AMD's 65nm Preview Part 2 - The Plot Thickens

[tylerw13]

http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2893

interesting....i thought amds 65nm processors would be some what close to the intel...but in most cases they dont even beat their older 90nm brothers....if you ask me its kinda like the freakin 4X4 thing....why even buy it????

 

[SunnyD]

Cooler & More energy efficient come to mind.

 

[Rock Hydra]

Originally posted by: SunnyD
Cooler & More energy efficient come to mind.

That's about all...

 

[Cogman]

Possibly overclockable? Not much reason though. Anyways right now AMD is not the way to go for upgrades (unless your going at a lower level)

 

[Duvie]

The step back in performance with Brisbane is truly puzzling;

This about sums it up....

I also think we can rule out real overclockability at the near future if Anandtehc is to be believed on the vcore coming on their 65nm parts. I dont see much of an improvement necessary for the 65nm part to achieve enough top end to even come close to Intel.

The fact is the E6400 a 2.13ghz part pretty much beat the 5000+ brisbane every time and often times by 10%. That is 2.13ghz 10% faster then 2.6ghz...That is partly to the fact AMD has gone with the approach of lower l2 cache which is a mistake in my opinion.

Since I have tested 4mb C2D's versus 2mb C2D I have seenfirst hand that many apps take advanatage of the larger cache pool...


IMO AMD is looking like INtel of a few years ago....blunder after blunder....

Lets hope they get it together with the K8L and the K10....

 

[BitByBit]

The increase in L2 latency could have been made in preparation for Barcelona, which could well use higher latency L2. That would make sense, since its shared L3 may reduce the need for faster private caches.
As the article mentions, there is little reason for increasing the latency when the most recent roadmaps are taken into consideration.

 

[BigDH01]

I just ran some quick numbers that don't really mean much with the data points I have but I thought I'd share. This has to do with scaling. I took Anandtech's review and looked at the scaling from 6300 -> 6400 -> 6600 and 4600 -> 4800 -> 5000. Like I said, with 3 data points this doesn't really mean much but I was curious to see which architecture scales better. The results are somewhat thrown off by the cache size of the 6600 and the increased latency of the 5000 and 4800 compared to the 4600. I only ran the encoding tests as I'm at work. I also went out to 3 significant digits which is beyond what would be allowed on these tests if I followed scientific rules, but I have to go out that far to find any sort of difference. Everyhing is presented in a per MHz basis. For example: I'd take the performance difference between the 6600 and 6400 and divide it by 266 to get difference by MHz. Basically, this measures performance per MHz. Here it is:

WME E6600 + .023/MHz
E6400 + .02/MHz

5000 -
4800 +.02/MHz

The 5000 performed worse than the 4800 in this case making these results difficult to trust.

DivX E6600 + .003/MHz
E6400 + .004/MHz

5000 + .001/MHz
4800 + .002/MHz

This test makes some sense. Scaling slows with a growing clockspeed but you can see that the Core 2 scales better here. It also shows that the extra cache of the 6600 doesn't help much.

Quicktime E6600 + .008/MHz
E6400 + .013/MHz

5000 + .011/MHz
4800 -
The 4800 performs identically to the 4600 in this case. I'm guessing this is because of the increased latency of the 4800. However, both the 90 and 65nm 5000 perform the same making the above conclusion seem dubious. This might be a testing error. However, scaling seems similar for both chips but it's difficult to tell with only one data point.

iTunes E6600 + .004
E6400 + .004

5000 + .003
4800 + .001

This one shows better scaling at higher clock speeds so might not be accurate. However, it shows the Core scaling slightly better.

All in all, these above results are worthless . However, I was bored at work and decided to do something to occupy my mind and thought I would share.

 

[Regs]

I agree Duvie. This is more or less a step backwards in performance yet a step forward in, I guess, a marketing ploy? Why sell something like this? Do people really care about a marginal improvement in energy use? Energy efficiency on old architecture means nothing. Energy efficiency in upcoming products mean everything. The whole idea of 65nm is so that they can almost guarantee a new high performing part that does not suck down power and give off enough heat to cook a burger. So what on earth is the reason for Barcelona? Are they really trying to sell a stepping stone to the public for upcoming products? The nerve.

To this day we still don't have a engineering sample of the K8L or one at least displayed to the public. Though after seeing the "predicted and virtualized" performance gains in AMDs online commercial, I seriously doubt AMD has anything yet to show in a real demonstration.

Though to be honest my patience ran thin months ago before the Core Duo even came out. While Intel was still making blonder after blonder, AMD was only offering what they all ready had made. A whole year seemed to be wasted by this whole "mega tasker" stuff that no one bought into though what some members in the fan world had to have it. I guess so they don't get much of a performance impact running two things at once...Which is another argument of its own.

I still don't own a dual core processor. Though at least Intel is making it worth while by actually upgrading to a dual core which performs better in all tasks on a computer and not just a selected suite of programs.

 

[atom]

Originally posted by: Regs
I agree Duvie. This is more or less a step backwards in performance yet a step forward in, I guess, a marketing ploy? Why sell something like this? Do people really care about a marginal improvement in energy use?

The incorrect assumption you make is that the transition to 65nm was for the consumer. The average consumer doesn't care about 65nm vs 90nm. They might care about energy efficiency, but I suspect most people don't. So why go to 65nm? Yields. More parts per wafer = more money for AMD.

Energy efficiency on old architecture means nothing. Energy efficiency in upcoming products mean everything. The whole idea of 65nm is so that they can almost guarantee a new high performing part that does not suck down power and give off enough heat to cook a burger. So what on earth is the reason for Barcelona? Are they really trying to sell a stepping stone to the public for upcoming products? The nerve.

That makes no sense at all....if energy effiency matters to you, it should matter to you whether it's a new architecture or not. And again like I stated above, your idea of why AMD made the transition to 65nm is wrong.

I mean cmon, did you really expect 65nm to bring much greater performance? People using phase change and other extreme cooling on 90nm chips have hit a hard wall, which most likely meant a limit on the architecture itself, not heat issues. Considering Brisbane was a dumb shrink I don't see how anyone could assume it would have brought a huge performance gain.

I'm surprised that Brisbane does worse than the 90nm parts but I'm not really surprised at all performance didn't increase. Then again it's still new and not many people have overclocked it. Hopefully the average overclock for the 65nm chips increases an extra 200-300 mhz. I think thats the best people should reasonably hope for. Nobody should expect their chips to be hitting 3.6-3.8 on air.

 

[Duvie]

Originally posted by: atom

Originally posted by: Regs
I agree Duvie. This is more or less a step backwards in performance yet a step forward in, I guess, a marketing ploy? Why sell something like this? Do people really care about a marginal improvement in energy use?

The incorrect assumption you make is that the transition to 65nm was for the consumer. The average consumer doesn't care about 65nm vs 90nm. They might care about energy efficiency, but I suspect most people don't. So why go to 65nm? Yields. More parts per wafer = more money for AMD.

Energy efficiency on old architecture means nothing. Energy efficiency in upcoming products mean everything. The whole idea of 65nm is so that they can almost guarantee a new high performing part that does not suck down power and give off enough heat to cook a burger. So what on earth is the reason for Barcelona? Are they really trying to sell a stepping stone to the public for upcoming products? The nerve.

That makes no sense at all....if energy effiency matters to you, it should matter to you whether it's a new architecture or not. And again like I stated above, your idea of why AMD made the transition to 65nm is wrong.

I mean cmon, did you really expect 65nm to bring much greater performance? People using phase change and other extreme cooling on 90nm chips have hit a hard wall, which most likely meant a limit on the architecture itself, not heat issues. Considering Brisbane was a dumb shrink I don't see how anyone could assume it would have brought a huge performance gain.

I'm surprised that Brisbane does worse than the 90nm parts but I'm not really surprised at all performance didn't increase. Then again it's still new and not many people have overclocked it. Hopefully the average overclock for the 65nm chips increases an extra 200-300 mhz. I think thats the best people should reasonably hope for. Nobody should expect their chips to be hitting 3.6-3.8 on air.

I didn't expect anything...The winchesters and 90nm were better then the 130nm newcastles but that had more to do with the advent of dual channel memory controller. However I didn't expect as anandtech pointed out a worsing of performance from 90nm to 65nm when there was no architectural changes...

I did expect lower vcore then what anandtech was stating...i expected 1.3v for all of the parts. A drop of only .05v is not going to buy AMD much headroom and that is something else I was expecting.

The 90nm winchesters came in and were able to OC about 300mhz further at least then the newcastles and were .1v less from the shrink....

The higher vcore assures us of higher TDP chips then what we thought. I see less headroom to OC then I was hoping. I basically see no real chance of much above 3ghz.


The fact these have been ouot for a bit yet I hear no raving results, speaks volumes of AMDs results....

Basically we can all skip this line and hope for greater out of the K8L.....

 

[StrangerGuy]

IMO, 65mn A64s are pretty much the same as the 65nm Pentium-Ds. Not to improve performance but only to test-drive the new process technology.

 

[lopri]

90nm Socket 939 A64 > 90nm Socket AM2 A64 > 65nm Socket AM2 A64

:disgust:

 

[Duvie]

Originally posted by: lopri
90nm Socket 939 A64 > 90nm Socket AM2 A64 > 65nm Socket AM2 A64

:disgust:

QFT....

I guess that is 2 step back or at least running laterally for the last 6 months...

 

[coldpower27]

Originally posted by: StrangerGuy
IMO, 65mn A64s are pretty much the same as the 65nm Pentium-Ds. Not to improve performance but only to test-drive the new process technology.

At least 65nm Pentium D resolved the heat issue some and could be overclocked a decent bit further, as well they added more cache and the speed of said cache didn't decrease,as well it's a MCM implementation so the yields would be improved. Pentium D 65nm was either faster or the same as the Pentium D 90nm with equivalent clock frequency.

This 65nm shrink for AMD, just doesn't look all that amazing. The die size while smaller is quite a bit larger then the average optical shrink size by Intel & AMD, the LV2 cache is now slower. They are also barely available.

I am quite surprised by the voltage levels, not much of a drop at all.

Guess we will have to wait till K8L/K10 derivatives to see what the 65nm process can really do.