Why no Quads?
I was wondering why Apple has not released a quad powermac yet... Does it have to do with the current architecture? Will the next chip Apple uses be able to be offered in quad versions? I use my powermac for music and even the dual gig which i have now is not sufficient. Ofcourse most musics apps aren't near being complete for OS X yet and don't support duals yet...
Here is hoping for the future
Here is hoping for the future
Comments
1 CPU: 133MHz per CPU
2 CPUs: 66MHz per CPU
4 CPUs: 33MHz per CPU
According to motorola, the G4 is only (yeah right) bandwidth limited when you are doing AltiVec on dual CPU systems (Mac OS X in other words).
However, with the 2MB of SRAM L3 per CPU, you probably would see some performance increase in small photoshop files, as the data just stays on the 1GHz effective L3, which isn't on the MPX (a front-side aka system) bus.
Barto
[quote]
originally posted by me
<strong>... I don't think you even need to get into feature parity and I/O to equalize Dual Xeon prices to DP 1Ghz prices, they're more expensive already.
G4s are cheaper and cooler and lower power than P4 xeons (and P4).
I'm no Apple apologist but as far as multi-CPU arrangements of 4 or more go, the price-performance edge rapidly shifts in favor of the PPC.
First let's deal with the cost issue for boxes we can buy right now. Anyone selling Dual Xeon boxes is doing so at a premium. They ain't cheaper than a DP 1Ghz, many cost a fair bit more. Granted they're probably scary fast, but they ain't cheap. (forget AMD for now) They require DP capable socket 601 boards, which aren't cheap, and they need XP-Pro (also not cheap, but pirateable if you're rolling your own). The costs quickly add up. It's more expensive than PowerMac, but it's also faster.
But the essential crux of Amorph's throw more PPC's at it solution is still quite valid. In fact, I'd say it's a compelling strategy given the PPC versus Xeon comparo.
4 1Ghz G4's @ 20watts dissapation each. 80 watts. It'll go comfortably in a QS size tower, with a strong but not prohibitively expensive PS.
4 2Ghz Xeons... AHAHHAHA!!! 75watts x 4 = 300 watts AHAHAHA!!! Not unless you have some jet turbine cooling fans and a HUGE PS and MoBo gussied up to deal with the power and heat requirements. It ain't happening. Even then, it certainly won't sit on your desk! And it only gets worse if you want to add more CPU's.
But that's making a false assumption that you could put MORE than TWO 'regular' Xeons in a box to begin with. YOU CAN'T. You need Xeon MP for that. Yep, Intel weenies everywhere take note: There are 3 desktop flavors of P4. Regular P4, Xeon P4 for dual configs, and Xeon MP for configs of 4 or more. Wanna guess how much those Xeon MP's cost? In March they were selling for $1177 for a 1.4 w/512KB L2 and over $3500 for a 1.6 w/1MB L2. Two months have passed, so prices are surely down, but even in a very generous estimate you're looking at $1000 dollars per chip!
Compare to the current G4 1Ghz at just a touch more for ALL FOUR CHIPS.
See a winner yet?
So what does Apple need to make this happen. Easy. Faster RAM and more of it. 4GB DDR266 would probably do the trick considering how well they've extended the life of SDR133. But what's more amazing, when you really think about it, is that using PPC's it could be done on a fairly conventional MoBo, in a regular ATX size case. Maybe the ultimate performance wouldn't be as good as if dual memory channels were thrown in but given the software and hardware integration, I bet they could do Powermac quads well before the law of diminishing returns kicked in, even on 4GB of DDR266. All, according to the magic calculator in my head, for no more than $1000 above a DP 1Ghz (excluding RAM allotment).
$4000 Powermac Quad anyone?
And that's using 300 dollar Ghz chips. Drop it down to $125 800's and you could start making comparatively affordable 4 ways right now.
</strong><hr></blockquote>
So there you have the cost per machine feasability. It's quite nice for PPC based machines, not so nice for x86 based quads. Same for heat and efficiency: You can cram more PPC's into a given space before heat and power consumption become a huge concern.
If Apple could use DDR266, 4 CPU's wouldn't be any more starved than 2 CPU's are (on SDR133). If the rumors of DDR333 are true, they'd even be a little better off. IF, a dual channel memory bus were used (effectively 128bit DDR) they'd have some impressive potential to say the least.
Still, methinks even plain DDR would open the possibility of PPC quads!
AIf Apple can't get the clock past 1.3 Ghz, quads might not be a bad idea. What with Quartz Extreme and the Next Gen Nvidia part due later this year we could have a real dramatic speed increase over what we are using now. Now it just has to happen
TITLE Mac OS X Server: Maximum RAM Supported Article ID: Created: Modified: 31274 11/14/00 8/15/01 -------------------------------------------------------------------------------- TOPIC This article discusses the maximum amount of RAM that can be used with Mac OS X Server. -------------------------------------------------------------------------------- DISCUSSION The maximum amount of RAM that Mac OS X Server can use varies depending on the version of Mac OS X Server. Mac OS X Server Mac OS X Server 1.0 can work with up to 1 GB RAM. Mac OS X Server 1.2 can work with up to 1.5 GB RAM. Mac OS X Server 10.0.x can work with up to 3 GB RAM. Notes: 1. Some computers may have RAM limits below those of Mac OS X Server. For example, you can install only 1 GB of RAM in a Power Macintosh G3 (Blue and White) computer. Although Mac OS X Server 1.2 can work with up to 1.5 GB of RAM, you obviously cannot install that much in this computer. 2. To allow MacOS.app to function correctly with Mac OS X Server and 1 GB or more of RAM, type the following at the command line prompt in a Terminal window: defaults write MacOS SysZoneMemorySize 0x10000000
That said, more than a dual is pretty much pointless at the moment since even one G4 is currently memory bound. Going DDR will certainly help that, but only testing an actual machine will tell us if >2 CPUs is a win. My guess is that it will be only in certain applications. A quad machine won't really be a big win until each processor has its own memory controller and RAM -- in this situation the RAM is very fast and not shared at all.
The current chipset is limited in how much memory it can address, but I'm not aware of any inherent limitations in how much MacOSX can address. The G4 can have physical addresses for 16 GBytes, with a maximum of 4 GBytes per process -- I'd be surprised if MacOSX doesn't support these limits. There have been rumours that Apple's new DDR chipset supports the full 16 GBytes. Wouldn't that be cool. :cool:
My guess would be fast DDR333 RAM (one bus) and 4 PPC's each with a pretty big (4MB or so) L3 cache.
Or
A dual channel DDR Mobo with regular 2MB cache PPC parts.
What are the issues?
What things would be muchg faster and what things wouldn't see much if any improvement.
If a quad PPC is only 30 % more efficient than an dualie in average it will not worth the prize. I think there is two sectors who need great raw power : 3 D rendering and video . 3 D rendering depend merely on the FPU, video can take in many case advantage of the Altivec stuff.
<strong>Why don't you tell us what you think a cost effective/feasible quad machine would like?</strong><hr></blockquote>
While I'm not as smart as programmer, the AMD hammer series looks pretty good.
1 DDR RAM controller on-chip per-chip. 1 CPU: single channel DDR RAM. 2 CPUs: dual channel DDR RAM. etc. etc.
The G5 is *apparently* going to do the same thing.
Barto
PS I know the 33MHz per CPU is over-simplified, but if every CPU is using it at once was my point. And I put in the disclaimer about the L3 Cache.
<strong>What are the issues?
What things would be muchg faster and what things wouldn't see much if any improvement.</strong><hr></blockquote>
One of two things has to happen: either the performance of memory has to catch up significantly to the performance of the processors, or each processor has to get its own memory. The first isn't very likely considering current processors can do ~60-80 instructions per uncached memory access, and even DDR will only drop that to 30-40... assuming the new processors don't have more functional units or higher clock rates. The second possiblity is what is rumoured for the G5 (and the G4 7500 if The Register is to be believed). When that happens 4x (and more) processor machines will be very effective because memory bandwidth increases every time you add a processor, instead of decreasing.