You think dual-core will come in s939 flavor?

K

krizzle

Gawd
Joined
Jun 28, 2004
Messages
902
With the dual-core hype and announcements just around the corner, I'm getting pretty anxious...
You guys think AMD will have s939 Dual Core CPU's by any chance?
Or will there be a whole new chipset revision, socket, and standards involved??
 
It looks like they are going to introduce the 940 part first, and then introduce the 939 after that...
 
they will definitely release dual core for 939, I believe AMD plans to stick with 939 for awhile.
 
there will be 939 dual core chips.
most current boards should only need a bios update to support them too.
 
Cant wait for dual cores..........I know my next upgrade...
 
Dual core should bring multiple advantages, other than the "Hyperthreading" style multitask performance. I assume, with s939, each core will have its own single chanel mem controller, and thus instead of one core handling a dual chanel controller, it will be two handling their own single channel. <-- seems to work great with s754's.

Thoughts?

Wait...
Why dual core on 940's??
You mean as in Opteron flavor? :confused:
 
Yes, they will be making an Opteron flavor as well. The dual core will be available for both server and desktop applications. Intel jumped on the bandwagon by proposing the same. I would think that each may have its own dual channel memory controllers. Isn't the dual core essencially two cores slapped together? Unless they are looking to save space, I think it would be pretty cool to have "quad channel memeory". Chances are you are right about the single controller per core, but its always nice to dream... =]
 
im just hoping apps wont have to be written from the ground up to support dual core.
 
Well guess what? They will. The application will need multithreading capabilities to take advantage of the second core, just like with HyperThreading. Otherwise, you'll notice a performance increase while multitasking under a multithreaded OS.
 
xonik said:
Well guess what? They will. The application will need multithreading capabilities to take advantage of the second core, just like with HyperThreading. Otherwise, you'll notice a performance increase while multitasking under a multithreaded OS.
Click to expand...

so its gonna take a few years before dual core is mainstream.
 
Yes, but when they do arrive, they will probably still be well ahead of the software at that point.
 
Jason711 said:
im just hoping apps wont have to be written from the ground up to support dual core.
Click to expand...
well not really. all programs support dual core, it's just whether or not the program properly splits itself up to take advantage of the extra core available. if not, pray that ms makes windows put different processes into different cores for you to generally speed stuff up ;)



Jason711 said:
so its gonna take a few years before dual core is mainstream.
Click to expand...
well intel kinda got the software people a head start with HT, so... dual core programs should be in full effect about a year after dual core comes out, if not before that

(note that this is merely a very educated guess, don't quote me on this later)
 
All apps have to do is compartmentalize tasks. It would be simple for most games to just have a separate process for AI and other menial tasks. The immediate benefit of SMP-on-Chip is P4 HT like multitasking as well as the ability to run a super cpu intensive app on one core while everything else uses the other core. So you could be encoding videos while playing Doom3 with zero performance hit.
 
(cf)Eclipse said:
well not really. all programs support dual core, it's just whether or not the program properly splits itself up to take advantage of the extra core available. if not, pray that ms makes windows put different processes into different cores for you to generally speed stuff up ;)




well intel kinda got the software people a head start with HT, so... dual core programs should be in full effect about a year after dual core comes out, if not before that

(note that this is merely a very educated guess, don't quote me on this later)
Click to expand...

perhaps by that time the proc will have come down to a managable price. *shrug*
 
The socket 939 chips will still only have a dual chanel memory controller, otherwise you're going to need a new socket and twice as many traces to memory, and that isn't very easy to do as they all have to be the same length as one another.
 
Jason711 said:
perhaps by that time the proc will have come down to a managable price. *shrug*
Click to expand...
they shouldn't be too expensive.. remember that the release of the opterons first will mean that the server market will take the largest price hit while amd is still tweaking the process. by the time they come to desktop use, they should have chips for the whole line of prices (minus low end.. <$200 to $300)
 
(cf)Eclipse said:
they shouldn't be too expensive.. remember that the release of the opterons first will mean that the server market will take the largest price hit while amd is still tweaking the process. by the time they come to desktop use, they should have chips for the whole line of prices (minus low end.. <$200 to $300)
Click to expand...

i hope your right.
 
Yeah, guess I'll stick with s754 untill dual core becomes worth it. And they are, indeed bringing dual core Opterons to the market first.
 
(cf)Eclipse said:
they shouldn't be too expensive.. remember that the release of the opterons first will mean that the server market will take the largest price hit while amd is still tweaking the process. by the time they come to desktop use, they should have chips for the whole line of prices (minus low end.. <$200 to $300)
Click to expand...
Haha, pardon my extreme skepticism.

AMD's costs are related to yields and die size. With any new processor design, yields are going to be a bit troublesome. So okay, the Opteron swallows up those costs, let's say. But still, we have to deal with a die that is more or less 200 mm^2 in area. That's Clawhammer 1 MB territory, and you remember those prices. Not only that, but AMD now would have a processor that is basically two CPUs in one. Like Intel, AMD would find this a great excuse to mark up that baby. I would expect similar prices as you see now, with an $800 to $1000 FX at the top, and $400-700 Athlon 64s for a while until they see fit to make a budget dual-core CPU.
 
i honestly wouldn't be surprised to see $1200 CPU's for the reasons stated above. Yields will be cut practically in half, maybe less with the 90nm process. These babies will fly, but so will all the money in your wallet.
 
When you can just about double your processing density with a proc and BIOS upgrade, that's much better than buying, installing, and transitioning to whole new servers. You pay for the privilage. Dual core on the desktop will yield its best improvements over the long-term as more pedestrian apps become SMP-optimized.
 
I cannot wait for DC A64's to come :D I miss my good old dual CPU rig (BP6 baby :D )
 
GodsMadClown said:
When you can just about double your processing density with a proc and BIOS upgrade, that's much better than buying, installing, and transitioning to whole new servers. You pay for the privilage. Dual core on the desktop will yield its best improvements over the long-term as more pedestrian apps become SMP-optimized.
Click to expand...

imagine how long a dual core/SLI rig will last you. :eek: :)
 
(cf)Eclipse said:
they shouldn't be too expensive.. remember that the release of the opterons first will mean that the server market will take the largest price hit while amd is still tweaking the process. by the time they come to desktop use, they should have chips for the whole line of prices (minus low end.. <$200 to $300)
Click to expand...

LOL your kidding right?

Its funny how people can have such short term memories :p The first Opterons launched about 4 months before the first A64's last yr. The prices for the clawhammers at launch were ridicuously high at launch last september. AMD hasnt launched a new cpu line at the $300 price point and under in how long?
 
I'd hate to see what the yield will be like on these things at first.. Dont' be expecting 2.6 Ghz Dual Core cpu's anytime soon.. we'll be starting over at 1.8 again..
 
The socket 939 chips will still only have a dual chanel memory controller, otherwise you're going to need a new socket and twice as many traces to memory, and that isn't very easy to do as they all have to be the same length as one another.
Click to expand...

It is theoretically possible that the dual cores will spilt the current dual channel memory subsystem into 2 single channel, 1 for each processor.
 
Retro Rex said:
It is theoretically possible that the dual cores will spilt the current dual channel memory subsystem into 2 single channel, 1 for each processor.
Click to expand...

and why not, the core doesnt even really need two channels.
 
My impresssion, from what I have read, is that this is not the case. The dual channel controller will be used as such. I don't think that this incurs any performance penalty, but it does simplify design by keeping the existing memory controller topology.
 
For a chip company, Dual Core is an excellent opportunity to make a rediculous amount of money. Consider:
They will be 90 nanometer, and I read in some magazine that they thus can produce ~70% dies per platter when compared to 130nm.
Instead of each core having their own ceramic plate and pins and copper heatspreader, you will only need one. Doesn't seem like much, but in volume manufacturing it will truly shine.
Just the gigantic e-penis attached to the term Dual Core sounds cool, and no doubt they will rape you for all they can offa that... like intel and their dirty Ghz... bah.

My forecast is that these nifty Dual-Cores will serve as the expensive flagships for quite some time... until AMD or Intel sees reason to make them mainstream. It's not as if singe-cores will be obscelete any time soon.

Ghz-wise, I don't see why they would cut down the speed. They still should operate at 2 - 2.6 ghz, since two 1.3ghz cores will never give the performance of one 2.6ghz core (ignoring multitasking advantages, that is.)

I just don't see how they can implement a whole 'nuther core without adding pins...
 
krizzle said:
Ghz-wise, I don't see why they would cut down the speed. They still should operate at 2 - 2.6 ghz, since two 1.3ghz cores will never give the performance of one 2.6ghz core (ignoring multitasking advantages, that is.)

I just don't see how they can implement a whole 'nuther core without adding pins...
Click to expand...

well, you have to remember, you're almost doubling the thermal rating of the processor.. two cores with a TDP of 68w in the same space? it's going to create alot more heat, and as such, they'll have to ramp down the clock speed. Not to mention simple yield problems with an entirely new chip, and you won't be seeing 2.6 Ghz on both cores anytime soon.. They've already said that the speed will be lower when talking about dual cores..

Also, you don't need to increase the number of pins, as the 940's and 939's were designed from the beginning to support dual core. They already have all the pins they need, AMD just has to drop the other core on..
 
Jason711 said:
and why not, the core doesnt even really need two channels.
Click to expand...
They're Distributed Processing designs, on demand bandwidth based not classical Intel sequential equal timeslices - "Symmetric Processing" designs.

The crossbar controller supports "simultaneous multiple data streams from multiple data sources to multiple data destinations", point-to-point channelized load balancing is automatic for AMD's Dual-Core with crossbar memory controller.

The crossbar controller is fully capable even if using a single memory channel for Dual-Core, the single memory channel will function then as if it had been dual-ported memory.

Pictorial examples below of "crossbar" AKA - NUMA (non-uniform-memory-architecture/access) by Mike Chambers of nvnews...

gf3_crossbar.gif


crossbar_s.gif
 
Well put.
But I don't think dual-core will have 2 cores the same size as an equivalent single core... e.g. they will have half the cache on each core, i think... and other halved components... no?

Bah, we will see.
 
no need for more pins


with the HT links, each core has two HT links, one cpu is connected to the other cpu intenally, one cpu is connected to the chipset. probably the same core that's connected to the memory, so one core is the one with faster access to the memory and system IO, the other is the bitch core. The bitch core is connected to the other core with it's HT link and can communicate with the entire system over that core.

It's the beuty of the K8 design.
 
Actually, I think there's going to be an internal crossbar that handles I/O with external logic. Both CPUs would connect to this crossbar, which in turn handles I/O with the outside world. At least, this is how it's done with multi-CPU Opteron systems, except the crossbar is off-die at the moment.
 
ROAR! when is this stuff coming anyway? I read all these dates... intel is this quarter, right? AMD is Q2 05?
 
Intel's solution might come out in Q3 2005, probably the Prescott-based dual core. AMD's dual-core Opterons are planned for halfway through next year, so maybe Q2/Q3. The dual cores from AMD for the desktop aren't coming out until the second half of next year. My guess is that you'll see the Athlon 64 FX version launched (paper?) in late Q3 or even Q4, with the Athlon 64 trailing behind that by a few months. Sound like fun?
 
I'm a bowl of mixed nuts (feelings) when it comes to dualcore.

When it comes to AMD, I think it's an amazing design. Like it was said by obiwansotti, the K8 was designed to easily scale from 1U to 2, 4, 8 and up. So, dualcore processors were inevitable. I don't believe there will be a bandwidth issue either. The K8 is a very bandwidth-efficient CPU, like the K7 was. The performance gain between a single channel and dual channel K8 setup is existant, but negligeable almost. I think the biggest boost between the two was 10%... Although it counts, it's not that big of a deal. And the HTT on the 90nm dualcore CPU will run at 5x200MHz, allowing even more bandwidth to the rest of the system and plenty for two. My only concern is memory bandwidth. I think there should be a push for new standard: faster DDR1 tech. Maybe 233MHz DRAM isn't a bad idea.

The Intel dualcore design is even more so flawed than the Prescott is. Intel claims that the core that their dualcore P4s will be using are different from your standard Prescott, but nonetheless. It's all about architecture at this point. The Pentium4 needs a lot of ressources (large cache, big bandwidth, etc) to flex its muscle. They're sticking two bandwidth hungry processors on one shared 1066MHz FSB and on a dual-channelled DDR533 (maybe DDR600) memory interface. What the hell are they thinking!?! The Prescott does well on FSB800 and Dual DDR400. In my opinion, you'd need more bandwidth. I can't say a shared 2MB cache won't work; it might. All we can hope for is that it won't cause more problems...

Now a dualcore PentiumM, that would work. I think that's the best thing that came out of Intel in a long time!

Then there's the heat issue that's yet to be seen... I can see it now: watercooling becoming a standard. I'd say it's something that Intel should be considering a tad more than AMD. The shrink to 90nm tech has brought down the power consumption of the K8 by quite a tad. I can see two 1.8GHz San Diegos working just fine with an elaborate air cooler.

Phew... I'm done
 
You must log in or register to reply here.
Back
Top