X-bit labs

Started by: Mamisano | Date 08/21/04

Comments: 8 | Last Comment:  08/24/04

[1-6]

1. AMD needs to choose 1 socket and stick with it. I wonder if they can make ALL desktop processors on the 939 socket, but cripple the low-end chips via a bridge so that it can only use single-channel ram. Say a 939 board with 4 DDR slots, with a "value" chip only 2 would work in Single channel. Then later on, a user can drop in a "mainstream" 939 which would support dual channel and the other 2 dimm slots.

Yes you are right.
I still don’t know why I can’t use a 939 pin processor in a 940 pin socket.
I still don’t know why 754 pin processors don’t do on a 939 socket.
They should have designed the socket from the beginning to accommodate lower and higher pins processors like Socket 1 or 3 in the past.

2. They've made it clear before that this was the eventual direction, that 754 would eventually only be used by the absolute lowest in their cpu line, and even then it would be phased out. 939 is something they said months ago, would expanded and be with us a long time. Typically you introduce high end items first, then you start bringing in mid and lower range later, so the mobo makers 'should' have been in on this information before. They offer high end boards for P4's and mid and low, so the same goes here. Soon you'll even add to the chipsets, CK08, NF3, 2-3 Via chipsets and SiS as well, so you'll have a variety of chipsets to offer high, mid and low range boards upon.

The only disturbing thing is the entire lack of upping the speed grade. It's not a very bad idea to fill in mainstream products to increase market share (we'd all like to see AMD do well there), but the high end is where you demonstrate what your chips are capable of in the future. And for some reason, the only way they've been able to sample chips getting to 2.6 is by halving the cache size. To my thinking that's basically saying they have to make technical compromises to get higher speed.

Has AMD hit a barrier too and they just aren't saying anything? Next month we might be seeing high end PCI-E chipsets and GPU boards to go with them, and it will be a bit sad if there is nothing new from AMD at that time. Mind you, what they have performs superbly, but I am noting when they are leading the market they suddenly don't feel much pressure to keep moving the bar up. I think maybe, that's because they can't...

3. True, high-end is the demo of what they can do for the future...But its mainstream is where the money is made from consumers. AMD has a slight issue called Socket A...They need to get rid of it...Since it still makes sales in Asia. (Cheap but fast CPU).

I reckon AMD has hit the barrier (just like Intel and IBM). The entire CPU market has stagnated, and news about new CPUs is dull as watching grass grow.

You say "What about dual core?"

Dual core is more like an addmission by all CPU makers that : "Look, we're screwed. We can't scale any higher...We'll start making things with the ability to do multiple things instead of making it faster".

Dual core won't come at really high speed...AMD's Opterons and A64 FX will be dual core from 1.6Ghz to 2Ghz...That's it...Speed will be in the form of very small increments. (over 90% of today's games don't even support multiple CPUs or cores, so you'll get a slower system with very good multi-tasking abilities.)

4. Well, the Athlon 64's are still scaling pretty well. They 3800+ is a nice example. Before that, their highest (non FX cpu) was 3400+, and this is still with 130nm chips. Thats a pretty nice performance jump, isn't it?

Intel seems to be stuck though. I wonder if they'll ever reach above 4 ghz...

5. Don't look at AMD CPU's "rated speed"...AMD's rated speed means shit all to me. Look at the real speed they're going at.

They haven't varied much have they? 1.6 to 2.4Ghz...With various features removed or added on. (smaller or larger cache, single or dual channel memory).

Haven't seen 2.5 to 3Ghz...Why?
How come they don't scale up speed like they use to?

For Intel, its marketing and maunfacturing related...For IBM/AMD, its manufacturing...It becomes really obvious when the market slows to a grinding halt and companies announce things that don't matter much because it isn't a speed scale.

6. What is the real speed they're going at? The mhz number? Oh please, take some hardware classes.

Or just look at the benchmarks. The 3800+ is *way* faster than the 3400+, and *way* faster than a Pentium 4 at the same clock speed.

MHz means *nothing* alone. Combine it with the IPC, and you get a meaningful number, and funnily enough, IPC is what AMD has been improving, and that is why they use the rated speeds. True, they're not accurate, but they're way better than just looking at the MHz numbers.

The point is that AMD's processors are still improving performance, whether or not they gain clockspeed.

Exactly. Clock speed isn't everything. AMD has been increasing the speeds of thier CPUs through improving the arcitecture, not raw-clock-speed. For example: K7 - K8 Transtion... What changed? On the outside, clock speed remained the same (at launch). But on the inside, we have a nurmerous number of changes... To name a few, we find a new SSE2 unit, Improved Branch Prediction, more Cache, an integreated memory controller, and a slightly longer + improved 12 Stage Pipeline. And guess what, they almost doubled thier preformance in some applications and kept the same manufacturing proccess.

Overall, the Athlon 64 didn't scale well in Clock speed, but it greatly improved IPC.

[1-6]