Cost reducing the X2

Vee

Senior member
Jun 18, 2004
689
0
0
This is an idea I'm just throwing out for your amusement and discussion.

I've had a few looks at Sempron64s lately. It's amazing how well they do for their price.
How well they do against the cheaper A64s.
Se64, Cheap 754 MB, 1GB, large hd, DVD-RW, 6600GT or X800GT, it all comes together at around $600. And that's a pretty good and capable system.

Is the maybe 20% - 25% better performance of a more 'edge' single core s939 A64 system at maybe twice the cost really significant enough?
It's an old question and you don't have to answer it. I don't want you to answer it. We have all come up with some answers to that, again and again. No, it's rethorical, and intended as background for this:

Singlecore clockrate is one kind of performance advantage. Dual core is another.
Let me put it like this - I think you're likely to be much happier with the dual core kind of performance advantage, in the long run.

So, doesn't that make the A64 look squeezed out of relevance, between the Se64 and X2?

Single core improves AMD's production volume, of course. And A64 single cores span the price range of interest for a large market group.

But are AMD moving towards dual core aggressive and fast enough?
The Manchester core is cost reduced, only 147 mm^2 compared to Toledo's 199 mm^2.
But is that going far enough?
A dual core reduced to 2 x 256KB L2 could maybe be ~121 mm^2, using a simple geometric estimate, and a 2 x 128KB L2 ~108 mm^2. And why be so particular about clockrate? Bring out 1.8GHz and 1.6GHz if it increases yields.
...And raise the price on X2 3800+. Because I suspect it's artificially low. For market positioning reasons. Not a good thing if you have to do that to your big seller.
 

Furen

Golden Member
Oct 21, 2004
1,567
0
0
Though making 256kB dual cores would make the die-size smaller (121mm^2 sounds about right), it would limit the CPUs to being semprons only. Most current Semprons are actually cache-disabled Venices, which gives AMD some flexibility on how to package them and allows it to change the balance between Semprons and Venices rather rapidly if the need arises. The same can be said about Toledo/Manchester cores. Most Manchesters out there seem to be Toledo cores, meaning that they physically have the full 1MB/core cache, which gives AMD the ability to package them as either Opteron dual-cores, X2 Toledos or X2 Manchesters. You could argue that the cache disabled Manchesters are Toledos with defective cache. While this is most certainly true on some chips, most of Toledos should have perfect caches unless yields are horrendous.
 

Vee

Senior member
Jun 18, 2004
689
0
0
No. AMD produce the Palermos and Manchesters precisely because they are smaller already on the die, thus yield and production goes up. Yes, there are defective cache disabled Toledos too, sold in the same package, 2 x 512KB, as Manchesters. And some retailers may market all 2 x 512 as "Manchester" regardless. But it's senseless to waste expensive production capacity producing cache to be disabled, and AMD tries to avoid that.

Edit: And it would not limit the CPUs to being "Semprons" only. (And what would that "only" be? And what would the "limit" be?) That's just a marketing game. It could just as well be an Athlon64 X2 3600+, or X2 3400+ and so on, just to confuse people even more.
 

Bona Fide

Banned
Jun 21, 2005
1,901
0
0
I'm guessing if AMD went forward with a dual-core Sempron (late 06-early 07) it would be a 2x256k part.
 

Furen

Golden Member
Oct 21, 2004
1,567
0
0
AMD is using the cache to differentiate the CPUs. They most certainly cannot do a 200MHz bump for every model, so they have decided that +512K of cache is about the same as a 200MHz speed bump. As much as we would like to get cheaper AMD dual-cores, most OEMs are more focused on absolutely rock-bottom prices, which even a Sempron dual-core would not be able to deliver. Another thing to think about is volume: AMD has no way of knowing with absolute certainty how well such a Sempron CPU would sell, so the design and validation effort may not be worth it for a low-margin SKU, which this would most-likely be. AMD's best bet to lowering prices is to ramp up dual-core production (which is not likely 'till FAB36 raises their production capacity), or to simply wait for the 65nm shrink, which should happen a couple of month's after Intel's.

Note: I'm calling it Sempron simply because of the cache size, AMD could call it whatever it wants and the above would still apply.
 

DrMrLordX

Lifer
Apr 27, 2000
22,939
13,024
136
Vee, I put together essentially that exact system back in March, buying everything new except the monitor and power supply and getting a 6800NU instead of a 6600GT. All told, it ran about $850 at the time, though I'm sure I could have done it cheaper had I shopped around a bit more. I definitely missed some good deals from Frys on harddrives.

It's a nice setup, but in certain games, it seems like either the 128 megs of RAM on my 6800NU or the 256k l2 cache on my OCed 2800+(2.3 ghz) hold me back. Either that, or it's just badly-coded graphics engines. Rome: Total War still runs like a dog, and Star Wars Galaxies also had framerate problems. However, all the FPS titles I tried, such as Chronicles of Riddick and Farcry, ran beautifully.
 

Vee

Senior member
Jun 18, 2004
689
0
0
Originally posted by: DrMrLordX
It's a nice setup, but in certain games, it seems like either the 128 megs of RAM on my 6800NU or the 256k l2 cache on my OCed 2800+(2.3 ghz) hold me back. Either that, or it's just badly-coded graphics engines. Rome: Total War still runs like a dog, and Star Wars Galaxies also had framerate problems. However, all the FPS titles I tried, such as Chronicles of Riddick and Farcry, ran beautifully.

Yes, it's quite possible the cache is hurting.
You should be able to test if it's the videocard's 128MB easy enough. Just turn off some options that cost a lot of memory. Like AA, for instance.

It's also quite possible you're hurting because the workload is not distributed on two cores. Other workload interferes with graphics engine. Even if two cache starved cores would compete more for the memory bus, I think it should overall do better.

Furen: Your second argument is very good, and I belive in every point.
Volume is AMD's problem, and I think that is why they only considered desktop dual cores late, and after Intel had announced theirs. I also think the Toledo X2 represent an effort that has been redirected from intended Opteron chips to the enthusiast market.
Chip area reduction is however still just as important on 65nm.
2 X 250 would CPU spec. wise, be 512KB L2, and thus an Athlon64 X2 :p

I seem to have been slightly misunderstood. I didn't ask for dual core Semprons. I was suggesting AMD hadn't pushed dual core aggressively enough, early enough, down market. And they can't do that without reducing chip area.

(I'm not asking for cheaper X2s from a personal aspect.)
 

Leper Messiah

Banned
Dec 13, 2004
7,973
8
0
Its more or less a paradigm shift for AMD, one that may not be entirely welcome among many enthusiasts. They've gone from being first in price by a fair ammount and being a little bit behind in performance, to being first in performance, but with more expensive SKU's. They have the respect of most of the DIY crowd, and businesses are starting to flock to them too. Of course, in AMD's eyes, they can command higher prices due to this respect. A lot of us forget that AMD is in this business to make money, not get good will of people.
 

eastvillager

Senior member
Mar 27, 2003
519
0
0
AMD doesn't seem interested in making low-end dual core procs, at least not while they've got two viable alternatives in that space already.