Things AMD can do to take on Conroe...

F

freeloader1969

2[H]4U
Joined
Dec 13, 2003
Messages
2,238
What does everyone think AMD will have to do to be competitive against Conroe? Some suggestions would be great.

Here's what I was thinking. Take the K8L (with extra floating point units), add a 4mb L2 cache (2mb per core), possibly some L3 cache, and if it's even possible to do in a few months, add 2 more stages to the Athlon 64 core.

I'm in no way a CPU engineer, so go easy on me! :D

My logic was that the extra stages may help the existing arcitecture scale over 3ghz. I realize the L3 cache is wishful thinking due to 90nm fabrication. Maybe at 65nm we'll see some. I also don't believe that the switch to DDR2 is going to make that much of a difference in current performance. A64's have never been bandwidth hungry anyhow.

Thoughts, suggestions?
 
Cache is good, but would drive the cost up severely. The cpu itself is going to need to become more efficient.
 
Bona Fide said:
Here's one. A 65nm process.
Click to expand...

Might help with reaching higher clocks, but the conroe per clock cycle is still better, thats where the main problem comes from. Cache is used to try to get past this, as the P4's did to attempt to overcome netburst.
 
wee96 said:
Might help with reaching higher clocks, but the conroe per clock cycle is still better, thats where the main problem comes from. Cache is used to try to get past this, as the P4's did to attempt to overcome netburst.
Click to expand...

Well a 65nm process will *probably* allow for a more efficient pipeline design, with maybe fewer, shorter pipelines.
 
Bona Fide said:
Well a 65nm process will *probably* allow for a more efficient pipeline design, with maybe fewer, shorter pipelines.
Click to expand...

:) Hopefully.
 
Anyone have a good link explaining pipelines. I have a vague idea of what they do and understand that too many can slow things down due to a failed branch prediction.
 
Intel and AMD's branch prediction unit are now very accurate.
 
freeloader1969 said:
Anyone have a good link explaining pipelines. I have a vague idea of what they do and understand that too many can slow things down due to a failed branch prediction.
Click to expand...

Take a look here http://arstechnica.com/articles/paedia/cpu/pipelining-1.ars/1

Empoy I think Freeloader is talking about the 20 stage P4's that employed hyperthreading to over come an incorect branch prediction so the processor wasn't completely useless between the clock cycles.
 
hyperthreading was not meant to overcome the incorrect branch prediction. when you have a misprediction, you need to flush the pipe (affecting a lot of instructions being executed out-of-order) and then fill it out again. this causes the waster cpu cycles.
 
How long has the a64 been out now? Over 2 years? I'm willing to bet they already have something up their sleeve unless they've just been doing nothing engineering wise in that time. Yes, I know they're working on am2, but merely making a ddr2 compatible a64 doesn't take 2 years (I'd be surprised if it took much more than 3 or 4 months). I am definitely curious to see what they come up with.
 
bobzdar said:
How long has the a64 been out now? Over 2 years? I'm willing to bet they already have something up their sleeve unless they've just been doing nothing engineering wise in that time. Yes, I know they're working on am2, but merely making a ddr2 compatible a64 doesn't take 2 years (I'd be surprised if it took much more than 3 or 4 months). I am definitely curious to see what they come up with.
Click to expand...

Quantum CPUs ;)

that is what they have up their sleve..
 
SSlaytanic said:
what are nm processes?
Click to expand...

The size of the transistor in the CPU. Smaller sizes = cooler operation and the ability to pack more transistors in the same place. nm is a billionith of a meter right?
 
empoy said:
hyperthreading was not meant to overcome the incorrect branch prediction. when you have a misprediction, you need to flush the pipe (affecting a lot of instructions being executed out-of-order) and then fill it out again. this causes the waster cpu cycles.
Click to expand...

but adding two ALU's helped this sitsuation out, at least in theory right? (or am I completely forgetting what I read about the reasons behind hyperthreading) Thats what I'm trying to get at.
 
Well, it costs the same for every chip I figure, a Sempron costs as much as the FX for them to make, So why not make their top of the line cpu's cheap and start working on better ones in the mean time

And if all else fails: Give out free hard drives or video cards with your purchase of a cpu! :D
 
J_I_M_B_O said:
Well, it costs the same for every chip I figure, a Sempron costs as much as the FX for them to make, So why not make their top of the line cpu's cheap and start working on better ones in the mean time

And if all else fails: Give out free hard drives or video cards with your purchase of a cpu! :D
Click to expand...

Semprons only have 256/512 kb of l2 cache and only 1 core, opposed to an fx-60 with 2 cores and 1 mb of l2 cache for each core. They wouldnt be able to sell and fx-60 at a semprons price and make any money off it.
 
TruthInRuin said:
Semprons only have 256/512 kb of l2 cache and only 1 core, opposed to an fx-60 with 2 cores and 1 mb of l2 cache for each core. They wouldnt be able to sell and fx-60 at a semprons price and make any money off it.
Click to expand...


sure they would.
 
J_I_M_B_O said:
Well, it costs the same for every chip I figure, a Sempron costs as much as the FX for them to make, So why not make their top of the line cpu's cheap and start working on better ones in the mean time

And if all else fails: Give out free hard drives or video cards with your purchase of a cpu! :D
Click to expand...

AMD's operating overhead is higher than Intels. If they could:

1) Lower the average cost of making each chip to what it costs Intel

2) Catch up to Intel in process technology, maybe get to 45nm a year or so sooner.

3) Design a processor that is about 30% better IPC.

then they will catch up to Intel in both profit and performance per market share.

Before then, you are absolutely right, they need to:

A) drop the FX name so they can drop the prices on those. Seriously drop their prices on all regular models

B) close the baby fabs, layoff some workers, and move the rest over to Fab36 so they can eek out more productivity.

C) make up some market share by shifting their main target market over to China.

D) move as much of their operation over to China as possible so they can cut costs and evade taxes.

or be prepared to lose some major market share. That's the way to survive. Of course, I don't condone all of these moves. I would not mind being able to grab an X2 for a hundred bucks or so. That might just hold me over until the fabled K10 comes out.
 
stealthy123 said:
sure they would.
Click to expand...
Selling their $1000 processors for $80 will get them nowhere fast at this point in time.

Plain and simple, a company whose products are based upon cutting-edge technology must research and develop new competitive products....or die. Drastically cutting prices at this particular point in time will only eat up their profit margin (i can almost hear the moaning on wall street now). They can heavily cut prices on current cpu's when their new flagship product is launched (heavy profits are made with early adopters, but this often only serves to recoup R&D costs)...
 
SJetski71 said:
Selling their $1000 processors for $80 will get them nowhere fast at this point in time.

Plain and simple, a company whose products are based upon cutting-edge technology must research and develop new competitive products....or die. Drastically cutting prices at this particular point in time will only eat up their profit margin (i can almost hear the moaning on wall street now). They can heavily cut prices on current cpu's when their new flagship product is launched (heavy profits are made with early adopters, but this often only serves to recoup R&D costs)...
Click to expand...

Well, sell less for more profit or sell more for less profit. I'm pretty sure I read somewhere it costs less than 80 bucks for AMD to make a chip. If they sell more, their image remains on everyones mind. If they sell less...too bad. Either way, sometimes they can recoup R&D and make money. At times like what is sure to come, they need to cut their losses.
 
I'd say the best way for them to pick up performance is through those additional FP units they're supposed to be adding. If they could take that dual core back to essentially a single core and optimize for some SSE instructions they could probably do alright.

The main point being that if they could process a full vector in ONE cycle there would likely be a nice boost from any application that is processing vectors. I'm guessing this is what conroe did that is speeding it up in certain applications.

Most games and encoding applications tend to use the vectors and processing those efficiently as possible would probably help significantly. It would sort of be like a half-assed vector based processor. Even if they didn't include all the functionality of a true vector processor to cut down on transistors and keep die size down it would probably give a significant boost. I'd have to say most applications that are extremely CPU dependent would make heavy use of SSE and vectors. The SSE multiplication instructions are probably the only ones they'd really need to worry about. I could be wrong here. Add, subtract, divide, sincos and some of the other operations aren't going to be extremely prevalent in most applications and might be not worth the expense of additional transistors. Although they'd be more than welcome if they could occur.
 
Could someone explain to me how AMD managed to get on top? I know that the AMD64 thing really got them some killer power, and hell, I was there watching. I remember when it came out.

But really, if they are so much smaller and much lower funded, how the hell did they get on top? And how the hell did Intel let them stay there?

I love AMD and hope they stay on top, but I'm prepared to go Intel reluctantly if things come to pass.
 
apHytHiaTe said:
Could someone explain to me how AMD managed to get on top? I know that the AMD64 thing really got them some killer power, and hell, I was there watching. I remember when it came out.

But really, if they are so much smaller and much lower funded, how the hell did they get on top? And how the hell did Intel let them stay there?

I love AMD and hope they stay on top, but I'm prepared to go Intel reluctantly if things come to pass.
Click to expand...

1. Intel's Netburst architecture was, for lack of a better word, horrible.
2. AMD [quite brilliantly] integrated the memory controller onto the chip, effectively negating latency for CPU-memory information transfer.
3. AMD's chips run cooler and consume less power.

As a result of 2 and 3, coupled with 1, AMD emerged as the victor, with a 2.0GHz processor that could rival an Intel 3.2GHz processor.
 
the truth of the matter is not what amd hasn't done. but what they are keeping secret from the compitition until the time is right. Anybody forget about socket F. Thats more advanced and uses way more pins and uses pads like the socket 775's. But about 1200 pins. That will have consitterable performance differences.

Do we have any idea how good that socket could be?

Anyways, AMD isn't lazy, they been in 65nm reasearch for years more then intel has. They are just not telling what they have done. They always have surprized us with 90nm tech. Its cept upto 65nm tech for this long from intel. What does that say to you? Despite AMD still useing 90nm tech, its still better then current intel tech. That what you can buy now a FX60 beats a EX Edition using 65nm's or matches it. What ya thinks going to happen when AMD uses 65nm's? Oh boy good bye intel for awile. See what I'm saying.

AMD really keeps too many secrets. They will anounce some surprising things sometime soon when the new sockets are released. THey only showed their minimal product line... The 1st cpus, not much to expect from the start. Intel just plays to early showing their best stuff their ace and AMD then knows what to expect and prepare theirs to whatever and then blow intel away as its always been am I right about the history of AMD and intel yes AMD has ALWays every time beat intel even when intel is ahead. Read, you'll see its never for very long. AMD will always be better.
 
apHytHiaTe said:
But really, if they are so much smaller and much lower funded, how the hell did they get on top? And how the hell did Intel let them stay there?
Click to expand...

AMD used to make knockoffs of Intel chips back in the Pentium and earlier days. But eventually they started in on their own designs and had modest success with the K6 and K6-2 chips. It seems to me they got on top when they realized that, as the underdog, they could take some bigger risks. Intel decided they wanted long pipelines and that clockspeed would still be king. P4 was scheduled to be at 10GHz by now - obviously it didn't happen. AMD went the opposite route and did pretty well with their Athlon chips. You could at least get a better bang for the buck, even if the performance crown went back and forth between AMD and Intel. Then AMD pushed us into the 64 bit era, put the memory controller on die, and brought out the dual core chips. Intel had to follow along to compete, especially since their chips just weren't scaling in clock speed like they had planned.

It doesn't strike me that Intel has really innovated much in the last few years, and I'm happy to see that change. Better competition means better products from both companies.
 
This thread has the potential for comedy gold

The whole premise of "lets get some people together who know nothing about cpu design to explain what AMD needs to do" is just begging for it

Here is my contribution

I think AMD should install mini-machine guns on their cpus (nanoguns you might say) so whenever they detect a conroe on the net, they can "blow them away"
 
needmorecarnitine said:
Here is my contribution

I think AMD should install mini-machine guns on their cpus (nanoguns you might say) so whenever they detect a conroe on the net, they can "blow them away"
Click to expand...

Damn, man I was about to go to bed when I almost fell off my chair. I don't take any of this "Cornrow benchie AMDIntelSuXmYc0X" thing very seriously, mostly just to see who knows how to smell BS. But, posts like this is what I live for.
 
needmorecarnitine said:
This thread has the potential for comedy gold

The whole premise of "lets get some people together who know nothing about cpu design to explain what AMD needs to do" is just begging for it

Here is my contribution

I think AMD should install mini-machine guns on their cpus (nanoguns you might say) so whenever they detect a conroe on the net, they can "blow them away"
Click to expand...

Yea your cool :)
 
Try and stay on topic guys. I don't think we'll see nano machine guns anytime soon :D

needmorecarnitine...If you don't have any ideas, then don't contribute. Just skip over this thread. I don't think there's a single CPU engineer on this whole board, atleast no one has ever claimed to be.

Some interesting ideas so far. We have extra cache, stronger IPC, 1 pass vector processing and more FP units. Sounds similar to what Intel has done, minus the FP units.
 
I think once we get to quad core, AMD's architecture will be more efficient than Intels, as it appears the Conroe was designed from the ground up as dual core.

Until that point, hopefully AMD can close the gap with DDR2-800. I don't see them taking the lead again for a while tho.

I'm looking forward to price wars, it means I sell more computers. :)
 
You must first consider what a cpu is made of. The same stuff is used to make the FX as the sempron. If they start thowing FX's out for $100 while they work on the new cpu, there'll be a feeding frenzy of people buying the FX
Only draw back: Buisesses will always stick with intel

I don't know a whole lot about processors but they need to work on power efficient , cool, and cheap cpu with more power than the conroe and they'll be owning intel in no time
And then of course, target buisnesses. (A buissness would like to buy somthing that sounds professional, not FX, sounds like children's toy)
 
HmmmDonut said:
The size of the transistor in the CPU. Smaller sizes = cooler operation and the ability to pack more transistors in the same place. nm is a billionith of a meter right?
Click to expand...


Not really but you have the right idea.

The size is actually the width of a single bit of SRAM. When a new process size is attempted they are typically tested and tuned by producing SRAM. SRAM is rather simple and repetitive but this makes it easier to detect problems which is why it is used. Each transistor type have different sizes so SRAM width became the standard. Just for notes a typical SRAM bit has 6 transistors.
 
I have faith in AMD. Does everyone remember what the pulled off with the TBread A -> TBread B? It was a miracle and that's what AMD will need to best Conroe.
 
Well, now that I think of it, I don't really care what happens to AMD or Intel, as long as we get powerful processors what so we care?
For some reason I've always favored Intel, but I bought an amd because it's more powerful
 
Serge84 said:
Anyways, AMD isn't lazy, they been in 65nm reasearch for years more then intel has. They are just not telling what they have done. They always have surprized us with 90nm tech. Its cept upto 65nm tech for this long from intel. What does that say to you? Despite AMD still useing 90nm tech, its still better then current intel tech. That what you can buy now a FX60 beats a EX Edition using 65nm's or matches it. What ya thinks going to happen when AMD uses 65nm's? Oh boy good bye intel for awile. See what I'm saying.
Click to expand...

AMD doesn't have issues with 90nm because of their architecture. look at the pentium m, it was fine on intel's 90nm, but the p4 wasn't because of it's design.

AMD won't see the same impact as intel with switching to 65nm if you're comparing it to how the P4 improved.
 
But with the looks of the way intel has been going it has been said intel will be on 45nm either before, or very soon after AMD his 65nm.... think this maybe true ?

perhaps AMD has lost it's crown and has been sitting on it's butt - i do recall sometime ago on anand, or another more reputible site they did an interview with AMD who, perhaps were just keeping a lid on things saying they had no major changes in line from the A64 - the move to DDR2 was the only major implementaion in M2 and from what many have said the added DDR2 latencies could actually do more harm then good ......


All i have said is only hear say, no hard facts - may the best man win, but i do think some have far too much confidence in AMD - sure they hit Intel in the nut's this last 2 years, but Intel also got smart in those last 2 years and dropped the dead P4 architecture and went back to the old, considerably more efficient, pIII roots.
 
Didn't AMD just buy some fancy superfast cache memory company and wasn't there some talk about 2 and 4 meg cache per core? I would think 4 meg per core and integrated DDR 800 (possibly 1000) would take up a lot of slack and equalize things while they roll out the next big thing.
 
You must log in or register to reply here.
Back
Top