PDA

View Full Version : nForce Speculation


arn
Jul 29, 2002, 03:36 PM
The Inquirer (http://www.theinquirer.net/?article=4663) speculates that chip positioning on the leaked PowerMac Motherboard may hint at an NVidia NForce 2 connection with the new PowerMacs:

The one chip in recent years to buck the trend is Nvidia's nForce northbridge, positioned at 45 degrees to a mobo's main axes. Alas the blurry snapshots of the alleged Apple product make the info printed on top of the part unreadable, but its positioning strongly suggests the inclusion of Nvidia technology.

The chip in question is pictured here (http://www6.tomshardware.com/mainboard/02q3/020716/nforce2-03.html) (Tom's Hardware), which in this photo does not resemble the chip on the leaked PowerMac motherboard.

nForce on Apple speculation (http://www.macrumors.com/pages/2002/07/20020716224903.shtml) was sparked by an out-of-place reference on NVidia's nForce 2 specs pages on one feature that is not currently available on Mac systems.

esome
Jul 29, 2002, 04:25 PM
this coupled with the quad-processor rumor at MOSR (http://www.macosrumors.com/) make for a nice mac fantasy.

mymemory
Jul 29, 2002, 04:26 PM
Interesting... interesting that we are getting more video ram than Mhz...Mmm, I couldn't care less about video ram at this point. With 8 or 16 I'm fine but where are the stuff we really care about?

arn
Jul 29, 2002, 04:32 PM
Originally posted by mymemory
Interesting... interesting that we are getting more video ram than Mhz...Mmm, I couldn't care less about video ram at this point. With 8 or 16 I'm fine but where are the stuff we really care about?

NForce 2 is not just a graphics card. It's a chipset/motherboard.

http://www6.tomshardware.com/mainboard/02q3/020716/index.html

When purchasing a PC, the chipset is all too often overlooked, which is unfortunate because the performance of a computer depends on it. The chipset in a PC can make as much of a difference in performance as, say, the difference between using an Athlon XP 1600+ and an 1900+. And this is no joke: while the first DDR chipsets for the Athlon were faster than their SDRAM ancestors, they were always significantly outclassed by the generations that followed. In the end, pure processor power is not enough if the system itself puts on the breaks.

job
Jul 29, 2002, 04:35 PM
Hmm...interesting news.

But would it really be possible to incorperate a G4 chip on the nForce motherboard?

cyberfunk
Jul 29, 2002, 04:39 PM
Now, I understand what DDR is, SDRAM that talks to the chipset twice each cycle.. but what is DUAL DDR ???

Furthermore.. how will we know, when the powermacs come out.. wether or not there is a DDR connect between the chipset and the CPU, not just the chipset and the memory ?

I'd be happy to have someone knowledgable explain.

Thanks, Funk

hvfsl
Jul 29, 2002, 04:55 PM
Dual DDR is what the Pentium 4 has, but I have no idea of how it works.

I can't see Apple ever using the Nforce Motherboard on G4 chips, but they may use them in Macs. I heard half a year ago that Manchester University (UK) had developed special code morphing software that allowed chips to emulate other chips. One of the examples they gave was the AMD Athlon emulating a PowerPC G3, where a 500Mhz AMD chip was 90% the speed of a 500Mhz PowerPC G3 chip. The G3 could only emulate the AMD chip at about 70% of the speed. I saw the article in the UK PCW magazine. Manchester University also managed to make other chips emulate each other like the Power4 and the MIPS.

So it is possible to run Mac OS X on a 2Ghz PC and be the fastest Mac so far. This is the only way I think that we will see the Nforce in a Mac.

etoiles
Jul 29, 2002, 04:57 PM
so, I thought the nForce was 'low end' or 'entry level' graphics, all bundled up in a nice package but still not targeted towards high end graphics...would this make sense in a new 'pro' model ? Could you use this onboard processor just to run Quartz extreme and a high-end card for the actual content ?

I am confused :confused:

cyberfunk
Jul 29, 2002, 05:08 PM
One more thing.. if you look closely.. you'll see that is simply the exact same chip as the daughtercard closeup pic shows.. if you rotate the daughter-card.. then everything matches up with the "wide view" (with the graphics card obscuring some stuff).

I'm sorry to say this, but thats just a G4 chip , nothing more.

Everything matches up, the little yellow things, and the other chips. From the closeup, we can clearly determine that it's not a nForce chip we're looking at, but a CPU. :(

I'd love someone to prove me wrong tho ! :)

[Apple Legal doesn't like these pictures to spread :) ]

job
Jul 29, 2002, 05:09 PM
I think the nForce does have a dedicated AGP slot, but don't quote me on that....:)

job
Jul 29, 2002, 05:11 PM
Originally posted by cyberfunk
I'm sorry to say this, but thats just a G4 chip , nothing more.

Very true...but I can dream can't I? :) :)

cyberfunk
Jul 29, 2002, 05:15 PM
as arn said: NForce 2 is not just a graphics card. It's a chipset/motherboard.


In particular the nForce2 is a chipset that controls the rest of the motherboard.. while nForce (1) was a IGP (integrated graphics Proc) , nForce 2 can either be a SPP or a IGP, that is to say.. it can be offered just fine without the integrated graphics. The nForce 2 is a chipset in it's own right, it is no longer only a integrated graphics solution.

In plain english this means that Apple could integrate the nForce 2 chipset into the MB, but still use removable graphics cards. I highly doubt that Apple would choose to use integrated graphics in it's pro line.

However, as I noted above.. I dont think thats an nForce 2 chip we're looking at.. it's a G4. If u need more info.. look at the tom's hardware link Arn gave us.

kenohki
Jul 29, 2002, 05:20 PM
Note that according to the posted specs the nForce2 does NOT support multiprocessor configurations. So I doubt Apple will use an nForce2 chipset as it is in it's current form. Pray for DDR400 though. It would help to feed much more data to the vector unit on the G4 (AltiVec) which is usually starved for data.

cc bcc
Jul 29, 2002, 05:23 PM
Now, I understand what DDR is, SDRAM that talks to the chipset twice each cycle.. but what is DUAL DDR ???


It's actually dual channel DDR, which works like RAID but then for ram. Delivers higher bandwidth/throughput than normal DDR.

edit: with 2 standard DDR modules you get almost twice the bandwidth

Mr. Anderson
Jul 29, 2002, 05:25 PM
Well, seeing the pic again, I have to go with it not being a nForce chip - but no one has asked why the damn thing is 45° off the rest of the board? What benefit does this have? Weird.....

D

kenohki
Jul 29, 2002, 05:29 PM
Originally posted by hvfsl
Dual DDR is what the Pentium 4 has, but I have no idea of how it works.

I can't see Apple ever using the Nforce Motherboard on G4 chips, but they may use them in Macs. I heard half a year ago that Manchester University (UK) had developed special code morphing software that allowed chips to emulate other chips. One of the examples they gave was the AMD Athlon emulating a PowerPC G3, where a 500Mhz AMD chip was 90% the speed of a 500Mhz PowerPC G3 chip. The G3 could only emulate the AMD chip at about 70% of the speed. I saw the article in the UK PCW magazine. Manchester University also managed to make other chips emulate each other like the Power4 and the MIPS.

So it is possible to run Mac OS X on a 2Ghz PC and be the fastest Mac so far. This is the only way I think that we will see the Nforce in a Mac.

Dual DDR is just DDR with two memory controllers and two channels to main memory.

As for code morphing software...ever seen HP's Dynamo software. It is somewhat of a code morphing software that runs PA-8000 code on PA-8000. Supposedly they get up to a 20% performance boost by using dynamic optimization of code at runtime above what a compiler can do at compile time. Neat stuff. Check out Ars Technica's review.

http://arstechnica.com/reviews/1q00/dynamo/dynamo-1.html

Aciddan
Jul 29, 2002, 05:29 PM
Originally posted by hvfsl
I can't see Apple ever using the Nforce Motherboard on G4 chips, but they may use them in Macs. I heard half a year ago that Manchester University (UK) had developed special code morphing software that allowed chips to emulate other chips. One of the examples they gave was the AMD Athlon emulating a PowerPC G3, where a 500Mhz AMD chip was 90% the speed of a 500Mhz PowerPC G3 chip. The G3 could only emulate the AMD chip at about 70% of the speed. I saw the article in the UK PCW magazine. Manchester University also managed to make other chips emulate each other like the Power4 and the MIPS.

So it is possible to run Mac OS X on a 2Ghz PC and be the fastest Mac so far. This is the only way I think that we will see the Nforce in a Mac.

Interesting stuff... I went looking and found this http://www.eetimes.com/story/OEG20011008S0028 article for those interested in reading about it. Here is a link to the company itself: http://www.transitives.com/

Interesting point in the article :- CPU makers may be a bit worried. Also, if I could run OS X on my Athlon, why would I take the plunge and move my desktop to Mac (as well as my lappy).

Still, an interesting technology though...

-- Dan =)

cyberfunk
Jul 29, 2002, 05:30 PM
Well.. nForce 2 or not (not, methinks, but anyhow.. ),

how will we know, when the powermacs come out.. wether or not there is a DDR connect between the chipset and the CPU, not just the chipset and the memory ?


Someone mentioned this being a problem, as, in the current Xserves, the new macs might have DDR to the chipset, but only SDR to the CPU from the chipset.... doesnt this defeat the point of DDR in the first place ?

awrc
Jul 29, 2002, 05:34 PM
Originally posted by cyberfunk
Now, I understand what DDR is, SDRAM that talks to the chipset twice each cycle.. but what is DUAL DDR ???


Dual DDR isn't really a specific technology, it's a feature of the architecture of nVidia's nForce (and nForce 2) chipsets. Specifically, it provides two independent 64-bit DDR channels to memory.

In a system using the built-in graphics, one of these channels is decidated to the graphics processor (which uses system RAM) allowing it achieve full performance without stealing memory bandwidth from the processor (which still has a full-width memory bus of its own).

In fact, the speed of DDR meant that on the original nForce chipsets, the onboard GeForce 2MX could be considered to be running at AGP6X, because it actually had more memory bandwidth than an AGP4X socket could provide.

With the onboard graphics disabled, and a video card in the AGP slot, both of those memory channels are, in theory, available to the system. My old PC has an nForce-based motherboard and, until they upgraded the BIOS and motherboard drivers to allow the system to more flexibly configure the memory controllers, it was back to the days of paired memory banks, where having DIMMs in banks 0 and 1 gave half the bandwidth of having the same two DIMMs in banks 0 and 2.

EddieB
Jul 29, 2002, 05:57 PM
If it is true about nForce 2 not being able to have multiprocessors, I have a reason why this wouldn't matter to apple, but then I have a argument against following that:

Was it the 603e or the 604e(or both or something like that) that supported multiprocessors? Apple gave them up when they moved to the G3, so maybe they could give them up when moving from the G4 to the nForce2(I'm not sure if they will or wont).

But then again back in the days of the old multiprocessors there weren't that many Applications in the Mac market that took advantage of them(compared to now that is), and I'm not even sure if the Mac OS during that time did.

Rower_CPU
Jul 29, 2002, 06:21 PM
Originally posted by dukestreet
Well, seeing the pic again, I have to go with it not being a nForce chip - but no one has asked why the damn thing is 45° off the rest of the board? What benefit does this have? Weird.....

D

I've seen similar layouts on PC mobos...it's a method of shortening the traces between chipsets, etc....that's about all I know.:)

3G4N
Jul 29, 2002, 07:37 PM
Originally posted by EddieB
Was it the 603e or the 604e(or both or something like that) that supported multiprocessors? Apple gave them up when they moved to the G3, so maybe they could give them up when moving from the G4 to the nForce2(I'm not sure if they will or wont).

The 604e and G4 support multiprocessing.
The G3 *does not* support multiprocessing.
That is why Apple didn't make them,
not because they "gave them up".
Their markets are screaming for MP systems.


also posted by EddieB
But then again back in the days of the old multiprocessors there weren't that many Applications in the Mac market that took advantage of them(compared to now that is), and I'm not even sure if the Mac OS during that time did.

There were quite a few MP-saavy apps back in the PPC 60x
days, most notably Photoshop and After Effects. The MacOS
was never MP-saavy, not before MacOSX. That was a major
reason for the move, all so that you and I could have rock-solid
MP-Macs to rule the world, and finish our work so we can go
home early!!

And Rower_CPU is right. My buddy makes the SW that optimizes
those trace lengths in creating circuits. The same should go for mobos.

And finally, that's a G4. Tell me it has no resemblence to
http://www.apple.com/g4/images/processorphoto112192000.jpg
Yeah right.

ddtlm
Jul 29, 2002, 08:33 PM
First, for some background, the nForce2 as it currently stands is a chipset for AMD's Athlons that continains a GeForce4mx core and uses two "side by side" channels of DDR RAM for both system memory traffic and video memory traffic. Each RAM channel can run at 133x2, 166x2 or 200x2, but the DDR-400 RAM really isn't available. Performance of the similar nForce1 was sometimes as high as a low-end GF2MX, and so we could assume (based on other info as well) that the nForce2 will sometimes get as high as a slow GF4MX, but usually will not.

First major point: it does not offer revolutionary graphics or system performance compared to the way Apple currently has things set up. The dual-channel RAM does not offer an appreciable performance boost for Athlon systems.

Sub-point: Dual-channel double-data-rate is not at all the same things as a single-channel quad-data-rate, such as seen on the Pentium4 FSB. Since "QDR" RAM does not currently exist, there is actually nothing directly comparable to dual-channel DDR other than quad-channel SDR, which I think has been done on some small-market high-end systems. (Quad-channel SDR is actually the better of the two, but very hard to do.) And there are even more options, such as 128-bit wide memory channels that are like dual-channel but are "joined at the hip", but lets not go there.

Second major point: nForce1+2 both support a nomal AGP slot, which means they can use faster video cards no problem.

Third major point: Since the nForce2 is using the AMD-compatible FSB, it would need changes to work with G4's. It is the AMD system bus that does not allow dual CPU's... this means that Apple could still support dual G4's on a nForce2-for-Apple motherboard.

Sub-point: A dual CPU Athlon needs dual FSB's as well, unlike dual CPU P3's, Xeons' and G4's, where the CPUs share a FSB. Because the nForce only supports one FSB, it can only support one Athlon. The FSB is incompatible with P3's or P4's (although this could be changed, and nVidia could offer an P4 version along side the Athlon and hypothetical G4 versions).

Chryx
Jul 29, 2002, 09:02 PM
The dual-channel RAM does not offer an appreciable performance boost for Athlon systems.

That's because the link between the Athlon and the northbridge is limited to 2.1GB/s, so any memory bandwidth over that point is pretty much unusable by the processor.

that goes for dual channel DDR AND for PC2700 ram.

It is the AMD system bus that does not allow dual CPU's.

Almost but not quite, the EV6 (Alpha) bus the Athlon uses is point to point (eg, you need an extra bus for each processor.)

Dual processor Athlon's are quite possible, eg Tyan Thunder K7X (http://www.tyan.com/products/html/thunderk7x.html)

pilotgi
Jul 29, 2002, 10:25 PM
Finally at the end of this thread ddtlm and Chyrx get the facts straight.

I want to go on record here that I think in August, new powermacs will have an nForce2 chipset with HyperTransport. Even with no upgrade in processor speed this will create a much bigger pipeline, thereby increasing performance.

When this comes true, I'll direct everyone back to this post and say:

"I told you so."

Chryx
Jul 29, 2002, 10:30 PM
Originally posted by pilotgi
new powermacs will have an nForce2 chipset with HyperTransport. Even with no upgrade in processor speed this will create a much bigger pipeline, thereby increasing performance.

bigger pipeline to what, from what exactly?

Chryx
Jul 29, 2002, 10:40 PM
Originally posted by cyberfunk
while nForce (1) was a IGP (integrated graphics Proc) , nForce 2 can either be a SPP or a IGP, that is to say.. it can be offered just fine without the integrated graphics.

FYI, Nvidia's nForce 415 series is "Nforce1" without the onboard graphics...

(iow, both nforce 1 and 2 are complete chipsets)

And even the boards that DID have onboard video also had an AGP slot (might be exceptions that I'm unaware of though?)

Asus Nforce mainboard (http://www.asus.com/mb/socketa/a7n266-e/overview.htm)

BenRoethig
Jul 29, 2002, 10:52 PM
Originally posted by pilotgi
Finally at the end of this thread ddtlm and Chyrx get the facts straight.

I want to go on record here that I think in August, new powermacs will have an nForce2 chipset with HyperTransport. Even with no upgrade in processor speed this will create a much bigger pipeline, thereby increasing performance.

When this comes true, I'll direct everyone back to this post and say:

"I told you so."

I too think there's a chance the new Power Macs could use a modified variant of the nForce2 SPP/MCP-T chipset with a couple additons.
4 dimm slots
10/100/1000 ethernet (No DualNet)
various other Mac specific changes.

Scottgfx
Jul 29, 2002, 11:23 PM
Originally posted by Chryx


bigger pipeline to what, from what exactly?

Right now the biggest bottleneck is the bus between the processor and the memory. nForce with the Athlon provides two channels of DDR memory. I believe the idea is that the processor (with the nForce) can only use about half of the available bandwidth wile the left over bandwidth can be used by the "IGP" or integrated graphics processor. The lowly consumer nForce actually provides some features that put it above the similar vintage VIA and AMD chipsets. Also, the nForce motherboards sill provide a complete AGP slot that will bypass the integrated video. Some friends of mine have done this.

While I think is very possible that Apple may be working with AMD on a nForce type chipset for the Mac, I do not think that the pictures we have seen have any relevance to this issue. The positioning and angle of the chips is a function of the PCB CAD program. You might tell the CAD program what chips are where, what leads connect to what, and how many layers of PCB it has to route through. Placing the chip at a 45 degree angle could be either to shorten or neaten the routing of the paths or perhaps just to show off the fancy new chip on the board. :)

Chryx
Jul 29, 2002, 11:45 PM
Originally posted by Scottgfx
Right now the biggest bottleneck is the bus between the processor and the memory.

If you check out what I was replying to, I was referencing his mention of Hypertransport, now... Hypertransport isn't a very good idea for the main memory subsystem :)

rugby
Jul 30, 2002, 08:08 AM
Well, the bottleneck from the cpu to ram is significant, the bottleneck of ATA/133 is bigger. Master/slave combinations kill performance. Yeah sure we can have 2 drives on each ata bus, but the cpu is in use to determine which drive to use for what each and every time the drive is accessed. I'm surprised that serial-ata isn't on this chipset. It's the future, no more bus architecture for hard drives, all serial baby.

topicolo
Jul 30, 2002, 08:33 AM
You know, ATi is also making an IGP...

bobartig
Jul 30, 2002, 09:45 AM
If the speculation that Apple will use nforce technology is based on the fact that the nForce has a tilted northbridge, then it's full of bunk. MSI boards (http://www.newegg.com/app/Showimage.asp?image=13-130-035-01.JPG/13-130-035-02.JPG/13-130-035-03.JPG) use that sort of arrangement all the time, and most of their northbridges come from VIA. And, VIA is the largest controller chip company in the industry. The fact that nForce tilts theirs is to reduce trace length, and therefore signal interference between components. Nothing revolutionary, or nonstandard here, the industry has been doing it for years. I'm not saying apple won't use an nVidia mobo in the future (god only knows what they're up to anyhoo), but this is certainly no indication that they will.

Chryx
Jul 30, 2002, 10:42 AM
Originally posted by topicolo
You know, ATi is also making an IGP...

According to the benchmarks I've seen, it's not even competitive with the Nforce 1 >:(

pilotgi
Jul 30, 2002, 11:18 AM
From the nForce2 pdf:here (http://www.nvidia.com/docs/lo/2036/SUPP/nForce2_PO_71502v3.pdf)

"NForce2 processors offload processing tasks from the cpu to deliver unmatched system performance.

…and an optimized 128-bit architecture reducing overall system latency.

With its efficient memory design and support for 3GB of 400 MHz DDR memory, nForce2 eliminate system bottlenecks and speed up everyday media-rich applications"

Of course, all these specs are referring to nForce2 being used with AMD cpus,
but I'm guessing they've developed one for the mac platform too.

ddtlm
Jul 30, 2002, 11:22 AM
Chryx:

Of course I said that exact same thing about how AMD can run dual Athlons with dual FSB's in my post. :) I have built several such machines.

Scottgfx:

There are actually two busses between the processor and memory, the FSB and the RAM's bus (which is not called "Rambus" cause that's a freaking company name). CPU's with only a single bus between memory and themselves are those with on-die memory controllers, such as AMD's Hammer series and I think Moto's 85xx series.

But however there is NO performance edge offered by current dual-channel nForce1's over single-channel DDR chipsets, such as the VIA KT333. Without a FSB fast enough to use it all, there is really no point to dual-channel DDR except if you happen to be using the built-in graphics (which the entry-level G4 towers could reasonably do, although I'd rather see somethng faster).

rugby:

Eh, only if you are grinding on disk all day. I'd take more memory system bandwitth any day (still running 5400RPM disks!).

topicolo:

Yeah, it sucks just like all the other integrated video setups except nForce.

OVERALL, I'd have to say that this rumor is silly, the 45deg angle of the chipset is meaningless, the nForce2 is not generaly useful to Apple, the nForce2 needs changes to both FSB and network controllers to work for Apple, and that no DDR FSB G4 is known to exist at this time. THere is no reason to suspect that this chipset will be used by Apple, and I would be surprised if it were.

ffakr
Jul 30, 2002, 11:55 AM
Originally posted by awrc


In fact, the speed of DDR meant that on the original nForce chipsets, the onboard GeForce 2MX could be considered to be running at AGP6X, because it actually had more memory bandwidth than an AGP4X socket could provide.


No, it doesn't mean AGP 4X runs at AGP 6X speeds.

AGP nX refers to the the real speed of the bus in relation to 66MHz bus. 1X=66MHz, 2X= 133MHz, 4X=266MHz.
The AGP spec states that the AGP bus is 32bit wide. There is NO double wide AGP bus (remember a standard SDRAM/DDR bus is 64bits wide, an interlaced bus is 128 bits wide)
This has NOTHING to do with system memory. AGP 4X provides [266MHz x 32bit] 1066 MB/s maximum through put.

AGP is its OWN bus. It doesn't matter if the main memory bus uses a ga-jillion way dimm setup and it has a petabyte/sec of throughput... AGP 4x is still AGP 4x.

Please, don't make stuff up.

ffakr
Jul 30, 2002, 12:10 PM
Originally posted by pilotgi
Finally at the end of this thread ddtlm and Chyrx get the facts straight.

I want to go on record here that I think in August, new powermacs will have an nForce2 chipset with HyperTransport. Even with no upgrade in processor speed this will create a much bigger pipeline, thereby increasing performance.

When this comes true, I'll direct everyone back to this post and say:

"I told you so."

haha, not in a million billion years.

You have no idea what nVidia would need to do in order to re-engineer an x86 chipset with an athlon bus in order to make it work with a Mac. Here is the short list....

It's the wrong endian
It is designed to issue x86 instructions! (how do you throw an x86 interrupt to a PPC processor?)
It is designed to interface with the wrong processor bus.
It doesn't know what the hell Open Firmware is.
It isn't designed to bootstrap from Mac ROM.

It would be (no joke) about as easy to make a nForce2 mac as it would be to make a PPC Athlon. Realistically, some of these issues are easer to tackle than others, but together they are a show stopper.

IF (and this is a HUGE if), nVidia makes a PPC compatable chipset, it would only be architecturally similar to nForce... it would have to be considered a different, new chipset.

ffakr
Jul 30, 2002, 12:14 PM
Originally posted by rugby
Well, the bottleneck from the cpu to ram is significant, the bottleneck of ATA/133 is bigger. .

NO, ATA/133 is NOT a bottleneck. The ONLY exception to this is master/slave which isn't really a bottleneck issue, but rather a design problem.

ATA/133 has 133MBYTES/sec of throughput. The fastest 7200rpm IDE drives don't sustain transfers much over 35MB/sec.
Current ATA standards are NOT a bottleneck... they have significantly more bandwidth than any one or two drives can even burst out.

The real issue with IDE is the poor design of the master/slave model... but this is easily overcome by adding channels.

Chryx
Jul 30, 2002, 12:16 PM
Originally posted by ffakr


No, it doesn't mean AGP 4X runs at AGP 6X speeds.

AGP nX refers to the the real speed of the bus in relation to 66MHz bus. 1X=66MHz, 2X= 133MHz, 4X=266MHz.
The AGP spec states that the AGP bus is 32bit wide. There is NO double wide AGP bus (remember a standard SDRAM/DDR bus is 64bits wide, an interlaced bus is 128 bits wide)
This has NOTHING to do with system memory. AGP 4X provides [266MHz x 32bit] 1066 MB/s maximum through put.

AGP is its OWN bus. It doesn't matter if the main memory bus uses a ga-jillion way dimm setup and it has a petabyte/sec of throughput... AGP 4x is still AGP 4x.

Please, don't make stuff up.

You're forgetting, the ONBOARD gf2MX on the Nforce is plugged straight into the "system controller" rather than being on the other side of the AGP bridge....

According to Nvidia it's theoretically equivalent to AGP @ 6x

'tis something of a moot point if there isn't memory bandwidth free for it however.

Chryx
Jul 30, 2002, 12:18 PM
Originally posted by ffakr

It is designed to interface with the wrong processor bus.


Just a heads up, we already know that the NForce is fairly modular in this regard, as it's been seen with both GTL+ (Xboxs use a variant of the Nforce chipset with meatier graphics hardware and a Pentium 3) and EV6 buses (Athlon/desktop Nforce)

Oh

The fastest 7200rpm IDE drives don't sustain transfers much over 35MB/sec.

Heads up number 2, Western Digital WD1200JB's can push about 50MB/s sustained.

ffakr
Jul 30, 2002, 12:20 PM
Originally posted by arn
The Inquirer (http://www.theinquirer.net/?article=4663) speculates that chip positioning on the leaked PowerMac Motherboard may hint at an NVidia NForce 2 connection with the new PowerMacs:

The one chip in recent years to buck the trend is Nvidia's nForce northbridge, positioned at 45 degrees to a mobo's main axes. Alas the blurry snapshots of the alleged Apple product make the info printed on top of the part unreadable, but its positioning strongly suggests the inclusion of Nvidia technology.


Geez, please stop posting this crap as likely rumor. This is totally wishful thinking...

Chip position doesn't mean anything. Not one hairy thing. That statement automatically qualifies the original rumor monger as a moron.

Second problem with the rumor...
nForce DOESN'T USE A NORTHBRIDGE!!!!!
The nForce chipset (and nForce2) doesn't use a traditional northbridge/southbridge design (neither do the newer intel p4 chipsets). Anyone who knew anything about the nForce would have known this.

The rumor is total, unmitigated, crap.

Please kill this line of rumors already. Apple already has a DDR chipset in production. I'll eat my hat if they don't use the same chipset in the next Mac towers.

Chryx
Jul 30, 2002, 12:26 PM
[QUOTE]Originally posted by ffakr
nForce DOESN'T USE A NORTHBRIDGE!!!!!
The nForce chipset (and nForce2) doesn't use a traditional northbridge/southbridge design


Indeed, I still tend to think of the piece of silicon with the AGP bridge and memory controller in it as "the northbridge" though...

Old habits die hard and all that.

ffakr
Jul 30, 2002, 12:41 PM
Originally posted by Chryx

The fastest 7200rpm IDE drives don't sustain transfers much over 35MB/sec.

Heads up number 2, Western Digital WD1200JB's can push about 50MB/s sustained. [/B]

If we want to pick nits:
pcextreme (http://www.pcextreme.net/wd1200jb.php) reviewed the drive you mentioned.

First off, this is a special edition drive and it is currently (to the best of my knowledge) the fastest IDE drive available... aside from the $20k solid state devices.

Second, here are the benchmarks...
sequential read, 47MB/sec
random read, 8MB/sec

The fastest transfer possible is 82MB/s which is a burst from the whopping 8MB of on drive cache.

Even 2 of these drives couldn't use up ATA/133 bandwidth on a sustained read... even if all the data was on one contiguous block. That would never happen though, because the IDE won't let you access a master and slave at the same time.

Even the best IDE drive available can only burst about 2/3s of ATA/133s bandwidth. I think my assertion still stands. This drive can sustain less than 1/2 of ATA/133s bandwidth with a totally contiguous block of data (no seeking).

topicolo
Jul 30, 2002, 01:07 PM
Not if you have an IDE RAID 0/1 device set up striping all that data

Chryx
Jul 30, 2002, 01:07 PM
Originally posted by ffakr

Even the best IDE drive available can only burst about 2/3s of ATA/133s bandwidth. I think my assertion still stands. This drive can sustain less than 1/2 of ATA/133s bandwidth with a totally contiguous block of data (no seeking).

Indeed, I wasn't picking that nit :)

Joshlew
Jul 30, 2002, 02:25 PM
nForce would do a lot for the speed of the next macs..................

Chryx
Jul 30, 2002, 02:57 PM
Originally posted by Joshlew
nForce would do a lot for the speed of the next macs..................

I don't think it would make THAT big of a performance difference, I would however like to see apple using something like the MCP on the Nforce chipset, 5.1 dolby encoding in realtime across the range?.. damn straight!

sturm375
Jul 30, 2002, 04:59 PM
Originally posted by topicolo
Not if you have an IDE RAID 0/1 device set up striping all that data

RAID is excellent for writing, but horrid for reading. So you would be able to Write large file very fast, and safely, however reading them back would take longer. I thing the Screen Savers posted some benchmarking scores, and found that on average, RAID is slower than standard Master/Slave. The only reason to use RAID, is for the security of redundancy.

Also on IDE RAID, you still can't write, or read from more than one disk at a time. That is part of IDE.

Chryx
Jul 30, 2002, 05:05 PM
Originally posted by sturm375


RAID is excellent for writing, but horrid for reading. So you would be able to Write large file very fast, and safely, however reading them back would take longer. I thing the Screen Savers posted some benchmarking scores, and found that on average, RAID is slower than standard Master/Slave. The only reason to use RAID, is for the security of redundancy.


There are different kinds of RAID, the two found most often are:

RAID 0 : Striping
literally slices the data into chunks and simultaniously writes/reads from all disks (provides a pretty much 100% boost to read/write speeds compared to a lone drive)

RAID 1 : Mirroring
maintains a live hardware backup, some controllers interleave the reads (gives a RAID0 like read performance increase)

(there's also things like RAID 0+1, which is a stripeset with a live mirror)


I would imagine the tests you saw were for RAID-5?

ktlx
Jul 30, 2002, 09:00 PM
Originally posted by sturm375

RAID is excellent for writing, but horrid for reading. So you would be able to Write large file very fast, and safely, however reading them back would take longer. I thing the Screen Savers posted some benchmarking scores, and found that on average, RAID is slower than standard Master/Slave. The only reason to use RAID, is for the security of redundancy.


There are various types of RAID, each for a different application. Here is a run down of the most common types and where they are good and where they are not so good:

RAID 0: This is often called striping and consists of chaining multiple drives together to make a single larger volume. If the data is spread across multiple drives (hence the term stripe) then read and write performance is very good. It approaches the speed of an individual drive times the number of drives in the volume. However if you loose a single drive, the whole volume is toast. So you pay a huge reliability risk because the chance of a failure is the MTBF of each drive divided by the number of drives in the volume. One area where this can be useful is in intermediate processing of things like video. You need the intermediate stages to work as fast as possible, but their life span is relatively short so you can risk a disk failure.

RAID 1: This is often called mirroring and consists of having one volume exactly mirror another volume. There is no speed increase but the reliability goes up. It is now the MTBF of both drives added together. If the data is read and written in a very sequential nature, this model can be very fast. However, the main reason to go with mirroring is redundancy.

RAID 01, 10, 0+1, etc. All these are names for the same basic idea. First you create two volumes using a RAID 0 stripe. Then you mirror the two volumes as you would two disks. Now you have the read performance of RAID 0 combined with the redundancy of RAID 1. However this is very expensive as you have to buy twice as much disk space. But if you have disk needs that have lots of reads and writes and must be able to withstand disk failures, this is about all you can pick.

RAID 5: You can think of this as RAID 0 with parity information. Additional data is written to the volume to allow the volume to continue to function if a single disk failure occurs. Since the parity information has to be written on each disk write, writes pay a significant performance penalty. You almost always have to use a RAID controller to get any kind of throughput. However, if no disks in the volume have failed, you do not need to use the parity information, so reads are as fast as RAID 0. RAID 5 is great if you have data that is written very infrequently but read quite a bit. You get the reliability of mirroring without the cost of RAID 0+1.


Also on IDE RAID, you still can't write, or read from more than one disk at a time. That is part of IDE.

This completely depends upon the implementation. Yes, if you have two drives in the RAID set and both are on the same channel, this is true. However IDE RAID controllers typically have multiple channels (between two and eight) so that you put each drive on its own channel. Once you do this, then you can write to multiple drives at the same time. I believe the thinking is that if you need more than four to eight drives in the volume, you really should be using some other means to access the drives (i.e. SCSI or Fiber Channel).

Scottgfx
Jul 31, 2002, 01:33 AM
Originally posted by ddtlm
Chryx:
Scottgfx:

There are actually two busses between the processor and memory, the FSB and the RAM's bus (which is not called "Rambus" cause that's a freaking company name). CPU's with only a single bus between memory and themselves are those with on-die memory controllers, such as AMD's Hammer series and I think Moto's 85xx series.

But however there is NO performance edge offered by current dual-channel nForce1's over single-channel DDR chipsets, such as the VIA KT333.

OVERALL, I'd have to say that this rumor is silly, the 45deg angle of the chipset is meaningless, the nForce2 is not generaly useful to Apple, the nForce2 needs changes to both FSB and network controllers to work for Apple, and that no DDR FSB G4 is known to exist at this time. THere is no reason to suspect that this chipset will be used by Apple, and I would be surprised if it were.

Understood, I've been away from my PCs for a few months, and forgot that Hypertransport only went between the north and southbridges.

I too think it's silly about the 45 degree angle issue. I've seen either network cards or modems with angled chips. Perhaps it has something to do with "alien" technology being used? :)