AMD 65nm to use half multipliers + 65w 5200+

[vapb400]

http://www.hkepc.com/bbs/itnews.php?tid=629089 - half multis
Eliminating the cach differences for half multipliers is rather interesting
http://www.hkepc.com/bbs/itnews.php?tid=629102 - fx-64 and 5200+ info

[lowfat]

Hopefully they will be downward half multi's open. I loved 1/2 multis back with A-XP days.

[einCe]

yeah im a fan of the half multi's too. can't wait

[VulgarHandle]

Well, dailytech has an article with more info.....
http://www.dailytech.com/article.aspx?newsid=3223

Single-core AMD Athlon products will remain on a 90nm Fab process with no foreseeable 65nm future. With the new 65nm Fab process AMD will be able to increase frequency increases of 100MHz increments unlike the 200MHz increments of current products. This will be interesting as AMD can theoretically abandon all 2x1MB parts in favor of 2x512KB parts with 100MHz frequency differences. DDR2-800 support will be brought to all 65nm processors including the Sempron. This should bring welcomed performance improvements to the AMD Sempron product lineup.

So, since single-core seems to remain 90nm, but Sempron will be 65nm, does this mean that Sempron's will be dual-core?

[nn_step]

Half multi

[STEvil]

but does the memory divider recognize the half multi or is it just gonna do full-steps still?

[ozzimark]

but does the memory divider recognize the half multi or is it just gonna do full-steps still?

i'd be surprised if amd changed the memory controller, though i know i would love to see 1/2 memory dividers..

[nn_step]

i'd be surprised if amd changed the memory controller, though i know i would love to see 1/2 memory dividers..

umm they just did..when switching from DDR to DDR2/DDR3 they have some tinker time to play with extra features

[ZX7891]

id like to see how these half multis work, never worked with 939s

[VulgarHandle]

id like to see how these half multis work, never worked with 939s

umm, they did on dfi lanparty's for sure.....but as has been stated, when you used the .5 multi's, the memory divider rounded up to next whole number(if you use 200x12.5=2500, memory divider would do 2500/13=192.xxx)

[STEvil]

I'd much prefer that they allow us to set the memory divider based on an integer value we select.

Say I had a 165 @ 3216mhz, and chose 15 as a divider for the mem to run it at 214.4mhz.. I want full range control of the mem!

[nn_step]

I'd much prefer that they allow us to set the memory divider based on an integer value we select.

Say I had a 165 @ 3216mhz, and chose 15 as a divider for the mem to run it at 214.4mhz.. I want full range control of the mem!

well that is already available on your s939 and sAM2 processors. Just not to the degree you desire.

[ozzimark]

umm they just did..when switching from DDR to DDR2/DDR3 they have some tinker time to play with extra features

true, but why leave 1/2 dividers in there and not use them? i know ddr2 could bvenefit a lot, due to all the "my ram is only running at 360mhz at stock!" stuff. it's a much bigger gap in mhz than it was with ddr1... and it's happening with the fastest supported speed, which didn't happen before. amd better get to work

[VulgarHandle]

is it possible for them to set enough dividers so we can raise ram speed by increments of 1?

[nn_step]

is it possible for them to set enough dividers so we can raise ram speed by increments of 1?

Yes, but only if your motherboard supports it. For example some DFI boards and MSI and Epox boards (s939) give you the option of DDR-500. which is about what you are thinking of.
The ram will run at 250Mhz and the HTT runs at 200Mhz

[gOJDO]

Yes, but only if your motherboard supports it. For example some DFI boards and MSI and Epox boards (s939) give you the option of DDR-500. which is about what you are thinking of.
The ram will run at 250Mhz and the HTT runs at 200Mhz

it requires apropriate CPU,K8 Rev E or latter.

[STEvil]

Yes, but only if your motherboard supports it. For example some DFI boards and MSI and Epox boards (s939) give you the option of DDR-500. which is about what you are thinking of.
The ram will run at 250Mhz and the HTT runs at 200Mhz

no.

No AMD motherboards support free range ram multiplier (cpu divisor really) control.

None, nadda, zip.

[_damien_]

No AMD motherboards support free range ram multiplier (cpu divisor really) control.

None, nadda, zip.

All he said is that there are dividers that runs the RAM faster than the HTT, and that a few mobos support a full range of dividers (except half-dividers). Which is simply true, but the DDR433, 466 and 500-dividers only work on rev E CPUs, like gOJDO said.

[ozzimark]

true, but the actual dividers that are available are what counts. atm, there is only integer dividers, so ram speeds will continue to make very coarse jumps in speed, regardless of what you do in the bios

[STEvil]

but they do not offer full range dividers/multipliers for the memory which was what started this branch of the discussion and what people misunderstood to to be cpu multipliers.

[_damien_]

but they do not offer full range dividers/multipliers for the memory which was what started this branch of the discussion and what people misunderstood to to be cpu multipliers.

I was certainly talking about dividers, and what you replied to nn_step in post #17 doesn't make sense either way. He said that certain 939-mobos provide divider adjustment in increments of one, and you said no (for some reason)

There are up to 10 dividers available. And how often do you need to run your RAM any slower than (CPU speed) / (CPU multi + 6) ?

[_damien_]

true, but the actual dividers that are available are what counts. atm, there is only integer dividers, so ram speeds will continue to make very coarse jumps in speed, regardless of what you do in the bios

Sure, half dividers would be nice. But half multipliers still help a bit, as they too provide more different combinations of CPU and RAM clocks.

[nn_step]

but they do not offer full range dividers/multipliers for the memory which was what started this branch of the discussion and what people misunderstood to to be cpu multipliers.

The reason they don't provide FREE RANGE Multis is because the extra logic required for it will SERIOUSLY IMPACT PERFORMANCE AND LATENCY.
We are talking 10-20Million more Transistors and WHOLE nS of LATENCY. Such a design change on the 90 or 65nm Process will Kill performance. On 45nm Perhaps it could be done without as serious of an impact but I doubt that AMD will follow that route

[ozzimark]

Sure, half dividers would be nice. But half multipliers still help a bit, as they too provide more different combinations of CPU and RAM clocks.

but the way you're stating it, it makes it seem like K8 doesn't have 1/2 multis right now


btw NN.. only 1ns of extra latency? i think it'd be a bit more

[_damien_]

but the way you're stating it, it makes it seem like K8 doesn't have 1/2 multis right now

What I meant is that even though there are no half dividers, the half multipliers provide better fine-tuning of the memory/CPU-speeds.