Bulldozer vs Netburst

[protokiller]

So if intel were to make a modern version of netburst using trigate (they never will) do you think it would be better than bulldozer?

More power efficient?

I guess assume 6-8 cores for the netburst product.

This is meant to be a fun discussion.

We can only speculate on the clock speed and pipeline length of the netburst product.

The battle of lame ducks.

 

[AbdullahG]

Which ever smells like ice cream and shoots rainbows and sprinkles out of it's heatsink...

 

[warmon6]

So if intel were to make a modern version of netburst using trigate (they never will) do you think it would be better than bulldozer?

More power efficient?

I guess assume 6-8 cores for the netburst product.

This is meant to be a fun discussion.

We can only speculate on the clock speed and pipeline length of the netburst product.

The battle of lame ducks.

hmm...... well..... from know what netburst was, i would say it would be an even bigger power hog than bulldozer if it was released as a 6 to 8 core cpu. That's with die shrinks and everything else.

as for performance, i just dont see modernized netburst beating bulldozer..... Like said above, with the possibility of even more power use than bulldozer, the clocks would have to be lowered to fit in a certain heat range. Also, you would have to change the netburst arch so much to make it better on IPC that it wouldn't be called netburst anymore (minor tweaks wouldn't be enough).....


All and all, bulldozer IMO would be better than Netburst

 

[gnomio]

Netburst

Bulldozer

 

[Pyree]

Why? Netburst is only going to be 1 core and go crazy on clock. You can't compare it with multi core cpu that is better suited for multi thread application.

 

[fazers_on_stun]

Hmm, this thread is like asking who is more beautiful - an 800 lb prom queen or a 3-legged rhino ..

Interestingly enough, a number of netburst ideas have resurfaced in modern CPUs, albeit in much improved form. If you read the realworldtech articles on both BD and SB, you'll see how much influence netburst actually had on both..

 

[gnomio]

Hmm, this thread is like asking who is more beautiful - an 800 lb prom queen or a 3-legged rhino ..

Interestingly enough, a number of netburst ideas have resurfaced in modern CPUs, albeit in much improved form. If you read the realworldtech articles on both BD and SB, you'll see how much influence netburst actually had on both..

HyperThreeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeads

 

[protokiller]

Why? Netburst is only going to be 1 core and go crazy on clock. You can't compare it with multi core cpu that is better suited for multi thread application.

Don't forget the Pentium D, it was multi core.

This is a therotical competition between a modern netburst (6-8 cores, much more cache, tweaks, built on Intel's trigate) vs bulldozer.

 

[FALC0N]

Don't forget the Pentium D, it was multi core.

This is a therotical competition between a modern netburst (6-8 cores, much more cache, tweaks, built on Intel's trigate) vs bulldozer.

The Pentium D was a hackjob multicore. Netburst was never meant to be multicore.

IF intel just took netburst, added multiple cores through the FSB, ala the Pentium D, and die shrunk it, it would still be slower than bulldozer. Bulldozer IPC is much higher and the FSB issue would be a bigger problem for netburst as you added cores.

 

[Pyree]

^+1

Netbust is not an archetecture with multi core in mind. That is why Pentium D is just as power hungry as other Netburst cpu and perform poorly compared with the true multi core cpu from AMD of that time.

Comparing Bulldozer to Netburst is unfair and yield no interpretable result.

Bulldozer is just so much newer and not a competitor of Netburst product.

It would be more constructive if it is comparing Bulldozer with Sandy Bridge.

 

[jsc]

The benchmarks do that, and mostly, there's no comparison,

Had BD been on time, things might have been different.

 

[jimmysmitty]

So if intel were to make a modern version of netburst using trigate (they never will) do you think it would be better than bulldozer?

More power efficient?

I guess assume 6-8 cores for the netburst product.

This is meant to be a fun discussion.

We can only speculate on the clock speed and pipeline length of the netburst product.

The battle of lame ducks.

The biggest issue with netburst was Intels 130nm and 90nm process. They were not that great. 65nm Netburst CPUs (Ceader Mill) had low TDP CPUs of 65w.

I think it would be a tie though. netburst also is not a good arch for this gen of software. It would probably have great thermals and low power usage on 22nm but I doubt ti would perform better than BD.

 

[loneninja]

Netburst has terrible IPC even compared to bulldozer, only way it could compete is with much higher clocks.

 

[protokiller]

^+1

Netbust is not an archetecture with multi core in mind. That is why Pentium D is just as power hungry as other Netburst cpu and perform poorly compared with the true multi core cpu from AMD of that time.

Comparing Bulldozer to Netburst is unfair and yield no interpretable result.

Bulldozer is just so much newer and not a competitor of Netburst product.

It would be more constructive if it is comparing Bulldozer with Sandy Bridge.

I really don't think communicating through the fsb was that bad, or at least was the thing that made it perform poorly.

The Core 2 Quad also communicated through the fsb and had two dies on one chip, a hackjob just like Pentium D and it did fine.

 

[zhihao50]

is someone willing enough to perform this with a pentium 4 or any old CPU so we can start comparing with exact figures. Its in this link. Getting pentium to 3.3Ghz might be a small challenge.
http://www.tomshardware.com/reviews/fx-8150-zambezi-bulldozer-990fx,3043-6.html

 

[holdingholder]

http://www.tomshardware.com/reviews/processor-architecture-benchmark,2974.html
Tom's already did a comparison of single core performance recently and included a netburst p4. As shown it did very poorly showing that netburst was a horrible design, despite being able to achieve high clock speeds, clock for clock, it can't compete with more modern cpus. I remember when the core 2's first arrived a 2.2 ghz core 2 would give you about the same performance as a 3.6 ghz pentium d.

 

[protokiller]

http://www.tomshardware.com/reviews/processor-architecture-benchmark,2974.html
Tom's already did a comparison of single core performance recently and included a netburst p4. As shown it did very poorly showing that netburst was a horrible design, despite being able to achieve high clock speeds, clock for clock, it can't compete with more modern cpus. I remember when the core 2's first arrived a 2.2 ghz core 2 would give you about the same performance as a 3.6 ghz pentium d.

That's an interesting article, something I found interesting was that the P4 beat most of amd's offerings in 3dmark 11 graphics score at 3GHz.

 

[zhihao50]

ok so on the first lame test single core bulldozer takes 1.73X i7 2600k's time. That put it at 2.48s on the lame test holdingholder linked. P4 holds at 4.22, which is 70% worse again and 55% slower on itune.

so the question is how much would the three die shrink, tri gate transistor, new instrusction sets (it would be logical if intel had continualed to develop netbust instead of switching to core) help?

 

[gnomio]

Why would Intel go backwards. Sure SB has some Netburst ideas but only the positives of it and they fly with it. Using the whole Netburst architecture will be taking a 100 steps back into the stone ages

 

[fazers_on_stun]

I really don't think communicating through the fsb was that bad, or at least was the thing that made it perform poorly.

The Core 2 Quad also communicated through the fsb and had two dies on one chip, a hackjob just like Pentium D and it did fine.

For average desktop use, FSB was fine - even 2-socket server it was OK. Netburst's main deficiency was poor out-of-order/branch prediction for such a long pipeline - when an incorrect branch occurred, it took a while to flush the long pipe so the CPU basically stalled out for quite a few clock cycles. Also I think it had a 2-issue decoder, which is a problem BD seems to share when both cores are active and sharing the front end.

 

[gnomio]

Yeah if your have a big enough cache. But once that cpu went on looking for data in the ram your system slowed down to a crawl