1st Hammer benchmarks, impressive!!

[Deviant]

This is just an early release Claw Hammer, so later versions should be more powerful, and have a higher clock speed. Also the Opteron is supposed to be better than this!

It's shows Hammer is 40% faster clock for clock than an Athlon XP, also a 800MHz Hammer beats a 1667 MHz P4 Willy.

These results look impressive. More here:
http://www.aceshardware.com/

[I.R]

Hehe AMD locked them at 800Mhz so partners couldn't OC them and get our hopes up :p..... hope this isn't an indication of things to come :eek:

[Deviant]

I read somwhere that Jerry Sanders (AMD CEO) said they wouldn't mak it impossible. Just enough to keep the masses from doing it.

Duron & T-Bird were to easy, but maybe they think XP has achieved the correct balance. I hope so, otherwise FSB overclocking only.

[_N_]

well, as long as they can make the future boards run at 33PCI while the FSB is whatever it is, eg the Abit BD7?? I think, we'll be OK.

they are making the OCing harder mainly cause before some shops would buy like a duron 700, unlock it, and sell it in a computer as an 800. and then ppl would say AMD's suck cause they are not stable.

[Solid Snake]

Yeah ... I heard about that. That would suck but I can understand AMD doing that.

[Volodkovich]

looks nice....real nice. Heh, thats only a 256kb cache one aswell...imagine this chip at 2.5 - 3ghz it will be released in...i imagine it whooping Intel ass, aswell as pulling AMDs stock up...I mean, like everyone is like "wait for hammer",i feel amd are really gonna start leading the market with this chip

[Binky Stunt Cat]

http://www.tecchannel.de/hardware/937/
german site, but has an english version of the article...

[kaz]

mmmm


sorry to say this but AMD R FOOOOOLSS

they r comparing 64bit chips with 32bit chips

wait til intel brings out there 64bit chips..... :D

[mird-OC]

firstly, the article mentioned has not been approved and is not endorsed by AMD.

secondly, if you read the article you'd notice this they said - "The test does not give any results about a performance with 64-bit software." - meaning they didn't even get to flex the 64-bit powers of the chip, which means they were comparing 32-bit against 32-bit... a fair comparison.

and thirdly, intel have already brought out 64-bit chips (for the high-end server market), and they were a big flop.

[Binky Stunt Cat]

and whats more, AMD have a lead over Intel in a 32 and 64 bit compatible chip for the transition.
the only reason Intel is doing so is coz AMD was in the first place.
Personaly i might just get a Hammer system in 2004, hopefully DDR II will be out by then.

[CrazySurfaNZ]

Hell Yeah!! Go amd, and good comment mird :) AMD rule... dont knock it till youve tried it :)

[_N_]

>>dont knock it till youve tried it
hehe, don't get me started,
basically, my AMD and WinME just don't go together (ME has some superior multimedia features over 2K),
I'll have to a full reinstall and make 2K my main OS soon,

>>wait til intel brings out there 64bit chips.....
hmm, wait we will, if you don't count the 4000$ per CPU itanium, it will take a while.

[CrazySurfaNZ]

yeah too right, can i ask one question tho... why arent u using win xp? its great... ive never had a problem with it... runs stable and great on my system

[swiftynz]

Originally posted by CrazySurfaNZ
yeah too right, can i ask one question tho... why arent u using win xp? its great... ive never had a problem with it... runs stable and great on my system
i suggest you read this (http://forums.overclockers.co.nz/showthread.php?s=&threadid=70). :D

i am really looking forward to this Hammer cpu. it wont exactly magically give AMD a 50% market share but it should improve it's standings somewhat. i await intel's response is with interest. :)

[GooGlo]

Originally posted by CrazySurfaNZ
yeah too right, can i ask one question tho... why arent u using win xp? its great... ive never had a problem with it... runs stable and great on my system

:D Hahahahhahahhahahaha!

Anyway....

Are there any confirmed speeds for the hammer? 800mhz seems a bit low...

[_N_]

CrazySurfaNZ, the reason why I don't use WinXP is that I got the legal CD (Pro version) and it started coming up with errors about 3 days after it was installed. IE errors, and Windows errors, so I gave up on the thought of using WInXP.
the OS looks good, but it was screwing up for me.

[Tiggerz]

Originally posted by Binky Stunt Cat
http://www.tecchannel.de/hardware/937/
german site, but has an english version of the article...

I read the review.. interesting to see the performance of the prototype cpu..

I do whish however, when people do reviews that they compare apples with apples. Comparing a perported 64 bit amd cpu with a 32 p4 cpu seems to me to be a bit stupid - especially when they then claim the former is faster.

I am sure the more technical people here would have worked this out and taken it into consideration. But the less technically informed wont be, and this in the long run could be more damaging to the industry. I am not surprised amd didn't want to comment on it.

I would be interested in seeing the clawhammer run alongside an itanium or itanium II. A cpu more closer in architecture.

[Viper_NZ]

Originally posted by _N_
>>dont knock it till youve tried it
hehe, don't get me started,
basically, my AMD and WinME just don't go together (ME has some superior multimedia features over 2K),
I'll have to a full reinstall and make 2K my main OS soon,

>>wait til intel brings out there 64bit chips.....
hmm, wait we will, if you don't count the 4000$ per CPU itanium, it will take a while.

Exactly what multimedia features is Me better at???

And intel's 800MHz Itanium has approximately 90MHz pentium performance @ IA32 code

[Deviant]

Originally posted by Tiggerz


I read the review.. interesting to see the performance of the I do whish however, when people do reviews that they compare apples with apples. Comparing a perported 64 bit amd cpu with a 32 p4 cpu seems to me to be a bit stupid - especially when they then claim the former is faster.


What can be fairer then a Hammer in 32 bit mode up against Intel in 32 bit mode, or are you that afraid that AMD need a few hundred pins tied behind their backs? Didn't you read Mirds post?

And AMD will go againt the Itanium, with the Opteron, and this isn't the Opteron Chip, so don't start comparing it with Itanium.

[Deviant]

For years people have been saying AMD is nothing, and will only ever play catch up to Intel, and that Intel bring out new inovative ideas. Not now!!! It's too early to say the tides have turned, but Intel have to watch out or it will be like 3DFX and Voodoo up against Nvidia, and we know who won there. It was only a few years ago that Nvidia were the small players in the graphics market.

[_N_]

>>Exactly what multimedia features is Me better at???

when I first started ripping DVD's I had problems doing it in Win2k, but that's OK now.

My main concer is not being able to record straight into DivX when recording of my TV Tuner.

I also like Windows Movie Maker, what's a good substitue for this?

[Tiggerz]

Originally posted by Binky Stunt Cat
and whats more, AMD have a lead over Intel in a 32 and 64 bit compatible chip for the transition.
the only reason Intel is doing so is coz AMD was in the first place.
Personaly i might just get a Hammer system in 2004, hopefully DDR II will be out by then.

The words bull and ox spring to mind here - in a friendly way.. Intel made their first 64 bit processor available to developers back in 1994. This current version - which you can buy now runs 64 bit WinXP which you can also buy now and will run 32bit applications. The 64bit AMD cpu which you cant buy now will also run 64 bit XP but you cant buy that now either because MS have only just announced they will support the AMD and it will take a while for them to add it. This does not sound like having a lead to me :)

Currently, sampling is the Itanium II, due to go into volume in a couple of months. In sept. oct. we should see the low power 0.13 micron deerfield sampling (have put my name down for one of them) and the Madison with the 3mb cache. Both are due to go into production round june next year..

Round middle of next year the montecito should start sampling for release in 04. This is based on the new 0.09 micron technology.

So this means that middle of next year Intel will have their 5th generation CPU samping, whilst AMD will be shipping their 1st gen units (which I hope will be good - coz I want one at some stage).


Also due to be released shortly is the new 0.09 micron SDRAM with 330m transisters - super high density stuff and I heard a rumour that rdram is on the way out [any one else heard this].

Also slated for release mid to early next year is the new 0.09 micron prescott based P4s.. Not sure if they will contain Yamahil technology (64/32bit) as Intel have gone a bit quiet on that one. My understanding that the remaining lifecycle of P4 was to remain 32bit only.


Finally on the AMD front could people be a bit clearer on which CPU they are talking about.. There is a big diff between the ClawHammer and the ClawHammer SP which will be the desktop version. Its like Xeon vs Normal in intel terms.

I thought thoroughbred should have been released this month.. Anyone know whats happening with this ?

[Viper_NZ]

Movie maker is on XP (although you said you had issues with it)

Personally I like 2k/xp for divx encoding. If the encoding app bleeds mem you can end up with 0% sys resources in the morning and have a really fun time with a half encoded file (old Xmpeg and a few others were quite bad, don't know about currently because I haven't used 9x in a while)

What model TV Tuner do you have?

[_N_]

flyvideo 98,

the dynalink one ate all my system resources in win9x

[Binky Stunt Cat]

*warning*
danger will robinson danger!!!
Intel fanatic on the loose!

As far as i understood it, the new 64bit chip from Intel was going to be solely 64bit, untill AMD stated their x86-64 work......
/me goes away to check up on this...

[mird-OC]

Originally posted by Tiggerz
Also due to be released shortly is the new 0.09 micron SDRAM with 330m transisters - super high density stuff and I heard a rumour that rdram is on the way out [any one else heard this].
i've heard that intel will support Rambus for some time yet. apparently there is some scepticism whether DDR and DDR-II can offer the same robustness and scalibility, given the inherent flaws in DDR design that have been revealed since it's conception. however in saying that there seems to be an underlying want to shift to to a single memory type across the board. i reckon Rambus will be around for at least the next two years anyway.

Finally on the AMD front could people be a bit clearer on which CPU they are talking about.. There is a big diff between the ClawHammer and the ClawHammer SP which will be the desktop version. Its like Xeon vs Normal in intel terms.
most of us here are only really interesting in desktop technology, referring to the clawhammer (be it SP or MP) rather than the sledgehammer.

I thought thoroughbred should have been released this month.. Anyone know whats happening with this ?
it's supposedly due for next week, meaning we should see it here by July.

[Binky Stunt Cat]

http://www.anandtech.com/cpu/showdoc.html?i=1295&p=1
there we go, thats about the x86-64 that AMD is using, as opposed to the IA-64 that Intel was going with at that stage.
While you could run 32 bit apps on it, you'd realy need a proper 32bit chip to run them well.

oh, and whats the point in having a 64bit setup if ur gonna only use 32bit apps???

[Deviant]

Originally posted by Binky Stunt Cat

oh, and whats the point in having a 64bit setup if ur gonna only use 32bit apps???

What's the point of buying a 64 bit CPU if you've only got 32 bit software? Well with Hammer you can slowly upgrade all your software to 64 bit, I thought it was damn obvious Binky the Stunted Cat.

[_N_]

yep,

but what they could do to save themselves some trouble is have a board running dual CPU with one 32bit and the other 64bit.
Average users won't need 64bit CPU's for about next 4 years, because it takes them that long to develop software which will take advantage of 64 bit CPU's.

[Tiggerz]

Originally posted by _N_
yep,

but what they could do to save themselves some trouble is have a board running dual CPU with one 32bit and the other 64bit.
Average users won't need 64bit CPU's for about next 4 years, because it takes them that long to develop software which will take advantage of 64 bit CPU's.

I have heard that when they moved from 8 to 16 and to 32 bit.

It takes approx 10 years to move all the software from one platform to another. First of all you need to release all of the compilers, then you need to update all the OSs, then teach everyone to program to the new model, then recompile and rewrite all the old apps, then get the corps to update all their software that works just fine the way it is.. :)

When the Intel 32bit processor came out - there were a total of 11 people world wide certified to write OSs on the processor (as far as I was aware at the time).. I suspect with amd at the moment, there are probably a lot less than this.

Itanium is pure 64bit - intel never said it wouldn't be.. It is capable of running 32 bit apps - in fact 32bit apps can be modified to call 64 bit routines - this is called mixed mode programming- something that hammer wont be able to do. (well at least the programmers manuals indicates you cant do it).

On a response to an earlier post, the new hammer cpus are being bandied around as 64 bit cpus so if that is the case - it is only fair to compare them to the itaniums.. if however, you guys decide that they are 32bit cpus, then fair enough compare it to a P4. Better still it would be good to refer to the SP model vs the non SP model..

It isnt a 64 bit cpu that can run 32 bit software.. according to the programmers doc (and correct me if I am wrong) - it can only be started up in either 64 bit mode or 32bit mode and has to be switched between the two (actually from memory I think it starts up in 32bit mode in order to initialize the hardware and has to be switched into 64bit mode). Hence there will be server versions and SP versions for the desktop - although this may change.

You wont see any 64 bit demos until someone updates their os to work with it.. I suspect this should be happening kind of nowish, so will probably see either a review of linux or xp64 soon.

It'll be good I hope - am looking forward to it.

[Deviant]

Originally posted by Tiggerz


On a response to an earlier post, the new hammer cpus are being bandied around as 64 bit cpus so if that is the case - it is only fair to compare them to the itaniums.. if however, you guys decide that they are 32bit cpus, then fair enough compare it to a P4. Better still it would be good to refer to the SP model vs the non SP model..

Compare 32bit benchmarks with 32 bit chips,
and 64 bit benchmarks with 64 bit chips. It makes sense to me.


You wont see any 64 bit demos until someone updates their os to work with it.. I suspect this should be happening kind of nowish, so will probably see either a review of linux or xp64 soon.


Where have you been, Microsoft has already demo-ed the XP64 with Hammer chips up and running a month ago.

[Deviant]

Originally posted by Viper_NZ


And intel's 800MHz Itanium has approximately 90MHz pentium performance @ IA32 code

If this is true, and I don't doubt Viper, then there's no way your going to start making comparisons of a 64bit chip with 32 bit software, but maybe Hammer will be an exception.

[KingJackal]

Originally posted by mird-OC
i've heard that intel will support Rambus for some time yet. apparently there is some scepticism whether DDR and DDR-II can offer the same robustness and scalibility, given the inherent flaws in DDR design that have been revealed since it's conception. however in saying that there seems to be an underlying want to shift to to a single memory type across the board. i reckon Rambus will be around for at least the next two years anyway.


RAMBUS have already alluded to the not-so-far-away introduction RIMM 3200, RIMM 4200 and RIMM 4800. These are 32-bit RDRAM modules, and they appear to be giving quite good yields!

RIMM 4800 are 32-bit RDRAM modules running at 600MHz DDR ( 1200MHz effective ). They are effectively 2 16-bit modules in one, so they can be run alone, and have DOUBLE the already large memory bandwidth of 16-bit RDRAM!!

PC2100 DDR SD-RAM: 2.1GB/s
PC2700 DDR SD-RAM: 2.7GB/s
PC800 RDRAM: 3.2GB/s ( in a pair )
RIMM 4800 RDRAM: 4.8GB/s ( ONE stick by itself )

ph333rrr.....
*just imagine an evil grin smiley here*

[CrazySurfaNZ]

Jeezzzz thats fast... ill be interested to see what happens with QDR ram... anyone know anything about this yet?

[Binky Stunt Cat]

*disclaimer*
i only care about 32bit performance of chips, due to the fact that i dont use any 64bit apps, nor know of any 64bit apps that would be used to perform the same functions as my prefered apps in 32bit (and no, i dont use XP)

[mird-OC]

afaik, 32-bit RIMMs are already in production and the new i850E chipset supports them.

and as bandwidth scores stand i think it's:

PC1066 = 4.2GB/sec
PC1200 (RIMM4800) = 4.8GB/sec

also, with the soon to be introduced 64-bit RIMMs we'll see RIMM 9600 modules... yes, that's right 9.6GB/sec... :D :D

oh and CrazySurfaNZ, Rambus IS essentially QDR... the only thing that comes close to keeping up is DDR-II, which is basically ultra-high speed DDR.

[Mashed_Penguin]

But rambus is an evil company that wants to take over the world, their products CANT be good... :D

[CrazySurfaNZ]

Yeah and no one has intel chips to use them with anyway... cos everyone loves and uses AMD :)


(waits for intel hoarde!)

[mird-OC]

well it all comes down to how DDR and DDR-II fair in the next year... DDR-II sounds promising on the surface but things don't look too good underneat, and it seems the industry is sitting on the fence a bit as to support for DDR-II. it's not entirely out of the question that AMD may even look into Rambus technology... who knows? :)

[CrazySurfaNZ]

Yep true... but im backing ddr solely because its cheap and royalty free :) (and it does a fairly good job!)

[KingJackal]

Yep true... but im backing ddr solely because its cheap and royalty free
The RDRAM royalties are due to a small chip on the RIMMs that RAMBUS has patent over. According to Anand ( I believe this to be accurate ):
The royalty is approximately 1 – 2% per RDRAM chip and 2 – 5% for the memory controller.
And besides, what do you care who owns what royalty rights? How does that affect your decision? Does it affect price significantly? Performance? Compatibility??? :confused:

Yeah and no one has intel chips to use them with anyway... cos everyone loves and uses AMD
AMD have a license to use RAMBUS technology. They just haven't been using it.

PC1066 = 4.2GB/sec
PC1200 (RIMM4800) = 4.8GB/sec

Yes - BUT that's with TWO modules. What about 2 of the RIMM 4800 modules? It's only a matter of time before chipsets support dual channeling these....

....but yeah, speed isn't the main issue with the new RIMMs. It's the ease of upgrades ( one module at a time etc ), and the fact that it should make the same amount of the same speed RAM cheaper ;).

Which, BTW, is all good :).

And as for QDR - I think that's more of a pipedream than anything close to an actual reality in the near future... :(

[Tiggerz]

Originally posted by Deviant

Where have you been, Microsoft has already demo-ed the XP64 with Hammer chips up and running a month ago.

Happily ignoring microsoft :)

[Binky Stunt Cat]

Originally posted by Tiggerz


Happily ignoring microsoft :)
dont we all??

[Tiggerz]

I thought this info might be usefull. This is from a discussion I had with someone recently talking about system design. It only discusses the MCH to memory architecture. If anyone has any questions or would like me to also talk about CPU to I/O bandwidth architecture just let me know..

Also, if anyone knows how the VIA chipsets work I'd be interested to know.


Ok, the way I see it on the ram front - and its been years since I did the training so correct me iffn I get it wrong here.

umm.. On yer bog standard desktop board you have a thing called a GMCH (Graphics and Memory Controller Hub) or MCH if the board on simpler boards in the case of the 845E.

This interacts with a number of components. One of which is your system bus. In the case of the 845E its operates at 100/133Mhz 4 samples per clock across a 64 bit databus. And has a throughput of 3.2GB/s.

You then have your suposedly high speed agp interface. This operates at 66Mhz, 4 samples per clock, 32bit databus, and a thoughput of a massive 1.066Gb/s (every wonder why that nvidia card was slow).

Then you have yer olde PCI 33Mhz slots running at an incredible 1 sample per clock on a 32 bit databus.. Giving a massive performance of 133Mb/s

Finally you have yer old faithfull DDR200/266Mhz ram running at 100/133Mhz 2 samples per second, 64 bit databus, giving 1.6/2.1 GB/s throughput.

All the above put together is termed bandwidth. So when someone talks about designing a system for memory bandwidth, they are talking about poking round with the above figures.

Now the bit that I left out above is the Hub interface for the memory controller hub (MCH). This sits between the cpu, ram and i/o controller. It has the grand specs of 66Mhz, 4 samples per clock, 8 bit databus with a 266MB/s throughput.

The above layout only has a single channel between the MCH and the ram.. So its called single channel ram for obvious reasons.

Now we go to the 850 chipset which uses RDRam.

This MCH is slightly different. First of all it has two channels between the MCH and the ram. Second of all it supports AGP fast writes. It also supports a 64kb line cache.

The system bus operates at 133Mhz and is quad pumped giving 533Mhz over the databus giving the 3.2GB/s memory bandwidth per channel (run in lockstep mode). In theory, this will give you 6.4GB/s Memory throughput, but it practice, most people only put in the wrong type of ram and end up with only 3.2.

Although - eat your heart out - those running nvida cards coz this will give a performance boost over the above system. Once the data gets out of the AGP slot that is.

The PC800 RDRam runs at 400Mhz. I should probably write something on this, but have forgotten why it was important..

Pretty much everything else here is the same.

Now lets look at a chipset that does not have AGP support. I'll use the E7500.

This is almost the same as the rdram chipset above. Except it has two DDR-SDRAM 200Mhz memory bus channels giving 3.2GB/s memory bandwidth per channel (again in lockstep mode).

It doesnt have an AGP slot. Instead it has a PXI-X slot 64bit databus running at 133Mhz. The other PCI-X slots tend to run at 100Mhz.

The point I am trying to make here is that many people think that RDRam means a faster machine. Well not really.. You can see from above that the real bottleneck it the AGP slot followed by the MCH and then your CPU. Also, once you have the data out of the AGP slot, the fastest you are going to get it across the system in any of the setups is at 3.2GB/s.

If you put an NVidia card in any of the top two boards, you should technically see absolutely no difference.. Any difference you do see will most likely be from the i/o controller.

I feel that if anyone makes a graphics card running on a PCI-X slot, then they will strike gold. Coz it will effectively double the AGP capability.. Sadly, no-one does this yet.

Out of non-interest one thing I did miss out was that PC800 RDRam actually only has a data throughput of 1.6GB/s per channel. When you have two channels you do actually add them together.. Hence the 3.2GB/s. Thats what lockstep does (had to look this up)..

[mird-OC]

that's a great post there Tiggerz. i think that'd make a nice appendum to KJs memory article :)

in my case i lean toward RD-RAM not for an increase in gfx performance, but to feed the bandwidth hungry apps i use. that's why i'm really looking forward to RIMM 9600 :)

stability in numbers is also a big factor. DDR is becoming more and more notorious for it's instability with largish memory configurations, and although the problem mostly lies in the the memory controller, it's no conincidence that RD-RAM consistantly outperforms every DDR platform in terms of robustness.

[Deviant]

Hey Tiggerz, correct me if I'm wrong, but the 1.066 GB/s bandwidth of the AGP card hasn't been slowing the video cards of today down yet from what I know.

It will soon be too slow, hence the AGP8x spec coming out. I agree that PCI-X will be the way to go when it is out in the mainstream but until then all we need is AGP4x or 8x.

The GPU does a lot of processing itself, and needs massive bandwidths for displaying all those pixels at high refresh rates as well as performing billions of calculations per second, but the actual data to the cards is small in comparison.

Think of the GeForce2's that you can still get in PCI form, and aren't troubled by the PCI bus. Admitedly, they might be a little slower, and limited to the GeForce2MX and nothing faster, but all the same they operate fine with a 133MB/s bandwidth on most games today.

[KingJackal]

Tiggerz, exactly what are you trying to say? That RDRAM isn't needed? Well, it's not supposed to be. RDRAM is a FUTURE memory specification. When it was released, it was most definately not meant for workstations, let alone PCs. It was meant for servers only.

[Tiggerz]

Originally posted by KingJackal
Tiggerz, exactly what are you trying to say? That RDRAM isn't needed? Well, it's not supposed to be. RDRAM is a FUTURE memory specification. When it was released, it was most definately not meant for workstations, let alone PCs. It was meant for servers only.

I was mainly putting things into perspective of where the ram fits in to the equation. To me Rambus = good. Here is the bit that I left out of the above.

Extending from my post above.. The three boards I mentioned were a desktop, mid-end and server board. You can see the architecure is the same for all three.

The design of the board very subtly effects the performance of the rambus technology (as it does with sdram).

The main component that effects performance are the length of the channel, the access time (45ns and 35ns) and the rams ability to transmit and receive at the same time. By channel I mean the channel from the MCH to the actual ram chip itself. This mainly affects the latency - or the time taken for data to transfer from the ram to the MCH (the actual time to travel across the channel is 1500 picoseconds). The fastest latency time is achieved by directly soldering the ram to the mainbaord. Next fastest is achieved by having ram-slots that you plug the ram into.

By default most MCHs only support 1 slot which can be up to 1 x 1Gb - or 2 x 1Gb slots if the MCH is using a duel channel bus. If there are more slots than this, then there will be a device called a memory repeater between the MCH and the ram. This is the slowest.

Higher end chipsets like the 860 and the e7500 for example will support two channels from the MCH to the ram. So you will see either 2 x 1Gb slots or 4 x 1Gb slots before a memory repeater is added.

Usually the rule is one channel from the MCH to Ram for each CPU - meaning each CPU will have its own bank of ram (and ram bus - pun intended).

Now rambus transmits 2 bits per clock per pin using a low power (1.2 - 1.4v) frequency. So, in order for them to get a significant number of bits across the bus they have to generate more clocks.. Hence the higher frequency the better the performance. Now the P4 has a system bus of 400Mhz which matches the 400Mhz of Rambus. You multiply the 400 x 2 bits = 800Mb/s x 2 bytes (16 pins) = 1.6Gb/s. Multiply this by 2 for dual [channel] ported ram and you get 3.2Gb/s.

This means to the un-initiated, that the rambus memory can handle whatever data the CPU is likely to throw at it.

This is good.

Now - where you have a situation where the ram cannot handle the output from the CPU - for example, you have a P4 chucking out 3.2Gb/s of data put ram that can only handle 2.1Gb/s of data, you get what is called a memory bottle-neck or bandwidth problem. The effect is that the CPU must remain idle while the memory catches up. This is usually the case with DDR ram.

This is not good.

Now we all know which manufacturer (and I am having a little dig here) who's name starts with something beginning with the letter before B - loves to use DDR ram. I even note that they list a bunch of techniques for getting round this in their programmers manual - one involves using MMX. :)

Now if you take your bog standard rambus and double the number of data bits ie. 32 rather than 16 and also increase the number of clocks (increase Mhz) then you are going to be able to shunt heaps more data across the bus than the CPU will need. Likewise, the CPU will never have to wait for the ram to catch up like in the above bottle-neck scenario.

This is good.

Why ?

Because not only does data go to the CPU and back, it also goes to the AGP slot and the PCI busses and the I/O controller as well.

In the original post I stated that the PCI slots are chucking out 133Mb/s the AGP (4x) was 1066Mb/s and the CPU will be chucking out 3.2Gb/s + whatever you get from the I/O controller. That poor old ram has to be able to cope with all that.

Hence the new PC1200 stuff 9.6Gb/s will be real real good.

When I said the AGP slot is a bottle-neck.. Remember that your application is going to be spitting data into ram at 3.2Gb/s, your CPU will be processing it and spitting out at the AGP slot at 3.2Gb/s.. Because the AGP slot is only 1066Gb/s there will be a bit of a bottle-neck here. I suspect thats why nVidia came up with there own bus design.

[KingJackal]

Yeah, pretty much. But with more and more memory being integrated into GPU's, and more and more GPU power, the bus is less of a bottleneck. As far as I can tell, The amount of RAM on a GPU ( even relative to system memory ) is increasing slowly. So, of course, is the cost of high-end cards. My point being that to a much larger extent now, GPU's are becoming more independant.

Eg:
In a 3D Game, it just shunts all the level data to the GPU. It may even all fit onboard, or close to it now. So the level load time will be permanent storage speed limited ( permanent storage is almost always slower than your buses ), and the rendering speed will be bottlenecked within the GPU. The AGP bus becomes more of a way to shunt control data than expensive video, texture, polygon etc data.

Of course, streaming CPU-rendered video makes a good counter-example. For instance, DVD ( assuming your GPU doesn't have DVD acceleration - another example of GPU independance increasing ).

...so as for the future? Well, it's obvious that yes, even AGP 8X sure won't last forever. But I think the reason why even that hasn't boomed yet in industry is there simply isn't the demand. More GPU independance through features like DVD acceleration, onboard T&L, Pixel Shaders, Vertex Shaders - it all means that the video output on your monitor relies a whole lot less on your GPU bus bandwidth.

[Deviant]

Originally posted by Tiggerz

When I said the AGP slot is a bottle-neck.. Remember that your application is going to be spitting data into ram at 3.2Gb/s, your CPU will be processing it and spitting out at the AGP slot at 3.2Gb/s.. Because the AGP slot is only 1066Gb/s there will be a bit of a bottle-neck here. I suspect thats why nVidia came up with there own bus design.

When KJ said what exactly are you trying to say, couldn't you do it in 30 words or less. Fortunately there is a word/post limit in this forum. What a read!

I still don't believe tha AGP slot is maxed out with data. Yes it's theoretically possible as you have stated, but I don't think it is that way with applications such as games that we use. (See my earlier post for reasons why)

[Joshsti_NZ]

haha, even though i have an Intel :rolleyes: .. i love teh AMD!! :D

[HomerNZ]

this morning i ate a banana............................... it tasted ok i guess..................

[Deviant]

Originally posted by HomerNZ
this morning i ate a banana............................... it tasted ok i guess..................

?????????? Why r u spamming? What good does this do for anyone?

[Grrr!!]

Originally posted by Deviant


?????????? Why r u spamming? What good does this do for anyone?

What does that do for anyone? +1 post count for HomerNZ.

[TB_]

Just for the record, Windows ME has much of the Windows XP bits and pieces. Which is why ME has multimedia stuff than 2k. It wasnt ready on the 2k code base til XP.

[el roffo]

however fast the hammer is going to be, amd better come up with a good marketing scheme...

for some reason i dont see the average folk buying an 800mhz hammer, because they could get a 2ghz or something for a similiar price...

imo the whole xp thing seemed to go well?