AMD to Disclose Details About Bulldozer Micro-Architecture in August

[informal]

This is just a HotChips presentation,relax guys.

[Manicdan]

Another thread about AMD versus Intel, just what I needed! Thanks guys! You're awesome!

yup, took less than 10 posts to go from, AMD makes an announcement, to, my chip is bigger than your chip

both brands have cpus that are great at their own thing, get use to the fact that you will never see one have a massive advantage unless you have a very small niche your trying to fill. since pricing is set so there is always some competition.

[Hornet331]

And what if AMD could clock an 8 core (4 modules) Bulldozer at 5 GHz while only consuming 95 Watts? And it does this while having about 95% of the IPC of AMD's current Phenom II chips. That's what I'm getting at. IPC is only part of the story.

I'm not claiming that Bulldozer will be anything like that, just that there's more than just IPC to getting great single threaded performance.

Sry this is whichfull thinking, without abandoning x86 thrers no way the can achive 5ghz with a modest thermal envelope and that much cores.

[Chumbucket843]

yup, took less than 10 posts to go from, AMD makes an announcement, to, my chip is bigger than your chip

both brands have cpus that are great at their own thing, get use to the fact that you will never see one have a massive advantage unless you have a very small niche your trying to fill. since pricing is set so there is always some competition.

almost every thread on amd or intel becomes some kind of argument for why one is better. its very annoying. i think some people need to learn how to enjoy tech.

[-Boris-]

No it isnot. It is shifting fast now that DX11 is out. All development environments and languages focus on parallel execution and are getting more support for parallel development.

A quadcore at 3GHz will have a significant lead over a hexacore at 2GHz or an octacore at 1.5GHz for a very long time in games. Even if they have the same IPC/core and theoretical performance.

amd spent more then 5 years designing the chip from the ground up .... yes it will be more then competitive....

Just as Intel did with Pentium 4? Except that Intel had tons of more resources of course.

[keithlm]

IPC is Intel's weapon right now, so it can't be ignored in any honest debate, especially since a quad-core 2.66 GHZ part is beating a hexa-core 3.2GHZ part. The point is moot anyways since the competitor product has higher ipc and can overclock even better with non-extreme cooling.

You mentioned "honest debate" in the same sentence where you then pretend that the chips are always running at their base clocks. If IPC is as important as you claim, then we must know the true running frequency in all benchmarks; otherwise any debate is worthless.

[QuadDamage]

Bulldozer? What Bulldozer.... Oh yeah Bulldozer i almost forgot about it. I'm not holding my breath.

[OhNoes!]

You mentioned "honest debate" in the same sentence where you then pretend that the chips are always running at their base clocks. If IPC is as important as you claim, then we must know the true running frequency in all benchmarks; otherwise any debate is worthless.

Ooops, here comes the TURBO debate again. I thought that argument would disappear with AMD's own rather aggressive TURBO implementation on Thuban compared to bloomfield. Anyway, what's the argument here, Intel has the ipc lead? Clock for clock, even penryn cores are faster than deneb cores.

[savantu]

And what if AMD could clock an 8 core (4 modules) Bulldozer at 5 GHz while only consuming 95 Watts? And it does this while having about 95% of the IPC of AMD's current Phenom II chips. That's what I'm getting at. IPC is only part of the story.

I'm not claiming that Bulldozer will be anything like that, just that there's more than just IPC to getting great single threaded performance.

??

IPC and frequency are conflicting aproaches. You can't have both, it's one or the other. Your theoretical Bulldozer is like having a Pentium 4 with the IPC of Core 2. Pigs will learn to fly sooner than you will get such a CPU.

To get a lot of IPC you need very wide cores with lots of execution units, complex decoders able to extract the parallelism out of the instruction stream and lots of buffers to keep a mountain of data in flight through the chip. All that limits the frequency you can get. In other words, if your goal is IPC, you give up on frequency. Complex circuits clock badly and burn a lot of power.

The best example is Itanium. Extremely wide, has typically 1.5-2x the IPC of Core/Nehalem but also clocks 2x as low. That's inspite of the fact that Intel trying to reduce complexity as much as possible, by making it in-order and moved the task of finding parallelism out of the chip and into the SW ( compiler ).

The other aproach is frequency. Netburst is a fine example, make it as narrow as possible, have half or two thirds of the execution units of other CPUs, but clock as high as possible. Do some clever stuff to hide cache miss penalty and raise unit utilizsation ( SMT ) and you have a speed racer design. IPC can't be high, you have fewer decoders which are simpler, few execution units and miss rate penalties are huge.

Middle aproach is a beefed up Nehalem, the current one leans more to the fat core, high IPC, than speed racer. Altough if you optimize it by hand as Intel does, you can get some impressive frequency too. But that's only possible if you have a few hundreds years of manhours available.

[-Sweeper_]

Anandtech is not middle of world . U read maybe 1-5 review, i read 40. TRy example encoding videos 6x in the same time and u will see diference. Now im in work, but later at home, can u send PM with reviews. Definetely, x6 1090T is in real multi thread aplication compared with i7 965 +-.PS: games are ok, but its not optimalized for more than quadcores (i mean not 10% load at others cores, its nothing). Watch in games at x4 965 and x6 1090T. 965 BE will better.

well, they are actually neck and neck in video encoding/3d rendering at the same clock, and bloomfield takes a pretty big lead in most of other tasks

http://img526.imageshack.us/img526/2295/img0028754.png
http://img249.imageshack.us/img249/3104/img0028843.png

[Helmore]

Sry this is whichfull thinking, without abandoning x86 thrers no way the can achive 5ghz with a modest thermal envelope and that much cores.

I'm only trying to say that IPC is only part of the story. We have pretty much no idea how Bulldozer works. What if Bulldozer has separate clock domains within a core? Not saying it will, but AMD might have just come up with something really innovative for all we know. Let's be realistic, what do we really know about Bulldozer? Not much at all AFAIK.

[cegras]

??

IPC and frequency are conflicting aproaches. You can't have both, it's one or the other. Your theoretical Bulldozer is like having a Pentium 4 with the IPC of Core 2. Pigs will learn to fly sooner than you will get such a CPU.

To get a lot of IPC you need very wide cores with lots of execution units, complex decoders able to extract the parallelism out of the instruction stream and lots of buffers to keep a mountain of data in flight through the chip. All that limits the frequency you can get. In other words, if your goal is IPC, you give up on frequency. Complex circuits clock badly and burn a lot of power.

The best example is Itanium. Extremely wide, has typically 1.5-2x the IPC of Core/Nehalem but also clocks 2x as low. That's inspite of the fact that Intel trying to reduce complexity as much as possible, by making it in-order and moved the task of finding parallelism out of the chip and into the SW ( compiler ).

The other aproach is frequency. Netburst is a fine example, make it as narrow as possible, have half or two thirds of the execution units of other CPUs, but clock as high as possible. Do some clever stuff to hide cache miss penalty and raise unit utilizsation ( SMT ) and you have a speed racer design. IPC can't be high, you have fewer decoders which are simpler, few execution units and miss rate penalties are huge.

Middle aproach is a beefed up Nehalem, the current one leans more to the fat core, high IPC, than speed racer. Altough if you optimize it by hand as Intel does, you can get some impressive frequency too. But that's only possible if you have a few hundreds years of manhours available.

What are you talking about? The fastest P4 released was 3.4 ghz (from wiki?), while core i7's under turbo (e.g. single core to single core) can approach that speed. You seem to have handily ignored things like transistor shrinkage as well as better design ...

[Manicdan]

AMD in thread title, and the trolls just swarm on in. lock please, the announcement has nothing to really discuss, and its been horribly derailed so quickly

[keithlm]

Ooops, here comes the TURBO debate again. I thought that argument would disappear with AMD's own rather aggressive TURBO implementation on Thuban compared to bloomfield. Anyway, what's the argument here, Intel has the ipc lead? Clock for clock, even penryn cores are faster than deneb cores.

You mention that IPC is extremely important and then you go ballistic when somebody points out that you can't blindly claim the base clock frequencies.

There is no debate: You must know the true frequency of both chips or you are not comparing IPC.

[Sn0wm@n]

Just as Intel did with Pentium 4? Except that Intel had tons of more resources of course.

your point is???

[OhNoes!]

AMD in thread title, and the trolls just swarm on in. lock please, the announcement has nothing to really discuss, and its been horribly derailed so quickly

I wonder why people only complain when it comes to AMD threads? Go look in the "Sandybridge," "Westmere-EP," and "Westmere-EX" threads. No one is complaining. This thread is about an announcement. What possibly can you discuss, possible dates of the actual announcement? No chip exists in a vacuum, every chip design is inevitably going to be compared to existing designs. The discussion right now is about "IPC," which is quite relevant to bulldozer since this is the one area AMD is lagging heavily (and is only competitive by ramping up frequency and adding more cores, plus aggressive pricing) IPC is very relevant imo.

[Helmore]

??

IPC and frequency are conflicting aproaches. You can't have both, it's one or the other. Your theoretical Bulldozer is like having a Pentium 4 with the IPC of Core 2. Pigs will learn to fly sooner than you will get such a CPU.

That's not what I said. I said, lower IPC than Phenom II while clocking higher and Phenom II has a lower IPC than Core 2 AFAIK.

To get a lot of IPC you need very wide cores with lots of execution units, complex decoders able to extract the parallelism out of the instruction stream and lots of buffers to keep a mountain of data in flight through the chip. All that limits the frequency you can get. In other words, if your goal is IPC, you give up on frequency. Complex circuits clock badly and burn a lot of power.

You could also go for a more hybrid approach, like double clocking those parts in a core that make sense. Clock domains within a single core in other words. You could for example run the schedulers and execution units at double the clockspeed of the fetch and decode stage. I'm not saying they will, but it's another approach.

As I just said, we know very little about Bulldozer and there is a very slight chance that AMD may really surprise us. I'm just being cautiously optimistic.

[Helmore]

I wonder why people only complain when it comes to AMD threads?

The reason is quite simple, AMD is the underdog and many people will always pity the underdog.

[Manicdan]

I wonder why people only complain when it comes to AMD threads? Go look in the "Sandybridge," "Westmere-EP," and "Westmere-EX" threads. No one is complaining. This thread is about an announcement. What possibly can you discuss, possible dates of the actual announcement? No chip exists in a vacuum, every chip design is inevitably going to be compared to existing designs. The discussion right now is about "IPC," which is quite relevant to bulldozer since this is the one area AMD is lagging heavily (and is only competitive by ramping up frequency and adding more cores, plus aggressive pricing) IPC is very relevant imo.

i dont complain often, but what caught my attention was how quickly it was derailed. notice the first thing you said was that its AMD fanboys that are crying, instead of pointing out reasons for keeping it open, you decided to start your post by attacking a group of people.

and i dont see discussions. i dont see anyone talking about design ideas to make things better. all i see is people trying to point out that what they have is better than what others have, a battle of epeens.

[gosh]

A quadcore at 3GHz will have a significant lead over a hexacore at 2GHz or an octacore at 1.5GHz for a very long time in games. Even if they have the same IPC/core and theoretical performance.

Hexacores at 2 GHz ?
New games that need performance will take advantage of parallel execution.

[OhNoes!]

i dont complain often, but what caught my attention was how quickly it was derailed. notice the first thing you said was that its AMD fanboys that are crying, instead of pointing out reasons for keeping it open, you decided to start your post by attacking a group of people.

and i dont see discussions. i dont see anyone talking about design ideas to make things better. all i see is people trying to point out that what they have is better than what others have, a battle of epeens.

I said "people" not "fanboys." It's a trend, a glaring one. By the way, what would you rather discuss?

[informal]

Again, people could you just relax for a second.Like Helmore said,very little is known about how Bulldozer module works ,let alone the IPC relative to Deneb or Bloomfield.Wait for August for more details and then we can discuss based on facts.

[JF-AMD]

yup, took less than 10 posts to go from, AMD makes an announcement, to, my chip is bigger than your chip

both brands have cpus that are great at their own thing, get use to the fact that you will never see one have a massive advantage unless you have a very small niche your trying to fill. since pricing is set so there is always some competition.

Yep, couldn't agree more.

Until you have benchmarks from both, all of this is just noise. There isn't anyone that has access to both parts to be able to make an honest statement.

The funniest thing I read is "it takes 6 of your cores to match 4 of Intel's". I guess I could have said "it takes 8 of your threads to match 6 of AMD's" but let's face it, those arguments aren't going anywhere.

Patience, 2011 will be a good year for everyone.

[QuadDamage]

What are you talking about? The fastest P4 released was 3.4 ghz (from wiki?), while core i7's under turbo (e.g. single core to single core) can approach that speed. You seem to have handily ignored things like transistor shrinkage as well as better design ...

Nah man, 3.8Ghz was the fastest P4, there was also a 3.73Ghz EE. Intel wanted to release a 4.0Ghz P4 but it got cancelled do to thermal throttling at stock clocks/juice.

[Manicdan]

The funniest thing I read is "it takes 6 of your cores to match 4 of Intel's". I guess I could have said "it takes 8 of your threads to match 6 of AMD's" but let's face it, those arguments aren't going anywhere.

that agrument is one of the best proofs that its the same damn chip, but people look at it as less cores or more threads.

i look at 2 things, the chips die size, and the price. the price matters to users more, the die size to relative profits (though size isnt what decides anywhere near the total cost)

cant wait for CMT so both sides can have 2 ways to look at the same chip, resulting in 4 methods to argue instead of just too.