Bulldozer press kit pictures

[Tikerz]

Can't wait

 

[LOL_Wut_Axel]

First, CISC and RISC are meaningless terms nowadays.

Second, the reason most people use (and have been using) x86 is because of compatibility and performance/price. Back in the '90s PowerPC, MIPS, and Alpha were all faster than any x86 processor of the day. The only saving grace for x86 back then was how much cheaper it was.

Yes, but beginning this century they started to fall in performance/watt by a huge amount. That's the main reason why Apple abandoned PowerPC and went for Core 2 and all following Intel microarchitectures.

RISC by design is better suited for hugely parallel workloads and therefore servers, virtualization, and render farms. It's also very good for very low-power/low-performance applications, hence why it's used for anything from calculators to tablets. It doesn't really offer an in-between, which is what desktops and laptops need. They don't execute for the most part hugely parallel workloads, but they do require good multi-threaded performance for some applications coupled with very high single-threaded performance and performance/watt. Currently the RISC architecture can't deliver that, and like you said before, cost meant a higher adoption rate for x86 in the '90s.

 

[Edrick]

Second, the reason most people use (and have been using) x86 is because of compatibility and performance/price. Back in the '90s PowerPC, MIPS, and Alpha were all faster than any x86 processor of the day. The only saving grace for x86 back then was how much cheaper it was.

IBM Power 1 gave us extensions like FMA back in the 90s which will not make it into x86 CPUs until BD and Haswell.

Alpha gave us IMC in the 90s which did not make it into intel CPUs until Nehalam.

They were very innovative, but very expensive.

 

[jhu]

RISC by design is better suited for hugely parallel workloads and therefore servers, virtualization, and render farms. It's also very good for very low-power/low-performance applications, hence why it's used for anything from calculators to tablets. It doesn't really offer an in-between, which is what desktops and laptops need.

I wouldn't go that far. For example, there's nothing in the SPARC v9 spec that says a sparc T4 is the most appropriate implementation. Similarly, there's nothing stopping Intel (other than lack of demand) from making an x86 version of a SPARC T4.

 

[StrangerGuy]

I'm also hoping that OBR is once again being a super troll. All these "it's going to fail" posts and then 'BAM!' on release date he says 'I fooled you all, I have this thing OCed to 5.5GHz on air and it is amazing.'

Not likely though, *sigh*.

If BD is so amazing, why would AMD say nothing about actual performance or even power consumption in official slides? You think a $100+ billion Intel wouldn't know that way before release?

 

[LOL_Wut_Axel]

I wouldn't go that far. For example, there's nothing in the SPARC v9 spec that says a sparc T4 is the most appropriate implementation. Similarly, there's nothing stopping Intel (other than lack of demand) from making an x86 version of a SPARC T4.

Well, CISC is pretty different from RISC fundamentally, so it's not that easy. They'd probably need a lot of R&D, including the costs of making a whole new architecture that 'behaves' similarly to begin with.

I'm not sure if there's a direct comparison, but I'm sure Sandy Bridge has much higher single-threaded performance than SPARC T4.

The fact that Oracle was able to get so many gains in comparison to their previous architecture in both integer and floating point performance is impressive. I also find it interesting that it uses CMT like Bulldozer to get smaller die sizes in comparison to CMP and be able to put more processing cores in it. For hugely parallel workloads, RISC makes more sense generally. It's also better for very low-power devices, but again, fails to strike the balance that CISC (x86) has for desktops and laptops.

 

[jhu]

Well, CISC is pretty different from RISC fundamentally, so it's not that easy. They'd probably need a lot of R&D, including the costs of making a whole new architecture that 'behaves' similarly to begin with.

Actually, they're not. Peal away the x86 decoding layer and current x86 processors are "RISC" processors.

The fact that Oracle was able to get so many gains in comparison to their previous architecture in both integer and floating point performance is impressive.

The fact that Sun/Oracle's versions of SPARC did not have out-of-order execution until the T4 probably has something to do with that.

For hugely parallel workloads, RISC makes more sense generally. It's also better for very low-power devices, but again, fails to strike the balance that CISC (x86) has for desktops and laptops.

All Intel or AMD would have to do is remove the x86 decode and we would get an instant mainstream "RISC" processor. But, again, there's nothing inherent to "RISC" that makes them highly parallel or low power. Nothing in the POWER, SPARC, MIPS, or ARM specifications dictates highly multi-core/multi-threaded designs or extremely low power designs.

 

[bronxzv]

Back in the '90s PowerPC, MIPS, and Alpha were all faster than any x86 processor of the day. The only saving grace for x86 back then was how much cheaper it was.

huh?!, what about the PPro (running 32-bit code)?, I remember some neat Intergraph workstation with dual PPro *faster* than more expensive MIPS, SPARC and Alpha based systems

 

[LOL_Wut_Axel]

Actually, they're not. Peal away the x86 decoding layer and current x86 processors are "RISC" processors.



The fact that Sun/Oracle's versions of SPARC did not have out-of-order execution until the T4 probably has something to do with that.



All Intel or AMD would have to do is remove the x86 decode and we would get an instant mainstream "RISC" processor. But, again, there's nothing inherent to "RISC" that makes them highly parallel or low power. Nothing in the POWER, SPARC, MIPS, or ARM specifications dictates highly multi-core/multi-threaded designs or extremely low power designs.

Seems to me that you're overly simplifying things. If Intel could design an X86 CPU that has comparable power consumption to ARM, they would've already made it. As of now I have not seen a single example of your statements. The inverse is true for ARM as well, and that truth is that it doesn't scale as well as X86 for high performance. Again, if what you're arguing was feasible it would've existed since years ago, or it would be in the process of making.

Your last sentences are simply wrong.

 

[RocksteadyDotNet]

IBM Power 1 gave us extensions like FMA back in the 90s which will not make it into x86 CPUs until BD and Haswell.

Alpha gave us IMC in the 90s which did not make it into intel CPUs until Nehalam.

They were very innovative, but very expensive.

<Censored>.

Intel were using an IMC back in 386 days.

 

[AnandThenMan]

Dumbass.

Was that really necessary?

 

[Vesku]

Guys, stay on topic and talk about that belt buckle. That just for reviewers or are we all going to get a little bit of Austin on our PIB?

 

[Dadofamunky]

Man, that is a SWEET press kit! Almost makes me miss my tech author freelancing days...

Jhu is right. It's amazing what was actually important to us back in the day. CISC vs. RISC is literally last century's war...

 

[RocksteadyDotNet]

Was that really necessary?

I retract the comment.

 

[Dadofamunky]

Was that really necessary?

+1

I retract the comment.

Taken. That must have posted two nanoseconds before my +1.

 

[sequoia464]

http://www.tigerdirect.com/applicati...?EdpNo=1375893

Saw it at overclock.net

Sabertooth 990 FX bundled with an FX-6100

not shipping for 7-21 days

 

[zlejedi]

I saw a guy buying a $270 motherboard in anticipation of the bulldozer release. I hope it's good for his sake... I can't imagine putting down that much money on an unreleased processor.

Why would you hope it for his sake?
Stupidity needs to be punished harshly only then there's a hope for humanity future.

 

[nenforcer]

What is that big ass commerative coin in the press kit, an attempt to bribe the journalists and reviewers?

 

[jhu]

Seems to me that you're overly simplifying things. If Intel could design an X86 CPU that has comparable power consumption to ARM, they would've already made it. As of now I have not seen a single example of your statements. The inverse is true for ARM as well, and that truth is that it doesn't scale as well as X86 for high performance. Again, if what you're arguing was feasible it would've existed since years ago, or it would be in the process of making.

Your last sentences are simply wrong.

Just because Intel or some other x86 manufacturer has not made a processor within the same power envelope as ARM does not mean it cannot be done. Also just because ARM hasn't reached the computing performance of x86 also does not mean that it cannot be done. Remember, absence of evidence is not the same as evidence of absence.

Consider that 20 years ago, if you were to tell people that MIPS would be relegated to routers instead of high performance computing, you'd be the laughing stock of everyone around. And yet here we are: MIPS relegated to routers and x86 in the majority of supercomputers.

 

[jhu]

huh?!, what about the PPro (running 32-bit code)?, I remember some neat Intergraph workstation with dual PPro *faster* than more expensive MIPS, SPARC and Alpha based systems

x86 has definitely had the performance/price crown. However, on a CPU vs. CPU comparison back in the day, x86 was pretty slow. Even the Pentium Pro:

spec cpu95
Intergraph 200 MHz PPro, 1 cpu
DEC Alpha, 1 cpu
Dell XPS, Pentium 133, 1 cpu
HP 9000, PA-RISC, 1 cpu
IBM, POWER2, 1 cpu

The above are published results from 12/1995. Looking at those FP results, Pentium is at the very bottom. PPro can hold its own, but is still slower than many of the others. Contrast to now, where you'd be hard pressed to find any spec benchmarks published in 2011 on processors that are not either Xeon or Opteron (four POWER results among the hundreds of Xeon/Opteron results)!

 

[3DVagabond]

I saw a guy buying a $270 motherboard in anticipation of the bulldozer release. I hope it's good for his sake... I can't imagine putting down that much money on an unreleased processor.

You think it would suck to buy a high end board and have the cpu suck? How about designing and producing/marketing a high end board and then have the cpu suck? Imagine how pissed all of these mobo manufacturers are going to be if BD sucks.

 

[Ancalagon44]

I doubt it will suck enough for it to not sell at all. For it to be THAT bad, it would need to underperform the price as well as run too hot. I think it will be priced right, which is most of what matters to consume. Plus it overclocks well which matters for e peen.

 

[StrangerGuy]

You think it would suck to buy a high end board and have the cpu suck? How about designing and producing/marketing a high end board and then have the cpu suck? Imagine how pissed all of these mobo manufacturers are going to be if BD sucks.

That didn't stopped Asus from releasing ROG boards for AM3, so why would they not do it for BD? Put flashy heatsinks, a different PCB color etc that costs peanuts to manufacture, charge a good $200 extra and there will still be people buying them. I'm would be stupid not to do that if I'm Asus.

 

[bronxzv]

CPU comparison back in the day, x86 was pretty slow.

a bit slow for specfp indeed, though from your link a PPro 200 MHz was leading in specint http://www.spec.org/cpu95/results/res9512/p034.html
(a tie with Alpha 21164 333 MHz) http://www.spec.org/cpu95/results/res9512/p023.html

also MIPS and SPARC submissions are missing, they were clearly behind Alpha IIRC and in the same league than PPro for fp (and behind for int)

 

[Voo]

Just because Intel or some other x86 manufacturer has not made a processor within the same power envelope as ARM does not mean it cannot be done. Also just because ARM hasn't reached the computing performance of x86 also does not mean that it cannot be done. Remember, absence of evidence is not the same as evidence of absence.

Yeah. As soon as we get any ARM CPU anywhere in the same performance range as a modern x86 CPU we'll see how their energy efficiency turns out (and then there's the fact how its measured, the usual numbers ARM throws around always exclude caches which completely skews the results)

The extra power for decoding x86 instructions is usually estimated to about 5&#37; (and that was before the sb cache so that should lower by now) and while 5% is nice, it's hardly that noteworthy - Intels manufacturing advantage could certainly balance that out. And ARM certainly has its own problems wrt decoding (different instruction modes with different lengths, predicates, instructions setting flags,..). Though they try to simplify that (thumb got rid of lots of stuff). And getting a high performance microprocessor isn't a simple undertaking, even design teams with decades of experience under their belt got that wrong often enough (pentium4, power6, original itanium,..). And ARM doesn't have anything similar to the teams of Intel or IBM. We'll see how that turns out, but I think it's doubtful we'll see ARM CPUs with comparable performance in the near future..


And claiming that RISC or CISC architectures are especially great at optimizing multithreaded performance or anything else can be easily proven wrong. Just look at all the surviving RISC architectures and their overall goals - you can't get a much wider range than the differences between IBM (single threaded performance using all resources available without bothering too much about cost), Oracle (throughput at the cost of single threaded performance, "simple" design with good power efficiency for low socket servers) Fujitsu (basically the old mainframe mentality? not really cared about those) and ARM.