Who thinks AMD will be out of desktop CPU market by 2014?

[IntelUser2000]

Sun's Rock chip was innovative with OoO retirement, but it was rumored it didn't work in real life, then it got canned.

The question is whether Bulldozer is like Barcelona, with AMD making bad decisions in timing, or the shared module architecture is flawed fundamentally. The only thing is we won't know either until AMD, or someone else gets it working properly.

I wonder if the latter is even true, when AMD has so far deviated massively from general industry processor design methodologies. Like opting for clustered multi-threading type of deal, when IBM, Intel, and Sun, opted for much simpler ones like SMT and SoEMT, and even FMT.

Another term for "innovation" that doesn't work might just be a "bad idea".

 

[Ajay]

The question is whether Bulldozer is like Barcelona, with AMD making bad decisions in timing, or the shared module architecture is flawed fundamentally. The only thing is we won't know either until AMD, or someone else gets it working properly.

I wonder if the latter is even true, when AMD has so far deviated massively from general industry processor design methodologies. Like opting for clustered multi-threading type of deal, when IBM, Intel, and Sun, opted for much simpler ones like SMT and SoEMT, and even FMT.

I'm thinking the architecture is the problem, given what I've just read over at Xbitlabs**: http://www.xbitlabs.com/news/cpu/display/20111012151052_AMD_Aims_to_Improve_Bulldozer_Performance_per_Watt_by_Up_to_50_by_2014.html

AMD is setting the bar pretty low - a 10-15% improvement over a one year period would be pretty decent, IFF* they weren't already 10-15% below where they need to be now.

I was seriously hoping that 'Piledriver' would show at least a 20% increase under the assumption that AMD would be able to significantly tweak the architecture, remove any major bottlenecks and GLOFO would improve the process (better yields, more headroom). Given that they are only planning to hit a 15% improvement at best, I think the architecture must be fatally flawed (visa vi Intel).

Another possibility is that after playing the Game of 'over promise and under deliver' - AMD could be switching to a more sensible 'under promise and over deliver' bit, but I'm probably just delusional.



*If and only if
** This is also in the AT review - apparently I forgot seeing it, duh.

 

[Ajay]

Good point. Add to that the improvements by using recompiled software.

I cannot for the life of me figure out why AMD doesn't develop a post compile optimizer for the Intel compiler. Basically, (if arch == amd.fx) AMD_ARCH = true;
And then add a reasonable number of if statements to the compiled code that would use AMD optimizations instead of Intel's in performance critical code segments.

 

[Olikan]

I'm thinking the architecture is the problem, given what I've just read over at Xbitlabs:

in many forums i'm reading that it's the cache.l2 cache is as slow as a l3 cache.

So, every time that the l1 cache miss a data, bulldozer loses a ton os cicles waiting for the l2

 

[IntelUser2000]

in many forums i'm reading that it's the cache.l2 cache is as slow as a l3 cache.

So, every time that the l1 cache miss a data, bulldozer loses a ton os cicles waiting for the l2

It's only slow because its so large. Phenom II had 512KB L2 per core, Bulldozer shares a 2MB one, and from the way the die shot shows, that's same as the L3 with 8MB spread over 4 locations. Latency is stil lower than L3 so that's still better than going to L3.

 

[bryanW1995]

A lot of people seem to be thinking this is the first time a new CPU architecture has been same/slower than the previous.

Need anybody mention the Netburst fiasco? The P4 had MUCH lower IPC than the P3 did. And was at first, slower. But later down the road once they started clocking up, they were much faster than P3's. Not on a clock for clock basis, but over all.

AMD is going to be just fine. Intel does not want AMD to go away. Because then they end up with the government looking at them pretty closely which could end up in having lots of little Intels. One for SSD's, one for CPU's, one for Chipsets, etc etc.

The problem with BD isn't that it's a slow and disappointing release, it's that BD was the hope/dreams that was announced when barcelona flopped in 2007 (with a projected release for BD of q4 2009 back in those days). Now, we're seeing a similarly aggressive improvement schedule for BD that, frankly, has almost no chance to succeed, either. AMD's only shot at hitting that schedule in my mind is to take the A team off the gpu/llano/bobcat/coke machine/whatever it's working on and give them the task of fixing this monstronsity, kind of like how jhh moved everyone around to fix the issues with fermi I and succeeded beautifully.

 

[bryanW1995]

Wrong.

http://www.anandtech.com/bench/Product/434?vs=53

AMD FX may not be entirely competitive with current generation Core i7 SB's but it wipes the floor with your moms Q6600.

That was a bit tongue-in-cheek, I admit. However, the Q6600 is a LOT better overclocker than 8150, and I don't run that Q6600 at stock. I've had it as high as 3.6, though it's down lower now just b/c mom doesn't need to browse her email any faster.

 

[bryanW1995]

Not even Intel would want AMD out of the CPU market. I don't think they are going anywhere. They will more than likely never beat Intel in the desktop CPU market (for any long period of time) because Intel has an astronomically huge advantage when it comes to R&D, manufacturing and money (something a lot people seem to forget). Intel needs AMD to stick around so they have some sort of competition.

I don't think it matters anymore. ARM is ramping up enough that by the time AMD withers/dies and spins ATI back off, there will be a mid/low end desktop and mobile/nettop replacement available. And ARM will be a lot tougher to fight off than AMD. I wonder what intel would pay to buy/kill ARM right now?

 

[Ajay]

That was a bit tongue-in-cheek, I admit. However, the Q6600 is a LOT better overclocker than 8150, and I don't run that Q6600 at stock. I've had it as high as 3.6, though it's down lower now just b/c mom doesn't need to browse her email any faster.

And the Q6600 didn't need ~500W to run @ 3.6GHz (which is what the FX uses @ 4.8GHz)

 

[IntelUser2000]

Voted No, because the graphics part with Llano and such will keep them going for a while. Maybe later. Even then bankruptcy doesn't mean they disappear.

 

[OVerLoRDI]

BD on the desktop is a turd. It may actually rock in a server. I hope so for AMDs sake, they need all the wins they can get.

Why is that? Sure lots of cores, but the power consumption is huge. Am I missing something?

 

[Ajay]

Voted no because of Llano, ATI and speculating that Trinity will be very successful in the consumer desktop market (so long as AMD delivers on time).

 

[Red Storm]

I don't think it matters anymore. ARM is ramping up enough that by the time AMD withers/dies and spins ATI back off, there will be a mid/low end desktop and mobile/nettop replacement available. And ARM will be a lot tougher to fight off than AMD. I wonder what intel would pay to buy/kill ARM right now?

ARM CPUs can't run any existing PC, Mac, or server applications, so it's not going to replace AMD in the desktop and server space.

 

[Olikan]

Why is that? Sure lots of cores, but the power consumption is huge. Am I missing something?

lower clocks

 

[Ajay]

Why is that? Sure lots of cores, but the power consumption is huge. Am I missing something?

Well, since the server chips will run at a lower frequency, GloFo should be able to tune it's process towards lower power consumption (vs higher performance). This could make a big difference. As IntelUser2000 has pointed out b/4 - typically, from a given process, you can get higher performance or lower power usage. I hope for AMD's sake, that GloFo's process yields much better performance/watt on it's server CPUs.

 

[RedString]

@OP, you are aware there is another end of the market don't you? AMD is still doing fine with the budget builders. I seriously doubt they will be out of CPU market by 2014.

 

[Rifter]

@OP, you are aware there is another end of the market don't you? AMD is still doing fine with the budget builders. I seriously doubt they will be out of CPU market by 2014.

I am, however with Intel launching the Pentium SB chips they are also destroying AMD in the low end now as well. AMD is only left with the under $99 market now and i dont see how they can make any money selling chips for $100 or less.

 

[hasu]

Lets fast forward a year.

We have ivys out that are 22nm and consume 50% less power at same clocks as sandys.

AMD tries to respin bd and improves power draw by 15%(still way behind 32nm sandy)but clocks them up to compete with 2600-2700ks.

another years goes by and nothing but pildriver delays,intel puts out high end ivys Es and kills the top end market.

6 months later we get haswell and its 20% faster than ivy with even better power consumtion.

What happens to AMD next?Do they give up or try to compete with 22nm haswell knowing that intel will have a shrink of that 8-14 months after launch.

Oh, boy! I am not worried about AMD at that time. I am worried what will happen to all of us then? We all will take a second mortgage to upgrade to the then new Intel CPU resulting in a brand new recession. And if AMD is out of business by then, we can all use that new Intel CPU for the next century or so.

 

[hasu]

I am, however with Intel launching the Pentium SB chips they are also destroying AMD in the low end now as well. AMD is only left with the under $99 market now and i dont see how they can make any money selling chips for $100 or less.

May be the only CPU which matters by then would be the sub $99.
Once you have all the mega horse-power game consoles and all your documents in the cloud, these CPU wars and trolling might just fade.

Edit: Sorry to be so Pro-AMD. When I see so much of energy against AMD, I dont think that it is natural.

 

[Onceler]

They should market the "8 core" for what it really is a quad and they should release a true 8 core if they want to remain competitive.

 

[lOl_lol_lOl]

AMD is continuously searching for market niches yet unexplored by Intel in an effort to survive. And Intel almost invariably dominates those areas once it takes them seriously.

- Value at the sub $200 range : Intel occupied this segment with SB Pentiums and i3 - i5.

- Integrated GPUs : only matter of time before Ivy's DX11 appears.

- Intel can even kill off AMD's server market share if Ivy's perf and efficiency gains can be implemented effectively in the server market.

Where does AMD go from here?

 

[Manoa]

Wrong.

http://www.anandtech.com/bench/Product/434?vs=53

AMD FX may not be entirely competitive with current generation Core i7 SB's but it wipes the floor with your moms Q6600.

that's BS, it doesn't beat a core 2 duo, per core per clock, that's why a 4-core SB beats 8-"core" BD, because it is more than 50% faster per core per clock than BD, it all starts with the per core per clock performance and then (if you are lucky to have such code that uses many cores) scales to as many as the cpu has. that's the first math to do before even asking how much code is so parallel, the point being IPC is way more important than cores, because IPC improves everything regardless, and more cores only improves parallel applications, the only reason it shows as better is because it is higher clocked - you can't compare 2.40 static to 3.60 with turbo to 4.20.

see for yourself: http://www.anandtech.com/bench/Product/54?vs=187