Will AMD ever regain the performance crown again?

[MDme]

With most if not all the C2D's getting very good OCs (3+ Ghz) range and the IPC advantage it has over current AMD CPUs, do you think that AMD will ever get the performance crown back from intel? I mean, just think about it: with current C2Ds being ~15-20% better than AMD clock for clock, a 3GHz C2D would be just like a ~3.6Ghz X2. Can't imagine AMD coming up with K8 like architecture that can clock like that. 65nm would help, but don't know if the other IPC improvements would help narrow the gap.

While I prefer AMD to intel, (though not a fanboy) I really hope AMD recovers because competition is good for us.

So share your thoughts and discuss. Please no flame wars just opinions, I know there is little if at all any data to project AMDs future offerings.

 

[hennethannun]

I think that AMD could definitely get back on top IF they can get get their 65nm process up and running without problems. K8L SHOULD get AMD back to where their top of the line processors are at least competitive with Intel, so long as it comes out fairly soon (say 6-9 months). AMD doesn't need to get all the way back to the top in one generation, but they need to make sure they are competitive with intel. If they can do that, they should be able to consolidate the market gains they made with K8.


the nightmare scenario for AMD is having serious problems with K8L and not being able to get it to market for a year or more. that will give Intel almost two full generations of lead time. And with Intel's already enourmous lead in market share, that would be disasterous for AMD.

 

[Kromis]

Now, I'm going to quote what I said some time ago about threads like these (I quoted a quote from a Magic card):

"Empires rise and fall, but evil is eternal"

Anything can happen. I think that AMD can become the usurper of the throne once again. But since I don't have any money to make a new rig, I really don't give a damn right now. Perhaps when I get a job and move out and stuff, then we shall see.

What is K8L again? Is that an architecture revision or something for the next core?

 

[tylerw13]

i honestly think they will...its always back and forth back and forth...after all that is technology!!! good for us as consumers....thats why having monopoly is a bad thing, that way you have companys always trying to outdo one another...isnt it good to have options!!!

 

[Furen]

I refuse to answer your poll simply because I do not have enough info to make an good assessment of whenit will happen. That said, I can almost bet that Intel's future architectures won't be such a huge improvement as Core. Core fixed many weaknesses inherent in the P6 microarchitecture and when compared to Netburst is a huge step forward. Going forward, however, Intel's redesigns will, very likely, be more gradual improvements upon its existing tech, whatever Intel wishes to tell us. AMD, on the other hand, still has room for lots of improvements. A big yet simple one is cache size. Load reordering, widening of L2 data paths, enlarging buffers, improving decoder performance and improving SSE perfomance are the other big things that we can probably expect to happen with K8L and may improve performance considerably. Just look at what the above did for Core 2. Whatever everyone wishes to tell you, Core 2 is Yonah on steroids.

 

[hennethannun]

K8L is what should be K9, ie the next derivation of AMDs micro-architetchure. for obvious reasons, K9 is not a good name for cpu architechture, so AMD wanted to go with K8.50 and finally settled on K8(Roman Numeral L). ie K8(50)

 

[Polish3d]

Not only is Core 2 around 20% faster per clock if not more on average (sometimes upwards of 50% faster) , but it overclocks like a mofo. People can take an X2 from 2.4 to 2.6 / 2.7ghz with a 4800+. An average OC for an E6600 is 1ghz HIGHER than that. Man I can't wait to get mine. And th y are debuting at nearly 1/3 the price of AMD's equivalents. I'm no big fan of AMD for how they charged (580 for my 3rd from top 4400+. What's an E6400 cost retail? 230 or so? LOL) .

Still, have to like AMD for putting out a quality product that forced intel to innovate. Where would we have been without AMD through crappy Prescott years... I hope they come back and beat intel with a better product and hopefully a better price for once

 

[harpoon84]

I would say yes, eventually, AMD will regain the performance crown. These things go in cycles, remember?

My memory is not that good but I think from the 90s it went like this:

Intel P2/P3 > AMD K6/K6-2
AMD K7 Athlon > Intel P3/P4 Willamette
Intel P4 Northwood > AMD K7 Athlon XP
AMD K8 Athlon 64 > Intel P4 Northwood/Prescott
AMD K8 A64 X2s > Intel P4-Ds
Intel C2D > AMD K8 A64 X2

History suggests AMD *will* strike back, whether it be K8L I'm not sure. The current gap in performance is quite large, ~20% clock for clock, plus C2D can clock pretty damn high if overclocking results are any indication (3GHz stock volts is almost standard). This suggests Intel could easily ramp up C2D clockspeeds in excess of 3GHz if they feel threatened.

 

[dmens]

Originally posted by: Furen
Going forward, however, Intel's redesigns will, very likely, be more gradual improvements upon its existing tech, whatever Intel wishes to tell us. AMD, on the other hand, still has room for lots of improvements.

so why does the K8 core have more room for improvement when compared to merom?

 

[Viditor]

Originally posted by: dmens

Originally posted by: Furen
Going forward, however, Intel's redesigns will, very likely, be more gradual improvements upon its existing tech, whatever Intel wishes to tell us. AMD, on the other hand, still has room for lots of improvements.

so why does the K8 core have more room for improvement when compared to merom?

Because of it's platform...it's quite possible to improve the K8 core to the level of C2D, and with the additional advantages of on-die mem controller and HT interconnects (not to mention an open platform for 3rd party development on those interconnects), the K8 should prove a superior overall system if & only if AMD can make core improvements (which to me seems likely, but as Furen says...we won't know till we see it).
This may indeed be counterbalanced by CSI in 2009, assuming Intel is successful there...but again we don't have enough info.

 

[hennethannun]

Originally posted by: Frackal
Not only is Core 2 around 20% faster per clock if not more on average (sometimes upwards of 50% faster) , but it overclocks like a mofo. People can take an X2 from 2.4 to 2.6 / 2.7ghz with a 4800+. An average OC for an E6600 is 1ghz HIGHER than that. Man I can't wait to get mine. And th y are debuting at nearly 1/3 the price of AMD's equivalents. I'm no big fan of AMD for how they charged (580 for my 3rd from top 4400+. What's an E6400 cost retail? 230 or so? LOL) .

Still, have to like AMD for putting out a quality product that forced intel to innovate. Where would we have been without AMD through crappy Prescott years... I hope they come back and beat intel with a better product and hopefully a better price for once

I'm not sure that C2d is quite 20% faster per clock than X2. the E6800 definitely outperforms the FX-62, but it is also clocked more than 100Mhz higher. I would say the performance advantage is probably 15-20% (so maybe 20% is the upper boundery, rather than the average). But you also have to remember that C2D is 65nm, while X2 is AMD is still on 90nm. Once K8 goes to 65, things will be slightly better (although i think C2d is still better than a 65nm X2).

But with the right architectural improvements (ie K8L) AMD could be very competitive. but i think that if K8L were coming anytime soon we would be hearing more about it from AMD. So K8L is still at least half a year off, and probably more.

as for prices, where have you been? the 4400+ retails for just of $200 now. and they are widely available. AMD is still fine in the low to mid-end market (although things will be rough for them once the C2D market settles down). they have almost completely ceded the high-end and enthusiast market to Intel for the moment. Aside from the FX series, there is not an AMD chip on the market that retails for more than $400.

 

[xenolith]

In the next year, AMD will have their work cut out for them in the performance market. Two years is a long time to hang on to the crown in this game. So who would vote never? Fanboys, ya gotta love'm.

 

[harpoon84]

Originally posted by: hennethannun
I'm not sure that C2d is quite 20% faster per clock than X2. the E6800 definitely outperforms the FX-62, but it is also clocked more than 100Mhz higher. I would say the performance advantage is probably 15-20% (so maybe 20% is the upper boundery, rather than the average). But you also have to remember that C2D is 65nm, while X2 is AMD is still on 90nm. Once K8 goes to 65, things will be slightly better (although i think C2d is still better than a 65nm X2).

If it's not, it's pretty damn close to it. There can be no exact figure as it all depends on the application and benchmark.

But when you figure that an E6600 @ 2.4GHz outperforms an FX-62 @ 2.8GHz (a 16.67% clockspeed advantage for the FX-62) I'd say ~20% is pretty accurate.

 

[Polish3d]

Originally posted by: hennethannun

Originally posted by: Frackal
Not only is Core 2 around 20% faster per clock if not more on average (sometimes upwards of 50% faster) , but it overclocks like a mofo. People can take an X2 from 2.4 to 2.6 / 2.7ghz with a 4800+. An average OC for an E6600 is 1ghz HIGHER than that. Man I can't wait to get mine. And th y are debuting at nearly 1/3 the price of AMD's equivalents. I'm no big fan of AMD for how they charged (580 for my 3rd from top 4400+. What's an E6400 cost retail? 230 or so? LOL) .

Still, have to like AMD for putting out a quality product that forced intel to innovate. Where would we have been without AMD through crappy Prescott years... I hope they come back and beat intel with a better product and hopefully a better price for once

I'm not sure that C2d is quite 20% faster per clock than X2. the E6800 definitely outperforms the FX-62, but it is also clocked more than 100Mhz higher. I would say the performance advantage is probably 15-20% (so maybe 20% is the upper boundery, rather than the average). But you also have to remember that C2D is 65nm, while X2 is AMD is still on 90nm. Once K8 goes to 65, things will be slightly better (although i think C2d is still better than a 65nm X2).

But with the right architectural improvements (ie K8L) AMD could be very competitive. but i think that if K8L were coming anytime soon we would be hearing more about it from AMD. So K8L is still at least half a year off, and probably more.

as for prices, where have you been? the 4400+ retails for just of $200 now. and they are widely available. AMD is still fine in the low to mid-end market (although things will be rough for them once the C2D market settles down). they have almost completely ceded the high-end and enthusiast market to Intel for the moment. Aside from the FX series, there is not an AMD chip on the market that retails for more than $400.

Don't ask me where I've been, read what I said. Now doesn't matter, AMD was forced to price it there.

When AMD was on top last year they released the 4800+ (equiv: E6700) for 1000 dollars, and their 3rd from top, 4400+ was about 580, nearly 3 times what you'll pay for Intel's 3rd from top, and the performance difference is larger this time around with Intel's new chip. I could have said it more clearly though

 

[dmens]

Originally posted by: Viditor
Because of it's platform...it's quite possible to improve the K8 core to the level of C2D, and with the additional advantages of on-die mem controller and HT interconnects (not to mention an open platform for 3rd party development on those interconnects), the K8 should prove a superior overall system if & only if AMD can make core improvements (which to me seems likely, but as Furen says...we won't know till we see it).
This may indeed be counterbalanced by CSI in 2009, assuming Intel is successful there...but again we don't have enough info.

merom isn't bus bound at 1P/2P, even if it uses the ancient fsb. 4P woodcrests already use better options. so that is a non-issue. p2p links are not force multipliers that enable greater leaps in performance. there's been no such thing for quite some time.

don't talk about platforms, it is such a dirty marketer word...

as for plugins, maybe they'll pan out, maybe not. but hyping something that may materialize in 2+ years as an advantage is crazy, imo.

 

[hennethannun]

Originally posted by: Frackal

Don't ask me where I've been, read what I said. Now doesn't matter, AMD was forced to price it there.

When AMD was on top last year they released the 4800+ (equiv: E6700) for 1000 dollars, and their 3rd from top, 4400+ was about 580, nearly 3 times what you'll pay for Intel's 3rd from top, and the performance difference is larger this time around with Intel's new chip. I could have said it more clearly though

Sorry, it's late where i am, and my post sounded a bit more insulting than I intended. didn't mean to offend.

as for the price strategy, I see your point, but i have to slightly disagree with it. AMD was introducing dual-core processing to the desktop market, and they have less production capacity than intel, so they really had no choice buy to charge a premium for the processors.

Intel has a lot more production capability, and they are introducing C2D into a market saturated with mid to high end dual-core processors.

In short, dual-core was not mainstream when X2 was introduced, and it is becoming mainstream now.

 

[buck]

I hope amd does take the crown back, i would prefer amd over intel but I want the better performance chip (as long as its not to expensive).

 

[gobucks]

i prefer AMD as well, but I have to say I'm glad Intel finally came out with a worthy product. AMD was getting a bit high-and-mighty over the last few years. it was especially annoying that they always seemed to make us wait forever for new technology to become affordable. every new product introduction (initial socket 754, which was DOA btw, socket 939, and x2) cost like $400+. Intel released a whole product lineup with all the new core 2 technology, and that's gonna force AMD to do the same in the future.

AMD will definitely regain the lead, but i don't think K8L will do it. the features won't help that much - the shared L3 will be slow, and i think they are even cutting the L1 to 64K, a huge mistake. Maybe K10 will add a wider execution path, shared L2 cache, and better branch prediction.

 

[DaveSimmons]

Some years ATI wins, some years nvidia does, some years AMD wins, some years intel does. Some times you feel like a nut, some times you don't.

 

[Viditor]

Originally posted by: dmens

Originally posted by: Viditor
Because of it's platform...it's quite possible to improve the K8 core to the level of C2D, and with the additional advantages of on-die mem controller and HT interconnects (not to mention an open platform for 3rd party development on those interconnects), the K8 should prove a superior overall system if & only if AMD can make core improvements (which to me seems likely, but as Furen says...we won't know till we see it).
This may indeed be counterbalanced by CSI in 2009, assuming Intel is successful there...but again we don't have enough info.

merom isn't bus bound at 1P/2P, even if it uses the ancient fsb. 4P woodcrests already use better options. so that is a non-issue. p2p links are not force multipliers that enable greater leaps in performance. there's been no such thing for quite some time.

don't talk about platforms, it is such a dirty marketer word...

as for plugins, maybe they'll pan out, maybe not. but hyping something that may materialize in 2+ years as an advantage is crazy, imo.

I'm not sure what your responding to here, but I don't think it's what I said...

1. I haven't heard that Merom was to be anything but 1P (same for the mobile version of K8L). And while they aren't bus bound, the latency of HT at 3k is certainly lower than the ancient FSB at even 1066.
2. There are no 4P woodcrests yet...
3. While P2P is not a force multiplier, it does reduce latency considerably (as does the on-die mem controller)...something that is highly beneficial to both K8 and C2D.
4. Plugins are already available (albeit only one that I know of), and IBM is already releasing HTX...so we are talking months, not years.
New IBM line with HTX

 

[Kromis]

Originally posted by: DaveSimmons
Some years ATI wins, some years nvidia does, some years AMD wins, some years intel does. Some times you feel like a nut, some times you don't.

Couldn't have said it better myself. :thumbsup:

 

[dmens]

Originally posted by: Viditor
I'm not sure what your responding to here, but I don't think it's what I said...

1. I haven't heard that Merom was to be anything but 1P (same for the mobile version of K8L). And while they aren't bus bound, the latency of HT at 3k is certainly lower than the ancient FSB at even 1066.
2. There are no 4P woodcrests yet...
3. While P2P is not a force multiplier, it does reduce latency considerably (as does the on-die mem controller)...something that is highly beneficial to both K8 and C2D.
4. Plugins are already available (albeit only one that I know of), and IBM is already releasing HTX...so we are talking months, not years.
New IBM line with HTX

i interpret what you said as c2d is restricted by the lack of p2p links, which is bs. fyi, i refer to all c2d as merom. there is a 4p woodcrest on the market right now, mac pro.

didn't xbitlabs did some tests on memory throughput on c2d, and the conclusion is that it is not high impact at all, as expected. there are so many more annoying bottlenecks than the bus. not to mention the bus bottleneck has the most options available to work around, as opposed to many fundamental chokes in the core.

link on useful plugins, please. is the cHT protocol even in use? if not, how would any plugin available now be any better than any other device using the pci bus?

 

[Viditor]

Originally posted by: dmens

Originally posted by: Viditor
I'm not sure what your responding to here, but I don't think it's what I said...

1. I haven't heard that Merom was to be anything but 1P (same for the mobile version of K8L). And while they aren't bus bound, the latency of HT at 3k is certainly lower than the ancient FSB at even 1066.
2. There are no 4P woodcrests yet...
3. While P2P is not a force multiplier, it does reduce latency considerably (as does the on-die mem controller)...something that is highly beneficial to both K8 and C2D.
4. Plugins are already available (albeit only one that I know of), and IBM is already releasing HTX...so we are talking months, not years.
New IBM line with HTX

i interpret what you said as c2d is restricted by the lack of p2p links, which is bs. fyi, i refer to all c2d as merom. there is a 4p woodcrest on the market right now, mac pro.

didn't xbitlabs did some tests on memory throughput on c2d, and the conclusion is that it is not high impact at all, as expected. there are so many more annoying bottlenecks than the bus. not to mention the bus bottleneck has the most options available to work around, as opposed to many fundamental chokes in the core.

link on useful plugins, please. is the cHT protocol even in use? if not, how would any plugin available now be any better than any other device using the pci bus?

1. Mac Pro is a 2P/4-core system, not a 4P system. The 4P C2D processors aren't due until mid 2007...
2. Link on the Xtremedata chip which drops in to a 940 socket (and yes, the cHT links are used).

 

[Regs]

I believe the K8 has a lot of room to improve. Look at all their bandwidth that it's wasting compared to Core. The K8L can deliver competitive if not more performance if it improves the memory subsystem. For instance, Johan De Gelas in an Anand articles explains how the K8 had a much lower memory latency than the P4 and about a 15-20% advantage over the Core. The Core however has a much lower cache latency as well as a much more efficient brand prediction unit to avoid cache mis-hits.

Then AMD will have to overcome Intel's micro-op fusion. I don't know how AMD will do it, if they even can, but they will have to deploy a trick of their own for their own architecture. Overcoming Intel's ability to decode will likely be one of their more challenging tasks. Adding another decoder without limiting bandwidth is no easy task.

 

[Furen]

Originally posted by: dmens

Originally posted by: Furen
Going forward, however, Intel's redesigns will, very likely, be more gradual improvements upon its existing tech, whatever Intel wishes to tell us. AMD, on the other hand, still has room for lots of improvements.

so why does the K8 core have more room for improvement when compared to merom?

Now, let me answer the question instead of letting Viditor answer for me:

The reason why K8 has more room for improvement is simply because it is crappier. The microarchitecture is pretty aged, if you think about it, as it is based on the K7. Sure, there were some improvements when the K8 was launched but these improvements did not really touch the backend at all and were relatively minor when taken as a whole. Kind of like how Yonah/Dothan/Banias were better performers than the P6 but nowhere near what Core accomplished. The memory controller IS big but AMD's buffers, data paths, L2 cache bandwidth, prefetchers, decode performance, execution unit performance, lack of load reordering and the like put the microarchitecture at a significant disadvantage. Intel's Core, on the other hand, does not have any glaring weaknesses you can point to. I'm sure there's still stuff that can be done to it to improve performance but you could say that it is the pinnacle of Intel tech RIGHT NOW.

EDIT: Regs, AMD has had micro-op fusion since the original K7 came out. If you mean MACRO-op fusion then I'm not sure how much of a benefit it really is. Just look at Core 2's x86-64 benchmarks, we don't see a huge weakness even though it lacks macro-op fusion in 64-bit mode.