May 2007 22:28, Randall R Schulz wrote:
otherwise). So then, if there is no speed advantage, what's the point in even having a 64 bit processor right now?
There does seem to be a speed difference. I serve hi res jpegs via apache and MySQL. The client sees the images smoothly and constantly loading and faster than the old 32 bit setup I had. Both were 1.8 MHz. Or maybe it's because of something else. I'm lost in dual core register alu cache sort of talk:-( Just my 0.02 Euros. Cheers, Steve. -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
M Harris wrote:
Although the analogy is going to be a little contrived, its something like the concept of cylinders in an internal combustion engine... there were cars made back in the 30s-50s with 10, 12, and 16 cylinders... but due to harmonics, balance, and other issues (8) seems to be the best (optimum) number of cylinders. von Neumann processors are going to be similar... my gut feeling is that 32 bit width is going to be optimum and that 64 bit is the beginning of the end of no returns. I mean if PCs really ever do need to have more than 4gig of real storage/virtual storage then.. .maybe. But also remember that the processor's instruction set complexity (and performance) play a role (are impacted).
Well, I guess in 2-3 years 4GB of RAM will be common. And it's definitely not only about bus, it's about register width (and therefore about ALU as well), which increased as well. 64bit registers help when computing numbers larger than "common" int. That was probably the idea behing ooooold 64bit MIPS, Alphas and such, whose real world speed may be compared with Pentium or even i486. 64bit CPUs are faster when working with large integers (which is not that frequent I presume), and they are a must when your application needs >3GB of memory. No other advantages here. I'm not sure where, but two or three years ago I saw a mysql and/or apache benchmark comparing 32bit vs. 64bit version performance on Athlon 64. If I remember correctly, 64 bit versions were faster by tens of percents. In other words, it depends on software as well.
I really think that multiple cores is going to be more practical (32 bit) than a wider bus. I would much rather have a quad processor (32 bit) right now than a 64 bit bus... which for the most part I do not need.
Well, talking about bus, 64 bit wide memory bus is here since SDR DRAM was introduced back in 1990's. Double channel efectively doubles that to 128bits I believe. Itanium 2 uses quad-channel memory bus and it gets quite confusing when considering multi-socket Opterons.... Multi-core - well, we get back to software then. Most people run single application and only small fraction runs more than 2 CPU intensive apps at a time. Realizing that most common today application software is single thread, you don't really need more than a double core. Quad core is investment for the future while the same goes for 64 bit CPUs. In fact, this discussion doesn't have a point. You can't buy 32bit only CPU (for common home use) anymore, at least there's no such worth mentioning. Yes, there's VIA, but it doesn't fit into "common home use". Cheers, Tosuja -- Petr "Tosuja" Klíma Mail: tosuja@tosuja.info Web: www.tosuja.info ICQ: 52057532 -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
|64bit CPUs are faster when working with large integers (which |is not that frequent I presume), and they are a must when your |application needs >3GB of memory. No other advantages here. | Recent cpu-cores have greatly simplified pipleines compared to pentium4,powerPC, etc. the cores in core2duo are infact pentiumIII with a simplified pipeline to speed up internal clock-speed. 32bits are just filling up the 32lsb bits of each registers/mem spot. thus no need for extra silicon to differenciate between 32bit and 64bit. powerPC had some logic which made it possible to crunch two 32bit numbers simultaineously. resulting is some really impressive openssl performance. -- MortenB -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
Hello,
On Tue, 15 May 2007 09:42:21 +0200
Morten Bjørnsvik
powerPC had some logic which made it possible to crunch two 32bit numbers simultaineously. resulting is some really impressive openssl performance.
I don't know PowerPC in detail, but is it the one by Altivec(SIMD)? Regards, eshsf -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
|-----Original Message----- |From: eshsf [mailto:eshsf@mbj.nifty.com] |> powerPC had some logic which made it possible to crunch two |32bit numbers simultaineously. |> resulting is some really impressive openssl performance. | |I don't know PowerPC in detail, but is it the one by Altivec(SIMD)? | Yep that was the name of it. Turning slightly opptopic now :-) On general computing a 2.8GHz Xeon were approx 1.5times faster than a 1.2GHz G4 powerbook. But with openssl the result were the opposite. The laptop were the fastest machine in the office! This were also the reason why apple always claimed Photoshop were faster on mac than on windows. They found the most intensive plugin using altivec and compared those to windows. The Cell used in PS3 is the next evolution of this technique. Properly tuned it is the fastest stuff on the planet. But it takes time to rewrite old code and learn the proper techniques. Thanks -- MortenB
On Tue, 15 May 2007 12:11:11 +0200
Morten Bjørnsvik
|-----Original Message----- |From: eshsf [mailto:eshsf@mbj.nifty.com] |> powerPC had some logic which made it possible to crunch two |32bit numbers simultaineously. |> resulting is some really impressive openssl performance. | |I don't know PowerPC in detail, but is it the one by Altivec(SIMD)? |
Yep that was the name of it. Turning slightly opptopic now :-)
On general computing a 2.8GHz Xeon were approx 1.5times faster than a 1.2GHz G4 powerbook. But with openssl the result were the opposite. The laptop were the fastest machine in the office!
This were also the reason why apple always claimed Photoshop were faster on mac than on windows. They found the most intensive plugin using altivec and compared those to windows.
It's cool. :) BTW, I found the following pages. it is a very interesting though it is a little old content. The Pentium 4 and the G4e: an Architectural Comparison http://arstechnica.com/articles/paedia/cpu/p4andg4e.ars Thanks, eshsf -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
Multi-core - well, we get back to software then. Most people run single application and only small fraction runs more than 2 CPU intensive apps at a time. Realizing that most common today application software is single thread, you don't really need more than a double core. Quad core is investment for the future while the same goes for 64 bit CPUs. Multiple cores are for the kernel... not so much for parallel thread
On Tuesday 15 May 2007 01:43, Petr Klíma wrote: processing with an app. In other words, at any given moment my little 'ol desktop machine has somewhere between 80 and 120 processes running on it... all through the same funnel... I mean ALU. With a quad core (or more) the kernel can assign those 80+ processes to the first available CPU.... the more the merrier... this potentially speeds everything up by an order of magnitude... assuming minimum resource contentions. We've been doing this for years really... one cpu funnel for normal ALU stuff, and the other one for video processing... starting back in 1980 with the VIC 20. When you start talking about VMs running, or multi-user app servers then the quad core (or multiple cpu) is a must. -- Kind regards, M Harris <>< -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
Our experience is that x86_64 does have a speed advantage in 64bit over 32bit -- this is for heavy compilation and running of computer algebra ==John ffitch -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
Pueblo Native
it seems that almost everybody I've talked to has advised me not to go with the 64 bit version of OpenSuse because there really is no speed advantage (if that's not good advice I'd like to hear otherwise). So then, if there is no speed advantage, what's the point in even having a 64 bit processor right now?
There is speed advantage for certain applications. Another difference for x86-64 is better handling of memory and data over 4 GB, Andreas -- Andreas Jaeger, aj@suse.de, http://www.suse.de/~aj/ SUSE LINUX Products GmbH, GF: Markus Rex, HRB 16746 (AG Nürnberg) Maxfeldstr. 5, 90409 Nürnberg, Germany GPG fingerprint = 93A3 365E CE47 B889 DF7F FED1 389A 563C C272 A126
On 5/15/07, Pueblo Native
it seems that almost everybody I've talked to has advised me not to go with the 64 bit version of OpenSuse because there really is no speed advantage (if that's not good advice I'd like to hear otherwise). So then, if there is no speed advantage, what's the point in even having a 64 bit processor right now?
Looks like in all the descriptions no-onen clarified that openSUSE, of _all_ the Linux distributions out there, is pretty much the best for 64bit support. Unlike Debian and Ubuntu, it's BiArch compatible, so you can still get your favourite 32bit applications on it (and of course hence you can have flash etc on your 64bit install). In summary, you will get a performance improvement with some applications with a 64bit install, and you shouldn't really lose out in any way (on openSUSE). Regards, -- Francis Giannaros http://francis.giannaros.org -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
Pueblo Native wrote:
it seems that almost everybody I've talked to has advised me not to go with the 64 bit version of OpenSuse because there really is no speed advantage (if that's not good advice I'd like to hear otherwise). So then, if there is no speed advantage, what's the point in even having a 64 bit processor right now?
I haven't heard about there being no speed advantage. However, even ignoring speed, you can handle bigger files, databases etc. -- Use OpenOffice.org http://www.openoffice.org -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
On Tuesday 15 May 2007 04:19, James Knott wrote:
Pueblo Native wrote:
...
I haven't heard about there being no speed advantage.
There certainly is a potential speed advantage, but you won't see it automatically just because you switch to a 64-bit processor (or to a 64-bit OS on a CPU that does both).
However, even ignoring speed, you can handle bigger files, databases etc.
You can handle bigger files and datasets in applications that need to keep their entire contents in primary storage concurrently, which is very few applications. Most applications bring file contents into memory incrementally or as needed and for most applications, I/O time will dominate the running time when operating on extremely large files. As processors get faster (or more numerous), the degree to which both primary (RAM) and secondary (disk) storage become bottlenecks just increases. If you have demanding computational tasks and need to increase the performance of your system when executing such tasks, you have to understand the demands they place on the system and balance your hardware for that demand. Otherwise, just switching to, say, a faster CPU will be mostly a waste.
-- Use OpenOffice.org http://www.openoffice.org
Randall Schulz -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
it seems that almost everybody I've talked to has advised me not to go with the 64 bit version of OpenSuse because there really is no speed advantage (if that's not good advice I'd like to hear otherwise). So then, if there is no speed advantage, what's the point in even having a 64 bit processor right now? There's already been lots of excellent discussion. In my work, I have been doing 64-bit computing for over 10 years starting with Digital's OSF/1 (now Tru64 Unix) on the Alpha. Linus actually ported Linux over to the Alpha about 1995. We all know about the memory advantage, which is a large linear memory model, but even for systems with under 3GB, a 64-bit application can take advantage of 64 bit integers. On the x86-64 chip in 64-bit mode you have twice as many registers available to both
On Mon, 14 May 2007 21:23:30 -0600
Pueblo Native
On Mon, 2007-05-14 at 21:23 -0600, Pueblo Native wrote:
it seems that almost everybody I've talked to has advised me not to go with the 64 bit version of OpenSuse because there really is no speed advantage (if that's not good advice I'd like to hear otherwise). So then, if there is no speed advantage, what's the point in even having a 64 bit processor right now?
I had the impression one needs a 64-bit cpu to count the amount of reboots needed for a M$ machine.... Oh no! Sorry i've forgotten that one uses floating point notation for such trivial calculations nowadays;-) -- pgp-id: 926EBB12 pgp-fingerprint: BE97 1CBF FAC4 236C 4A73 F76E EDFC D032 926E BB12 Registered linux user: 75761 (http://counter.li.org) -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
Pueblo Native wrote:
it seems that almost everybody I've talked to has advised me not to go with the 64 bit version of OpenSuse because there really is no speed advantage (if that's not good advice I'd like to hear otherwise). So then, if there is no speed advantage, what's the point in even having a 64 bit processor right now?
If you only have 2 gig of memory, 32bit will be faster. The thing that is great about 64bit, is it can handle memory above 2 GB natively, with out doing anything special. The reason that 32bit is faster is it only has to pass around 32 bit pointers vs 64 bit pointers. On the other hand, the faster memory handling above 2 GB with 64bit OS makes up the difference . Now memory isn't the only factor, but is good enough for this discussion. Things that use a lot of math also benefit with 64bit pointers, as the math is just faster. So unless you are running a mission critical DB or some vector graphics stuff, 64bit won't help either. Also keep in mind that the program needs to be 64bit for it to take advantage of the 64bit math. Hope this helps. -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
On Tuesday 22 May 2007 21:01, Tom Miller wrote:
...
Now memory isn't the only factor, but is good enough for this discussion.
In large part, memory _is_ the only factor. Or, at least, a very large one. Memory bandwidth, that is. Processors speed increases have continued to outstrip that of both primary and secondary storage, though on a per-CPU (core) basis, that has pretty much come to an end. Now more CPUs (or cores) are being placed on a given chip or board, and the effect is the same. So for many (I'd venture to say most) applications, memory bandwidth vs. demand is the dominant performance factor. Thus, 32-bit systems will quite often outperform 64-bit ones. Unless, of course, you really need something that is only possible with the 64-bit system (very large virtual address spaces, that is).
Things that use a lot of math also benefit with 64bit pointers, as the math is just faster. So unless you are running a mission critical DB or some vector graphics stuff, 64bit won't help either.
Don't you mean a 64-bit memory bus? Bigger pointers are all downside. If you must address that much memory, then there's no alternative, but it's all overhead, regardless.
...
Randall Schulz -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
Tom Miller wrote:
Pueblo Native wrote:
it seems that almost everybody I've talked to has advised me not to go with the 64 bit version of OpenSuse because there really is no speed advantage (if that's not good advice I'd like to hear otherwise). So then, if there is no speed advantage, what's the point in even having a 64 bit processor right now?
If you only have 2 gig of memory, 32bit will be faster.
It may just be my imagination, but x86-64 seems to be faster on my dual core 3.2GHz with 1 G memory. It seems to me that the place I notice it most is in Yast2 when it's refreshing the install repo's. It seems to process the information faster. Everything else seems to go just a bit faster, but Yast2 is where I notice it most. As I said, it may just be my imagination. However, I wouldn't hesitate to load it with just one Gig memory. -- (o:]>*HUGGLES*<[:o) Billie Walsh The three best words in the English Language: "I LOVE YOU" Pass them on! -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
Tom Miller wrote:
The reason that 32bit is faster is it only has to pass around 32 bit pointers vs 64 bit pointers. On the other hand, the faster memory handling above 2 GB with 64bit OS makes up the difference .
Given that data is transferred in parallel, that is all bits at the same time, how does it take longer to transfer 64 bits? -- Use OpenOffice.org http://www.openoffice.org -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
On Wednesday 23 May 2007 04:11, James Knott wrote:
Tom Miller wrote:
The reason that 32bit is faster is it only has to pass around 32 bit pointers vs 64 bit pointers. On the other hand, the faster memory handling above 2 GB with 64bit OS makes up the difference .
Given that data is transferred in parallel, that is all bits at the same time, how does it take longer to transfer 64 bits?
Because fewer of the data units that the program requests can be transferred per unit time when the program is sequentially accessing word-sized units, which is a very common case and one the the things caches and fetch-ahead mechanisms serve to accelerate.
-- Use OpenOffice.org http://www.openoffice.org
Randall Schulz -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
participants (15)
-
Andreas Jaeger -
Billie Erin Walsh -
eshsf -
Francis Giannaros -
Hans Witvliet -
James Knott -
Jerry Feldman -
jpff -
M Harris -
Morten Bjørnsvik -
Petr Klíma -
primm -
Pueblo Native -
Randall R Schulz -
Tom Miller