Interested in dual Cores? Got some 64 bit benchmarks.
AMD Athlon 64 dual core prototype / cinebench = GO TO PAGE 2
Ok, it's a good indicator of what we can expect in performance increases from Dual cores on a 64 bit processor. It's translated through google because it's at CBIT in germany or somewhere, but It's all relevant info IMO.
If you look at the red stripe towards the top it shows the Atthlon (one chip) with two physical cores beating out a dual (2 chip) Opteron each with one physical core.
There is more to the early prototype benches than that, but you'll have to read it your self. It's all good.
Ok, it's a good indicator of what we can expect in performance increases from Dual cores on a 64 bit processor. It's translated through google because it's at CBIT in germany or somewhere, but It's all relevant info IMO.
If you look at the red stripe towards the top it shows the Atthlon (one chip) with two physical cores beating out a dual (2 chip) Opteron each with one physical core.
There is more to the early prototype benches than that, but you'll have to read it your self. It's all good.
Comments
BTW the speed bump is very small : 2-3 % between a dual core and a dual CPU, but now if you compare a dual single core computer with a dual dual core one, the difference will be bigger ...
Originally posted by Powerdoc
The results of the benchmarks are logical.
BTW the speed bump is very small : 2-3 % between a dual core and a dual CPU, but now if you compare a dual single core computer with a dual dual core one, the difference will be bigger ...
You do mean 2 to 3% is One dual core CPU vs. Two single core CPUs right?
And do you mean Logical, or logical as in logical CPU's because only the Hyperthreaded processors were using logical CPU's. They counted cores as physical CPU's because they really are.
Passing from the 1 to the 2 Cores processore Athlon 64 Dual Core it has lowered the time of elaboration from 77,5 second ones to 41,4 second ones: draft of an increment consisting, equal to approximately 1,87 times the power of obtainable elaboration with the same operation using a single processore.
That's close to 2x (1.87x) the speed as a single core processor for that operation. I think it's impressive. That's Similar to the graphics results as seen using SLI. All operations average is almost 2x faster using SLI. (if the app is coded to use it)
Originally posted by onlooker
You do mean 2 to 3% is One dual core CPU vs. Two single core CPUs right?
And do you mean Logical, or logical as in logical CPU's because only the Hyperthreaded processors were using logical CPU's. They counted cores as physical CPU's because they really are.
Yes it was one dual core CPU vs Two single core CPU
I say logical, because the latency and perhaps the bandwitch between the two interconnected cores is much faster and efficient than between two separates CPU
That's close to 2x (1.87x) the speed as a single core processor for that operation. I think it's impressive. That's Similar to the graphics results as seen using SLI. All operations average is almost 2x faster using SLI. (if the app is coded to use it)
Dual core and more generally multicore or Cell like architecture are the way to go. Hyerscalar architectures has reach his limit, and it seems that the Mhz race has slowed down a bit.
The only cons with MP or multicore architecture, is that the software must be optmized to take advantage of such artichecture. Of course it's always an advantage if you run several applications altogether.
Originally posted by mikenap
Multitasking performance has always been crucial in my business. I own a small print shop that offers a variety of services, from large format printing to web design, scanning, etc. Often times it's me running the shop, doing quotes in filemaker, emailing in Entourage, wasting too much time here, as well as the heavy lifting in PS, etc. What has keept me (one of the reasons at least) so anchored to the Mac platform is OSX's ability to multitask. I have a dual processor quicksilver with alot of Ram. The PC's in my office seem to falter big time in this type of enviornment when running many apps together and freequently switching between them. Would this problem improve with multiprocessor PC's in the future? Is the reason OSX cranks like this on my main workstation because it's multiprocessor? Or is OSX a superieor OS for this type work?
AFAIK Windows XP supposedly can multitask, but it's OS X 's mostly unseen Unix side that makes it a better at it running multiple applications at once. I used to know all the technical aspects of this stuff when OS X first hit the shelves, but it's so standard now that I never think about it any longer, and thusly I forget.
Originally posted by Henriok
Using benchmarks from a dual core Athlon won't help us much figuring out what we are to gain from a dual core G5. The setup is radically different since the Athlon has an on die memory controller serving both cores thus severely reducing latency between the cores. A dual core G5 will rely on the off die system controller witch will cause over head. A 2x dual core machine will have a system controller managing 4 CPUs.. a heavy load indeed. Or is the plan to use two system controllers with a bank of RAM for each dual core processor, like dual processors Opterons are handling it today. Ouch.. this is getting more expensive the more I think of it.
How do you know? Doesn't the Power4, and 5 have on die memory controllers?
I thought the 9xx Dual core PPC's were based on this design.
Originally posted by onlooker
How do you know? Doesn't the Power4, and 5 have on die memory controllers?
I thought the 9xx Dual core PPC's were based on this design.
Neither POWER4 nor 970 have an on die memory controller, the POWER5 do however.
How do I know 970MP won't have on die memory contoller? Three reasons:
1) The kept the moniker "970" and i think it would be changed if they included such an important feature as a on die memory controller.
2) The three page teaser Morpheus shared with us some time ago, that outlined the differences between 970FX and 970MP, didn't mention any on die memory controller, just a new PowerTune, dual core, increased L2 cache and a larger pin count.
3) The increased pin count should also indicate that the cores doesn't share any resources, but i don't know how much more pins the 970MP will use. If it's just a couple, this would be a moot point, but it it's doubled, it's a sure sign of lack of memory controller.
970MP isn't a POWER5 derivative, it's "just" two 970GX cores on the same die. 970GX is 970FX with larger L2 cache and some updated power saving features. Nothing fancy really.
I would guess that a POWER5 derivarive would include features like on die memory controller and SMT, and would change its name to.. let's say PPC 975. There would probably be a dual core version of it too, the 975MP, and a 65 nm version later on, the 975FX.
Originally posted by onlooker
How do you know? Doesn't the Power4, and 5 have on die memory controllers?
I thought the 9xx Dual core PPC's were based on this design.
Only the Power 5 has an on-die memory controller, in fact the 970 has a pretty high memory latency beacuse of the seperate controller. 135ns best case for G5 vs. 95ns best case for G4.
When the real numbers come out for other applications, I wouldn't expect anywhere near 87% increases benchmarking using a singe application, of course there are many other benefits of using dual core processors.
It's not really important to me to have multithreaded apps, as OS X uses them so such effect already, (games and 3d apps perhaps) productivity can be amazing with OS X and two processors.
Now quad processors...
Originally posted by Henriok
Neither POWER4 nor 970 have an on die memory controller, the POWER5 do however.
According to IBM they had multicore on the Power4, and the Power4+ , but if these were prototypes that never saw the light of day or not, I don't know, who's to say they will not include them again. Anything is possible, They have been working on this for ages longer than AMD, or intel.
I'm not saying they will for sure be there, but I'm keeping faith, and hope up, and my options open. If Apple was wise enough to request it, It could easily be on them.
Article LINK
The best known Power5 attribute is its integration of two discrete RISC cores on a single chip. Announcements from AMD, Intel, and Sun Microsystems (Profile, Products, Articles) regarding upcoming multicore processors focused attention on this aspect of Power5, but multicore was also a feature of its predecessors, Power4 and Power4+. According to IBM, Power5 is fully compatible with Power4 executables. The wonder of multicore is that it delivers the pipe dream of more speed in less space without a marked increase in heat. But as you?ll see, multicore is not simply SMP on a chip.
Originally posted by onlooker
According to IBM they had multicore on the Power4, and the Power4+ , but if these were prototypes that never saw the light of day or not, I don't know, who's to say they will not include them again. Anything is possible, They have been working on this for ages longer than AMD, or intel.
The POWER4 and POWER4+ (180 nm and 130 nm versions) are dual core processors but they don't contain a memory controller.
A memory controller is a device that arbitrates the CPU's traffic to the main RAM, a part of what's called the North Bridge. If this device is mounted on the same die as the proceso it can talk to the RAM quicker. The response time to RAM is quicker, latency is lower, overhead is cut, complexity on the mother board is reduced, heat is reduced, it's easier to keep cool, transistor count can be kept down. This is a good thing.
But it's a more complex processor design, combining many different tasks. It's also not a modular design and it's harder to keep the development of the CPU and memory controller separare wich can be useful if a processor will be used in different environments.
If you add another CPU they often have to talk to each other trough the memory controller (processes share RAM). If this is two separate processors they have to use an off die memory controller. Just like it's handled by all dual core Macs today. If you have two processor cores on the same die, it would be quite inefficient if you had to take the traffic between the two cores off die. This will be the case witch 970MP, POWER4 and the upcomming crop of dual core Pentium/Xeons.
This is however NOT the case of the current crop of Athlon and Opteron processors that have had an on die memory controller since day one. A memory controller that's been geared to take on another core in the future.. witch is now. This is not the case of the current versions of POWER5 either , nor the future dual core G4s by Freescale as they too have an on die memory controller.
Originally posted by onlooker
Duh.. My bad. I lost track of what we were talking about.
No worries..
Currently in another thread:
970MP confimed by IBM!
Originally posted by onlooker
AFAIK Windows XP supposedly can multitask, but it's OS X 's mostly unseen Unix side that makes it a better at it running multiple applications at once. I used to know all the technical aspects of this stuff when OS X first hit the shelves, but it's so standard now that I never think about it any longer, and thusly I forget.
Multitasking in the PC world has been around since at least Win95, on through Win2k, and continues on XP. It is only since OSX, that Apple "caught up" with a pre-emptive multitasking functionality. I say that in quotes because it depends on what consider a step forward vs. being "caught up". Classic OS actually had an advantage in user responsiveness with its cooperative multitasking setup (for all the downsides that are attributed to CMT). When the user invokes a task, it gets done "then", not eventually. Multitasking competing processes did suffer somewhat, but it certainly was not unworkable. Now we have OSX that, like Windows has for sometime, uses PMT to balance competing processes more effectively. The downside is that the UI is a step less responsive (unless you throw brute CPU at it). So by moving to OSX, we did get some multitasking benefits, but not w/o giving up some things in the area of UI responsiveness (hence the steps forward actually incurred steps back, as well).
IME, once we get to PMT sort of environments (modern Windows vs. modern Mac OS), the quality of the multitasking is really more dominated by the CPU you can throw at it, rather the technical details of how that PMT exists in the OS. From that respect, Windows and OSX have both become considerably more processor hungry (though, I would still concede that OSX manages to still make do with older/lesser CPU's much better than XP). You can flog either OS with some heavy multitasking, but you better have some strong CPU(s) to leverage, or you will be stuck in molasses, regardless of how the PMT is accomplished in the OS.
Where Windows may have fallen on the wayside somewhat with multitasking performance is not necessarily in the OS (other than it being resource hungry), rather the hardware it has had to run on along the ages. For a while, it had to run on PIII and early P4 processors with cutdown caches and/or limited system buses to main memory. So they never could achieve their potential when it came to heavy multitasking. Now we have higher-clocked P4's with larger caches, but the latency of their monumental pipelines is not really the greatest thing to mix with rapidly switching, multiple, competing tasks. What about Athlons and Opterons? Well, it's not like MS has had the greatest political climate to optimize to anything other than Intel, historically (though, this barrier may be crumbling as of late). In some alternate reality where MS had really made concerted efforts to exploit AMD hardware, we might observe/have observed a completely different level of multitasking capability in Windows. Does any of that really mean OSX is better at it? It's not clearcut for me, and I maintain that the impression of how well it does it can be made or broken on the level of hardware you run it on.
So there you have it, according to my 2 cts.
Originally posted by Henriok
No worries..
Currently in another thread:
970MP confimed by IBM!
I know man, that is the Sh*t of all Sh*t!!!
BTW, thanks for that lengthly explanation Randycat99
Multitasking is being able to have more than one program open for use at the same time. Style of process/thread multi-programming doesn't matter. Mac had that with Multi-finder since the System 6.0.0 days. The older Switcher was just too clunky to really consider a true multitasker.
Multi-programming is a technical term for an OS design that allows more than one process to concurrently use the processor, whether cooperatively or preemptively. Again Mac OS has had that since System 6.0.0 (cooperative version) as evidenced by some of the desk accessories and background printing.
Win 95/98 had a mixed bag of preemptive techniques MS marketed as preemptive, but were largely still cooperative as anything that touched DOS or 16-bit code was cooperative only. Since most of those OS's calls were 32-bit wrappers around the old 16-bit code you can see what was really going on with the promises of preemption in 32-bit clean code. Win2K was vastly better but still had some issues in isolated spots with legacy application support and OS shims for backward compatibility. WinXP finally went fully preemptive with a DOS emulation layer rather than running some things on top of DOS. The emulation layer removed the constraints of cooperativeness just as Classic has done for OS9 emulation.
Seems more like parallel development rather than anyone playing catch-up. The rest I pretty much agree with.
Originally posted by Hiro
Randycat, multitasking has 2 separate but closely related common uses. The below 2 paras cover those different uses in a way some OS types have proposed to use to remove ambiguity in the use of the term multitasking. It can get confusing using the same word whan you compare MT from a users perspective with MT from a machines perspective.
I sense we are disputing over simply a shift in vernacular terms vs. the textbook technical definitions.
Quote:
Multitasking is being able to have more than one program open for use at the same time. Style of process/thread multi-programming doesn't matter. Mac had that with Multi-finder since the System 6.0.0 days. The older Switcher was just too clunky to really consider a true multitasker.
Multi-programming is a technical term for an OS design that allows more than one process to concurrently use the processor, whether cooperatively or preemptively. Again Mac OS has had that since System 6.0.0 (cooperative version) as evidenced by some of the desk accessories and background printing.
I think we can agree that the vernacular term "multitasking" has taken on more meaning than just being able to have more than one application open at a time. Just as System 6 had this feature, so was there in Windows 3.1. By the time System 8.x and Win95 were on the scene, "multitasking" had inherited additional nuances in meaning to include multiple processes could be running (which you would have described more technically as "multi-programming"). Of course, I think the masses have heard of the term cooperative/pre-emptive multitasking more likely, than "multi-programming". I'm not about to embark on a crusade to reinstate the academic useage as the more correct term for the people, but your point is taken.
Quote:
Win 95/98 had a mixed bag of preemptive techniques MS marketed as preemptive, but were largely still cooperative as anything that touched DOS or 16-bit code was cooperative only.
Hey, it was a start. I never meant to give the impression that Win95 had its pre-emptive-ness baked entirely from the start.
Quote:
Seems more like parallel development rather than anyone playing catch-up. The rest I pretty much agree with.
Very true, but you have to agree (or not...) that the PMT theme was much farther along in development and useability when Windows users had 2k and XP, while the only "field ready" option for Mac users (not talking about a "just about ready beta"