AOA Forums AOA Forums AOA Forums Folding For Team 45 AOA Files Home Front Page Become an AOA Subscriber! UserCP Calendar Memberlist FAQ Search Forum Home


Go Back   AOA Forums > Hardware > AMD Motherboards & CPUs

AMD Motherboards & CPUs Questions or comments on AMD products?


Reply
 
LinkBack Thread Tools Rate Thread
  #1 (permalink)  
Old 13th December, 2002, 10:10 PM
Chief Systems Administrator
 
Join Date: September 2001
Location: Europe
Posts: 13,075

ARGH! 8RDA has too many clocks!

Lets really examine this nForce 2 board, the EPoX 8RDA+.

We know that the SPP (Northbridge) and MCP-T (Southbridge) are connected together via HyperTransport, rather than the more traditional PCI. If you look closely at the HyperTransport bus in the nForce 2 implementation, it operates at 800MB/sec. We also know that is a point-to-point bus, meaning that it only connects two devices together. It's also packetised, so instead of handling individual bytes, it handles packets of data, rather like most PC parts. (Hard disk, Ethernet, SCSI, floppy and so on). This is partly why the nForce can provide a feature called "StreamThrough". "StreamThrough" just means that the HyperTransport bus can guarantee dedicated bandwidth for particular data transfers. This means that the packets of data containing video information coming from the network connector will always be able to travel across the bus when they need to. They don't get pushed around by other devices such as the hard disks, or USB devices. Cool huh?

In addition, the memory can run asynchronously from the FSB. All this is important! Running things asynchronously means that each device run asynchronously needs it's own clock to keep it's own timing.

Before, when everything was running synchronously, there was a master clock that regulated everything. Each different frequency required was simply a division of the master clock. In one fell swoop, all this has changed!

The FSB and memory bus can run asynchonously, so each must have a separate clock. We also know that the AGP slot can run at a fixed clock - this is asynchronous, so it too needs it's own clock.

So far, we have 3 clocks, all independent of each other, and we haven't looked outside the SPP yet! Remember that HyperTransport bus? Well, that needs it's own clock too to regulate the transfers across it. In fact, for every 8 bits, HyperTransport needs a seperate clock. An 8bit wide HT bus needs one clock, a 16bit wide HT bus needs two clock, and a 32bit wide HT bus needs 4 clocks! nVidia's implementation only uses 8 bit wide transfers, so it only needs a single clock!

We're up at 4 clocks now, and none of them are generated from the same source.

Lets move down to the southbridge. This is where information gets a little more sketchy. We already know from the SPP that there's no master clock. Where does the SPP derive it's clocking from then? It would make little sense to derive it from the FSB, as the MCP-T doesn't connect to the FSB, nor is the MCP-T's connection to the SPP regulated by the FSB. It would make more sense to derive the MCP-T's clocking from the HyperTransport bus, as there is only one clock in this case. If the nForce2 chipset had a wider HyperTransport, then it would be more difficult, as there would be more clocks for the HyperTransport bus. An HT bus clock runs at 100MHz

Let's look at what needs a clock on the MCP-T. Hmm, this could be a nice place for a list...
  • Dual IDE interface (266) ?
  • PCI bridge (133) (33)
  • Audio Processing Unit (150) ?
  • Audio Codec Interface (1) ?
  • Ethernet (25) (20)
  • LPC Bus (Low Pin Count) - The remnants of the ISA bus. (1)
  • Dual USB (50) (20)

Remember, ALL of these are integrated onto the MCP-T. The first number in brackets illustrate the approximate bandwidth needed to run the various parts flat out. The second number illustrates the clock rate needed.

With each of these devices needing different speed clocks, it makes little sense to tie them all together into the non-existent clock generator. Remember that the SPP doesn't need the clock generator, as it has 4 separate clock generators in silicon just to handle it's own interfaces and keep things running asynchronously.

A lack of clock generator, and a radically different architecture to the standard Northbridge/Southbridge system calls for a lot of differences. This includes the MCP-T having it's own clock generator that runs independently of the SPP. Independence means that the audio, PCI, IDE and Ethernet all run in spec all the time. There's no question about PCI locks or an IDE lock, as that would require a complete redesign of the nForce2 chipset! Items like Ethernet and Audio are highly clockspeed sensitive. If you changed the clock for the Ethernet, your machine could no longer communicate with other devices on the network. The audio should be fairly obvious what would happen!

All in all, this points to all the peripherals running in spec regardless of what the FSB, Memory bus or AGP bus do.

As far as items like IDE disk corruption go, there's more to this than meets the eye. People are already beginning to trace some of this strange corruption through to memory, rather than hard disk. Why this should be is odd, but may be related to the caching all modern OSes use. Modern OSes rely heavily on data cached in main memory. This cache can be a sizable chunk of memory (On my 512MB W2K system, the cache is nearly 200MB in size!). Any slight corruption of the cache will cause nasty side effects as this corrupt data gets used. As this data usually includes the various control structures on the disk (FAT for FAT, MFT for NTFS, I-Nodes for EXT2FS and FNodes for HPFS), this can be very dangerous. Couple this with the fact that the control structures cached will be accessed EVERY time the OS wants to do something with the hard disk, and these become heavily used parts of memory.

Phew.. Hopefully that'll give a bit of a better idea about why the nForce2 architecture is so radiacally different from VIA and other's implementations. This means that things you used to understand may be totally different! Fortunately, as time goes on, people will understand the nForce2 better and better.

Áedán
__________________
Any views, thoughts and opinions are entirely my own. They don't necessarily represent those of my employer (BlackBerry).

Last edited by Áedán; 13th December, 2002 at 10:14 PM.
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #2 (permalink)  
Old 13th December, 2002, 10:28 PM
Daniel ~'s Avatar
Chief BBS Administrator
 
Join Date: September 2001
Location: Seattle Wa.
Posts: 45,606

Front Page stuff! Which is where I've put it! ":O}
__________________
"Though all men live in ignorance before mystery,
they need not live in darkness...
Justice is foundation and Mercy ETERNAL
."
DKE

"All that we do is touched by Ocean
Yet we remain on the shore of what we know."
Richard Wilbur

[img]/forum/attachments/random-nonsense/16515-sigs-dan_drag.jpg[/img]
Subscribers! Ask Pitch about a Custom Sig Graphic

Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #3 (permalink)  
Old 13th December, 2002, 10:35 PM
EluSiOn's Avatar
Member
 
Join Date: November 2002
Location: USA
Posts: 549

Áedán,

How long you have done the research and reading the white paper then came up with this post? I am just curious.

Very useful information and now I understand NForce 2 more.
__________________
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #4 (permalink)  
Old 13th December, 2002, 11:13 PM
Holst's Avatar
Member
 
Join Date: September 2001
Location: Leics UK
Posts: 4,528

Excelent.

Nice to see somebody applying some logic to the question.
__________________
No longer Epox Tech.

Best of luck in the future all my friends.
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #5 (permalink)  
Old 13th December, 2002, 11:18 PM
Chief Systems Administrator
 
Join Date: September 2001
Location: Europe
Posts: 13,075

Quote:
Originally posted by EluSiOn
Áedán,

How long you have done the research and reading the white paper then came up with this post? I am just curious.

Very useful information and now I understand NForce 2 more.
Far too long! (Just doing the writing part of it took several hours!) There's actually a fair chunk of information over at nVidia's site, in their various papers, but it's not always as obvious as it should be. There's also plenty of stuff about HyperTransport over at AMD site, which should come as no surprise really. There's a few other sources for the information, which I should have noted, but didn't! (I'm sorry! I'm no good with citations!)

One of the things that motivated me to write the thing in the first place was the amount of confusion over how the nForce2 works internally. Most people here don't have a great understanding of how various things like busses and the like work, never mind the whole concept of asynchronous clocks and how they're implemented. On the other hand, I wouldn't expect people without some kind of electronics background to jump in and understand this stuff first hand. It's also not helped by a bunch of websites who claim to give indepth information, and fail because their writer hasn't grasped certain points.

I'm gonna stop here, otherwise I'm going to turn it into a mini paper in it's own right...

Áedán

(You should see my bookshelf sometime. It's mostly technical stuff!)
__________________
Any views, thoughts and opinions are entirely my own. They don't necessarily represent those of my employer (BlackBerry).

Last edited by Áedán; 13th December, 2002 at 11:20 PM.
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #6 (permalink)  
Old 13th December, 2002, 11:43 PM
mookydooky's Avatar
AOA Staff
 
Join Date: December 2001
Location: Everywhere you want to be.
Posts: 7,903
Send a message via MSN to mookydooky

You know Aedan, if you were gonna do this, you could have let me know and saved me a couple of hours of reading...
__________________
Need a laugh? Take a peek at
AOA's Mookydooky's "Just for laughs!" Joke Topic


AOA Team fah

我喜欢大屁股, 我不能骗你..... 他们弟兄不能否认......
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #7 (permalink)  
Old 14th December, 2002, 12:19 AM
Member
 
Join Date: October 2002
Posts: 101

Quote:
Originally posted by mookydooky
You know Aedan, if you were gonna do this, you could have let me know and saved me a couple of hours of reading...
at least you were willing to research it for yourself. you did your own reading instead of being spoon fed like the other people who just made accusations and statements based on incomplete information, while not being willing to actually research the subject

well done Aedan
__________________
2100xp 200x11.5 aiuhb 0204
swiftech mc462a rev1
epox 8rda+
512mb corsair xms3200 cl2 (2 256 sticks) 5-2-2-2
40gb 7200rpm wd 8mb cache
9700 pro 3.1 based omegas
56x cdrom
24x10x40 cdrw
winxp pro
19" trinitron
600w ps
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #8 (permalink)  
Old 14th December, 2002, 01:10 AM
kowalski's Avatar
Member
 
Join Date: October 2002
Location: london
Posts: 202

Re: ARGH! 8RDA has too many clocks!

Quote:
Originally posted by Áedán
Lets really examine this nForce 2 board, the EPoX 8RDA+.

**SNIP**

Áedán
i think perhaps nVidia got the concepts for this system of completely independant clocks from their graphics card architecture. i've always found it interesting that the GPU and RAM on the cards can be clocked totally separately in a way that's totally different to the traditional mobo divider async approach we've been used to.

it wouldn't have taken a genius at the company to say one day "hang on, let's apply this philosophy to mobo architecture. i think it may like, rock!" (would take quite a bright spark to implement it though )

and ... ker-ching...!
__________________
Epox 8K5A3+ | XP1700+ AROIA-Y @1851MHz (10x185) | Thermalright SLK-800 + SmartFan II | 512MB Samsung PC2700 CAS2 3-6-3 1T | PNY GeForce 4 Ti4600 @333/742 | Western Digital 120GB Special Edition | 431W Enermax PSU | CoolerMaster ATC-201 case | Iiyama 21" VisionMaster Pro | Win2k sp2

13,659 3DMarks
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #9 (permalink)  
Old 14th December, 2002, 01:13 AM
EluSiOn's Avatar
Member
 
Join Date: November 2002
Location: USA
Posts: 549

mookydooky,

Sometime it is good to be fed info like this... but sometime it is good to read yourself... LOL.... I know both you and Áedán spend alot of times reading Nvidia NForce2 white papers. All the hours you two has put in is not gonna be wasted but benefits me alot and also many others. It is always good for people start finding out the fact on their own instead of, I know this and that base on someone else statement or finding. It creates too many speculations that way. I have copy and paste this post and created it as a PDF. It will be inside my Guides directory where keep many good documentation on various subject.

Áedán,

I don't have bookshelf, I hardly pick up any books other than programming books, but I collect PDFs. I found reading easier on computer or using the text to speech applications. LoL, I am lazy bum and thanks for this info fed.
__________________
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #10 (permalink)  
Old 14th December, 2002, 03:05 AM
popcicle's Avatar
Member
 
Join Date: November 2002
Location: Miles From Civilization
Posts: 113

Primo


Pops.
__________________
....it ain't over 'till the Fat Lady sings....

Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #11 (permalink)  
Old 14th December, 2002, 05:23 AM
Daniel ~'s Avatar
Chief BBS Administrator
 
Join Date: September 2001
Location: Seattle Wa.
Posts: 45,606

Re: Re: ARGH! 8RDA has too many clocks!

Quote:
Originally posted by kowalski


i think perhaps nVidia got the concepts for this system of completely independant clocks from their graphics card architecture. i've always found it interesting that the GPU and RAM on the cards can be clocked totally separately in a way that's totally different to the traditional mobo divider async approach we've been used to.

it wouldn't have taken a genius at the company to say one day "hang on, let's apply this philosophy to mobo architecture. i think it may like, rock!" (would take quite a bright spark to implement it though )

and ... ker-ching...!
Wouldn't be at all surprised if your right.
__________________
"Though all men live in ignorance before mystery,
they need not live in darkness...
Justice is foundation and Mercy ETERNAL
."
DKE

"All that we do is touched by Ocean
Yet we remain on the shore of what we know."
Richard Wilbur

[img]/forum/attachments/random-nonsense/16515-sigs-dan_drag.jpg[/img]
Subscribers! Ask Pitch about a Custom Sig Graphic

Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #12 (permalink)  
Old 14th December, 2002, 02:07 PM
Chief Systems Administrator
 
Join Date: September 2001
Location: Europe
Posts: 13,075

Re: Re: ARGH! 8RDA has too many clocks!

Quote:
Originally posted by kowalski
i think perhaps nVidia got the concepts for this system of completely independant clocks from their graphics card architecture. i've always found it interesting that the GPU and RAM on the cards can be clocked totally separately in a way that's totally different to the traditional mobo divider async approach we've been used to.
Given they've managed to keep the core and RAM seperate (ie asynchronous), I would be surprised if they didn't do exactly the same thing with the memory controller on the SPP. It certainly appears to work fairly well on the gfx cards

P.S. clock dividers are not async - they're sync because they're based on a common high speed clock. There's a known correlation between the various clocks.

Áedan
__________________
Any views, thoughts and opinions are entirely my own. They don't necessarily represent those of my employer (BlackBerry).
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #13 (permalink)  
Old 14th December, 2002, 02:24 PM
kowalski's Avatar
Member
 
Join Date: October 2002
Location: london
Posts: 202

Re: Re: Re: ARGH! 8RDA has too many clocks!

Quote:
Originally posted by Áedán
P.S. clock dividers are not async - they're sync because they're based on a common high speed clock. There's a known correlation between the various clocks.
yeah you're right, that's what i meant . the 'traditional' way of basing them all on the same clockgen at different ratios as opposed to totally free.

surely it helps though to have things running off a common source though? it can't be good for timing purposes to have high frequency streams converging, offset to each other by uneven fractions of waveforms - like meshing two cogs running at different speeds without any gearing. or is this the common misconception they've managed to overcome? they must have some device at each bus interface that retranslates the data stream onto the other bus...
__________________
Epox 8K5A3+ | XP1700+ AROIA-Y @1851MHz (10x185) | Thermalright SLK-800 + SmartFan II | 512MB Samsung PC2700 CAS2 3-6-3 1T | PNY GeForce 4 Ti4600 @333/742 | Western Digital 120GB Special Edition | 431W Enermax PSU | CoolerMaster ATC-201 case | Iiyama 21" VisionMaster Pro | Win2k sp2

13,659 3DMarks
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #14 (permalink)  
Old 14th December, 2002, 03:59 PM
Chief Systems Administrator
 
Join Date: September 2001
Location: Europe
Posts: 13,075

Quote:
Originally posted by kowalski
surely it helps though to have things running off a common source though? it can't be good for timing purposes to have high frequency streams converging, offset to each other by uneven fractions of waveforms - like meshing two cogs running at different speeds without any gearing. or is this the common misconception they've managed to overcome? they must have some device at each bus interface that retranslates the data stream onto the other bus...
You have to ask yourself where the problem might lie. HyperTransport is good for 800MB/sec. The various devices on the HT bus don't add up to 800MB/sec, so the MCP-T is free from bottlenecks. No problems in the MCP-T then.

Once we get to the SPP, this is where all the high throughput stuff meets. We have the FSB, the memory bus, AGP and the other end of the HT bus.

The memory is the most critical item here. If one device writes to memory, and another reads the same bit of memory, the nForce2 chipset needs to ensure that the second device gets the changed data.

In terms of handling the difference between buses, a device called a FIFO or BiFIFO is useful. It's basically a small bit of memory with two ports. Stuff that goes in one port, comes out the other in the same order, and vise versa. By using this to handle the difference, small variations can be buffered, just the same way an MP3 player buffers a stream played across a network.

For bigger differences, some form of arbitration is needed. PC already have arbitration logic anyhow, due to the possibility of other devices accessing main memory.

If the AGP card is accessing memory, and the CPU wants access, then the CPU has to wait until the block of data the AGP card wanted has been sent. FIFOs can help if it's a write accces, as they can buffer a small block and deliver it when the receiving device is free.

Arbitration usually has some form of priority as well as guards against devices taking too long. The PCI bus does something similar with guard times. If a PCI bus master is on the bus too long, the PCI timer will eventually time out on the device, and it will have to go back and request the bus again. In the mean time, another device may have managed to get access.

I've no idea how this is implemented in nForce2, and I suspect if I had access to such details, I would have to sign an NDA, meaning I couldn't tell anyone!

Áedán
__________________
Any views, thoughts and opinions are entirely my own. They don't necessarily represent those of my employer (BlackBerry).
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #15 (permalink)  
Old 14th December, 2002, 05:17 PM
Daniel ~'s Avatar
Chief BBS Administrator
 
Join Date: September 2001
Location: Seattle Wa.
Posts: 45,606

Áedán, how about encapsulating some of this into a "Part two " for the front page?... You don't have to if your afraid! ":O}
__________________
"Though all men live in ignorance before mystery,
they need not live in darkness...
Justice is foundation and Mercy ETERNAL
."
DKE

"All that we do is touched by Ocean
Yet we remain on the shore of what we know."
Richard Wilbur

[img]/forum/attachments/random-nonsense/16515-sigs-dan_drag.jpg[/img]
Subscribers! Ask Pitch about a Custom Sig Graphic

Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #16 (permalink)  
Old 14th December, 2002, 07:36 PM
cloasters's Avatar
Asst. BBS Administrator
 
Join Date: September 2001
Location: Location, Location
Posts: 21,956

Thank you many times for writing and posting this superb explanation of the formerly confusing nForce2 chipset, Aedan!

I had a look at some newspapers to see their reaction to Aedan's post,

Financial Times: Dont' miss it!

Le Figaro: Another reason to hate the English speaking world!

Asahi Shimbun: Wow. It's a must see!

The Straits Times: A smash hit in any language.

Variety: Aedan, boffo on the tech front!

The New York Times: New Tech Star Born.

Not bad, eh?
__________________
When the world will be better.
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #17 (permalink)  
Old 14th December, 2002, 07:43 PM
Daniel ~'s Avatar
Chief BBS Administrator
 
Join Date: September 2001
Location: Seattle Wa.
Posts: 45,606

Yeah, but what did the big guns have to say? Did he for example get a mention on AOA? LOL
__________________
"Though all men live in ignorance before mystery,
they need not live in darkness...
Justice is foundation and Mercy ETERNAL
."
DKE

"All that we do is touched by Ocean
Yet we remain on the shore of what we know."
Richard Wilbur

[img]/forum/attachments/random-nonsense/16515-sigs-dan_drag.jpg[/img]
Subscribers! Ask Pitch about a Custom Sig Graphic

Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #18 (permalink)  
Old 16th December, 2002, 03:22 AM
monkeynuggets's Avatar
Member
 
Join Date: December 2002
Location: missouri
Posts: 6

Question

Well this is really good news, but it conflicts with other info from a supposedly reputable source saying that the MSI and Leadtek nforce2 boards do not have the pci locked. but it looks like you dug deeper than they ever did, i hope you are right
__________________
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #19 (permalink)  
Old 16th December, 2002, 05:07 AM
Daniel ~'s Avatar
Chief BBS Administrator
 
Join Date: September 2001
Location: Seattle Wa.
Posts: 45,606

Aedan speaks for AOA.
__________________
"Though all men live in ignorance before mystery,
they need not live in darkness...
Justice is foundation and Mercy ETERNAL
."
DKE

"All that we do is touched by Ocean
Yet we remain on the shore of what we know."
Richard Wilbur

[img]/forum/attachments/random-nonsense/16515-sigs-dan_drag.jpg[/img]
Subscribers! Ask Pitch about a Custom Sig Graphic

Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #20 (permalink)  
Old 16th December, 2002, 05:13 AM
monkeynuggets's Avatar
Member
 
Join Date: December 2002
Location: missouri
Posts: 6

mmk wutever that means
__________________
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
Reply



Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)
 
Thread Tools
Rate This Thread
Rate This Thread:

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is Off
HTML code is Off
Trackbacks are On
Pingbacks are On
Refbacks are On


Similar Threads
Thread Thread Starter Forum Replies Last Post
Unnusual 7600GS clocks? Arkanius Graphics and Sound cards; Speakers and other Peripherals 11 3rd May, 2007 12:15 PM
GPU clocks madcatmk3 General Hardware Discussion 0 11th November, 2005 04:18 PM
8RDA3+ and Barton XP-M 2600+ clocks Novis_cz AMD Motherboards & CPUs 18 24th August, 2004 02:17 PM
2500+ Clocks.. Gotenks AMD Motherboards & CPUs 4 31st July, 2004 11:28 AM
8RDA+ 2400+.. argh, too many pluses Parak AMD Motherboards & CPUs 6 2nd December, 2002 08:39 PM


All times are GMT +1. The time now is 11:19 AM.


Copyright ©2001 - 2010, AOA Forums
Don't Click Here Don't Click Here Either

Search Engine Friendly URLs by vBSEO 3.3.0