Printable View
Sure!Quote:
Originally Posted by gOJDO
Systems form above companies will be available to purchase at the mentioned date, but they allready are building this computers using 65nm CPUs.
If you're Dell for example you will not put new Box on sale before you build up adequate amount in magazines, yes? Otherwise you will end up with many people complying about tight supply or lack of availability.
That is my point of view.
PS. You're quick in responding :toast: .
Before Core2 was released, there were a lot of reviews and benchmarks. On the release date we were able to buy it, although its availability was low.
If we can buy any 65nm K8 CPU today or this week, then we can consider it as non paper launched.
You can either have information about K8L and Yorkfield or look at a blank page.Quote:
Originally Posted by GoThr3k
I bet nvidia wish they could keep their secrets as good as this ! :D
Regards
Andy
I bet everyone does, at this rate in January or March when someone can actually buy one they will find out that they come packaged in a retail bannana or something crazy.Quote:
Originally Posted by zakelwe
No, the 1.25/1.35v thing means they sometimes drop down lower, but at max performance they are 1.35vQuote:
Originally Posted by Mikael
Quote:
Originally Posted by terrace215
or it could mean that there will be normal models and EE models as with the 90nm AM2 CPUs.
cause if it were just a part of C&Q then i doubt theyd mention it.
Are you questioning the reliability of roadmaps and data sheets previded to and from mobo manufactuers from AMD and intel directly? Then I could very well say York will never clock at 3.73ghz on stock because its wrong when in reality it WILL indeed clock at that speed. AMD will release up to 3.5ghz K8L's in 2008 to counter it. And run HT3 at full speed. The links arn't the same anymore man. Do you understand the HT3 links are different from standard HT links they are nolonger fixed at one speed.Quote:
Originally Posted by gOJDO
See they run at what clock the K8L main clock is at for model number and greater differences since the pads are also in the cpu for hyper transport AMD made the hyper transport FSB rely on clock speed and give greater model differences. No one CPU with HT3 will have the same link speed. They made it flexible with different model's according to clock speed. So the HT3 link is nolonger independent.
Lower clocked CPU's will run 1.6ghz hyper transport. Highest clocked cpus will run 2.6ghz hyper transport. That is what HT3's max is. 5.2GT/s. You don't understand how K8L will use HT3 links do you? Its not the same. All mobos that are capable of HT3 will say supports HT3 speeds up to 2.6ghz Hypertransport or something close to that. Maybe less when the AFX's only do 2.9ghz on launch. They are launching a new arc. Its only natual to expect a new ark to start at a week speed and raise over time like the old days. K8L by far is not limited at 2.9ghz that would be a joke in its self. York is clocked at 3.7ghz for a reason you know agenst the AFX.
Infact we won't know anything performance wise until there is real physical hardware to show to bench agenst. We can all think how K8L might perform but we will never know until we can OC the new arc. Its amazing how Conroe OC's so well and is a new arc. We should expect the same from K8L because its not K8 by a long shot. Not one part of the cpu is the same if compared die for die to K8L. 65nm K8's are arc limited at a sertent speed. It has nothing to do with fab process only arc scalability and performance. K8's arc is 3 years old, K8L is totally new like conroe. Don't expect any less then the unexpected because as far as anybody is consirned its limits are unknown.
If this is not a paper launch then I don't know what is:rolleyes: We've already known about the December 5th date, now they're saying "Hey everyone they exist but you can't touch!"..
Wait you didn't your 65nm K8 in the mail yet? :rolleyes:Quote:
Originally Posted by red
Don't you know how to order parts? :p:
Why is AMD keeping the Brisbane die size a state secret?
Only the Inq makes reference to it, saying they are not putting it in the official info packet, but it is somewhere in 120-130 mm^2 range.
http://www.theinquirer.net/default.aspx?article=36162
This would be pretty bad, if true.
90nm equiv is 183mm^2.
If 65nm is 125mm^2, say, that is 68% of the size of 90nm.
A good 65nm process shrink should've accomplished something closer to 58%, or 105mm^2.
HTT clock is not dependend on the CPU clock!Quote:
Originally Posted by Serge84
HT links have never been fixed at one speed(freqfency). Unlike HTT2 and HTT1, HTT3 offers dynamic link width and dynamic clock adjustment of transmit and receive links. It also offers ACPI compliant dynamic power management. HTT3 has the same concept and works exactly like HTT2 and HTT1 and it is backwards compatible to both HTT2 and HTT1.
http://www.hypertransport.org/images/features_chart.gif
Quote:
The new 3.0 standard nearly doubles the clock speed and bandwidth of the previous HyperTransport 2.0 specification. In addition, HyperTransport 3.0 supports a variety of new features including AC coupling mode, hot plugging, un-ganging mode and dynamic power management for the support of extended signal transmission distance, typical of backplane and chassis-to-chassis implementations.
Quote:
HyperTransport technology devices are designed to operate at multiple clock speeds from 200MHz up to 2.6 GHz, and utilize double data rate (DDR) technology transferring two bits of data per clock cycle, for an effective transfer rate of up to 5.2 Gigatransfers/sec in each direction. Since transfers can occur in both directions simultaneously, aggregate transfer rates of 20.8 Gigabytes/second and 41.6 Gigabytes/second can be achieved with 16-bit wide and 32-bit wide link configurations respectively. To allow for system design optimization, the clock speed of the receive and transmit links may be set at different rates.
http://www.hypertransport.org/Quote:
It is important to note that HyperTransport 3.0 specification is fully backward compatible with prior 2.0 and 1.x devices. This means that existing investments in the technology can continue to be leveraged in the future.
http://www.hypertransport.org/docs/tech/HT3pres.pdf
http://www.hypertransport.org/tech/tech_specs.cfm
http://www.hypertransport.org/tech/tech_specs_conn.cfm
Quote:
Originally Posted by terrace215
AMD's Gen 1 65nm is a combination of 65nm and 90nm technologies, I recall.
That sounds highly unlikelyQuote:
Originally Posted by Shadowmage
My first guess is that 65 nm will bring AMD up to speeds are Intel doing now, and that's very optimistic.
And my second guess is that K8L will be performing something like a C2Q are doing now.
Just when AMD finally seems to be getting it right again, Intel will launch 45 nm and take the lead, again. I really hope I'm wrong.
I really think that AMD could have done some changes earlier, very small changes have been made with the K8 core since the release in April 2003. But then again they have been very successful with it, and they don't have all that cash Intel got for R&D.
Yeah, I'm just speculating. And I pretty much prefer AMD.
I never said it was anything of the sort. lol Your funny. Its not great at all. York will do great on CSI infact AMD doesn't stand a chance with HT3. HT3 was made to only allow more bandwidth and insure there would never be a bottleneck between 16cores on a 4 socket machine. It will not increase performance only allow K8L to use its full power and scale better with lower Lancery.Quote:
Originally Posted by brentpresley
HT3 hasn't even been finalized yet. We know almost nothing about it eather. But it's not for performance only for bandwidth and anti-bottlenecks. Thats why CSI was made so Conroe could show its full power and have no bottlenecks. Thats why york is going to kick so much ass if it is on csi hopefully.
To the other guy...
K8L will have a HT3 link at different max settings for the cpu and mobo at the time its launched. Those reversions are only for the mobos chipset max speed. They have nothing to do with the max speed of what ever rev they put in the cpu and each cpu will have a greater or smaller capable HT3 pad. Because there will be a few stages of HT3 put into mobos the 1.5ghz in 2007 in AM2+ up to the 2.6ghz in 2008 AM3. All the CPU's like how semprons only support 800mhz Hyper transport of today. There will be cpus like this in the lower budget and then about 1400mhz or 1300mhz hyper transport 3 for those budget models or something close to it. Then 1.5 to 1.7ghz HT3 for the X2's some of the higher end ones will go up to 2.1ghz but this is at high clock speeds and prob performance rated parts, then 1.8ghz to 2.1ghz for the X4's and AFX's. Its said on there between night and day the proof is right before your eyes the max GT/s each cpu can support. Hello!
http://badhardware.blogspot.com/2006...y-rainier.html
pop...
http://photos1.blogger.com/blogger/7...M7OCMYeXam.jpg
Says right here. According to this chart... the HT3 link depending on model and series will support from 3.0GT/s to 3.4GT/s for the Legacy models, 3.0GT/s for Mainstream, and 4.0GT/s to 4.2GT's for Performance and Ultimate performance models. Can you read the chart? As pane as day to my eyes. Those speeds are strictly depending on the clock speed do the math. If you can't see that then your hopeless.
I/O Speed is the GT/s of what the cpu supports at what clock speed. For example you see 2ghz to 2.9ghz that would mean 2ghz would have a 1500mhz HT3 link. Then the one at 2.9ghz would have a 2.1ghz HT3 link. Wow wasn't that hard to figure out. What ever model inbeween would have a different HT3 link speed. So if this is wrong then the MB manufactures are wrong because it says the Source. Are you calling perfessionals the lier? All you showed me was a link to what HT3 chipsets would support. Not the CPU's at the time. Says in the link I just gave that it would support between so and so speed for a different model and series according to that chart ofcorse. Only logic can tell you that. You have to see it for yourself in what they mean by 3.0GT/s to 4.2GT/s matches the other charts perfectly right below it that say I/O Bus Speeds for K8L. Doy!
I'm pretty sure I'd listen to reliable sources other then yours. Sorry but you didn't prove anything to me at all here. I hope you learned something about what I was talking about in HT3. Doesn't matter what the max chipset supports to a point, cus you can OC the chipset, but again its just saying the max the chipset can support. But the K8L's will support multi speeds in HT3 so depending on the chipset you can put in a faster HT3ed cpu for a faster chipset. BOTH support different MAX HT3 Speeds and the link can be set in the bios to X10 or whatever they will have it like. This is spread over a few months time. K8L won't even fully support HT3's max until later and there chipsets will be limited in the same way. The REV's of both matter only if you want to run the max the cpu's HT3 link can take. Doesn't matter for anything else cus it will run higher CPU models HT3 speeds only at the chipsets max default speed. Not the cpu speed. I mean it doesn't make OCing limited because both have a max only for the HT3 link at max you can turn that down ofcorse in a oc. I'm just saying thats what the max multiplier will have to support HT3 link REV so and so for chipset and cpu. They will not come out with a chipset or cpu that runs max HT3 yet until 2008. They will both have different HT3 speeds and multipliers for each model.
Serge, Yorkfield is a die shrink of Kentsfiel with little improvements, as it is based on Wolfdale (wich has 6MB L2 cache and SSE4) and FSB1333.
CSI and Intel's IMC will only show up in next generation (Nehalem), wich will be native quad core with HT (8 logical cores), in late 2008...
@Serge84
:nono:
Dude I have learned nothing from you. Your are puting words into my mouth and make conclusions about something that I have never said. I hate to repeat my self, but I'll repeat to point at your missleading statements:
Do you realise that what you said is not true?Quote:
Originally Posted by Serge84
Here is what I answered:
There are reasons why all K8L CPUs will not run HTT at same freqfency. For some of those reasons, all K8 CPUs are not running HTT at same freqfency(all Sempron/all s754/all sS1 at 800MHz, Athlon64/X2/FX/Opteron on s939/s940/sAM2/s1207 at 1GHz).Quote:
Originally Posted by gOJDO
I have never said that K8L CPUs will run the HT links at same/top speed. Although your epxlanations are wrong and missleading, are not necessary for me and can go down the toilet. Skip reading/learning from blogs and start learning from realible sources like from the inventors and developers of the hardware/tehcnology you are talking about. In this case that would be AMD and HyperTransport. I allready posted link of HTT3 paper (500 pages of technical documentation), but looks like you haven't opened it yet.
Please read the HTT specifications, how it works and how it is implemented. Also read some papers about K8L and its power management. I hope that you will come to the conclussion that the speed of the HTT is not dependend on the CPU freqfency.
A little help: If AMD want they can use the max HTT3 freqfency on 100MHz K8L CPU. The cores on the K8L will have independend clocks and voltage. That means that one core can run at top freqfency, while other at bottom. So, where is the logic with HTT dependecny on the CPU clock?:rolleyes:
Do you know that the HTT3 on the K8L will have power management also?
Your also making conclusions in a war that stops here. Just give it a rest and we will see whos right the MBM's or you ok cool down man its really just speculation then if you don't beleave what the MBM's are trying to say, believe what you want, I don't care. And K8L supports HT3 and standard HTL. If you put a K8L into AM2 it will lose HT3 features and power management features being able to run with multi power planes and voltages independent for each core. So what it runs a little slower and may have a minor bottleneck maybe not. Will not run as good and heat up more because all cores are running at the same volts. Even needing a 100A socket won't matter. K8L is desined to work in AM2 even with a lower. "Amp-level" it might need a little more volts is all to fix that. Yes I know what K8L's power management means. It also works similar to conroes power management for areas in the die that are not in use and will turn on and off depending on when not needed or when needed. Pretty cool stuff.Quote:
Originally Posted by gOJDO
A long long time ago, even before 90nm, tomshardware ran complete set of benchmarks on Athlon64's with HT reduced from 5x to 4x, and even 3x. Performance was nearly identical across the board.Quote:
Originally Posted by brentpresley
HT's impact on performance is turning out to be nearly as big a computer myth as AGP's impact on video card performance (ie tomshardware also did a whole road of AGP tests, and performance only started substantially degrading when you got to AGP2x)
I'm by far no expert, but transistors on the die are all sorts of sizes. Smallest length, smallest gate, highest performance transistors are used where needed most in critical path sections. Other rules for fan-out, matching, drivers, dynamic logic, repeaters etc.. still apply ofcourse.Quote:
Originally Posted by Shadowmage
(BTW: I was like !!! when I read 9 metal layers, wow, thats a lot of complexity)
The tests were also performed here. All the way from 2x to 5x. Cant quite remember the name of the person who did them, though.
Have you seen tests performed with 2+ cores? I have not, and have not seen links nor articles relating or referring to the change in performance either.
For 1P CPUs, the speed of the HTT is irelevant, be it for singlecore, dualcore or quadcore CPU. Twice K8 cores will not double or increase the traffic between the CPU and the northbridge chipset, if the rest of the hardware remains same.
thank you. all communication between cores on the same cpu die for amd occurs within the chip. no HT links needed :toast:Quote:
Originally Posted by gOJDO
so for single socket setups, this is true:
http://eclipseoc.com/index.php?id=6,47,0,0,1,0
ozzi you seriously need to rerun those tests with more than just a single Geforce4MX 440. A CHEAP X1600XT would be more realistic