Discussion in 'Article Discussion' started by arcticstoat, 31 May 2011.
Hmm a different story was presented here:Xbitlabs
If that is true it doesnt look good for AMD
They are all scaremongers I tell ya
AMD have obvioulsy had some quality issues from Global Foundries, which is why they went from a 'per waifer' price to a 'per working core' price earlier this year.... all the specualtion surrounding these delays are just the standard 'fill in the gaps' guestimates to improve webhits on their site - especially that tool from semi accurate.
The Real Reason:
Bulldozer Delayed until September - Delays are expected due to a new amitious marketing plan - Originally it was thought AMD were going with 'Neon Pink' packaging, but have since re-evaluated the packaging and, in a stunning turn of events have now opted for for an un-conventional 'blue/amber/white lightning' combo - this has thrown the packaging supplier into turmoil.
A 'source close to the presses' is quoted as saying
"There is a global shortage on 'Electric Blue' ink, as this is the chosen box colour we simply can't get the boxes out in time, the lads are devastated as it will be several weeks before the presses will run again......"
More to follow :0)
I was waiting for these hoping to build a new rig based on them. Would have been the first AMD rig I have had in about 5 years. Guess ill wait for LGA2011 now :-(.
Some pretty gooney comments on here^^
It's a long pipeline design, it runs faster for the same level of output. Integrate that into your thinking.
And why on earth would BD be lined up against LGA2011?? They're aimed at entirely different segments. It will be a p67 alternative plain and simple.
Who said they were up against LGA2011? Its going to look embarrassing when theyre releasing Bulldozer that cant even compete with SB (according to that source) and Intel have IB around the corner.
more time to save up money and I think for the most part when it comes to Intel and AMD, that once they get to B3 then its an ideal cpu
A long pipeline does not affect CPU speed in a positive way.
Think of it this way. We have a factory in which we have four workers, who each put a wheel on a car. Each worker takes (lets say) 5 seconds.
In a different factory, we have a single worker who does the same job as four other workers, again in 5 seconds (but this time for all wheels).
The same numbers of cars are coming out every second, because whilst in factory 2 there's one car being worked on at any one time, there are four cars being worked on at any one time in factory 1. So it all works out - a car is coming out every five seconds in factory one, and a car is coming out every 5 seconds in factory 2.
This is where the clock speed comes into effect - if we decrease the time it takes for each tyre to be fitted, it decreases the amount of time each tyre takes to come out.
So at full load, there's no difference between a long pipeline and a short pipeline.
At part load, however, there's a much bigger difference - if we only have one car being worked on at a time, factory 2 takes 1/4 of the time as factory 1 - because it can take a car from start to finish in 5 seconds, whilst factory 1 will take 20 seconds.
CPUs make mistakes, remember - and every time they make one the entire pipeline must be emptied and it must start filling it again.
This is one of the reasons why Sandy Bridge is faster than the old Pentium 4s. Sandy Bridge has an (IIRC) 12 stage pipeline, the last P4s had a 21 stage or 22 stage (I can't remember) pipeline.
In general a shorter pipeline is an advantage because it stays full for more of the time (mistakes made take less time to correct). You'd only want a longer pipeline if the way your pipeline worked meant you had to use the longer pipeline.
So no, a long pipeline does not help the CPU overall speed, and might be a disadvantage.
Though delays definitely suck I'm actually kinda glad to know this now, if it's even true then gives me more time to save for the upgrade since I'll wait for a revision that overclocks instead of jumping on the 1st thing that comes out...
Thank you for telling granny how to suck chicken ovulations.
"It's a long pipeline design, it runs faster (the clock runs faster) for the same level of output (across a general basket of loads)." This is essentially true and didn't warrant a cuddle-me-elmo response. Rephrasing though for the sake of things:
"it's a long pipeline design therefore it's clockspeed will tend to be higher than that of a short pipeline design. The risk of pipeline stalls and branch miss predictions is higher in a long pipeline design and the number of cycles required to flush the pipeline in such circumstances to allow a restart is greater also. The long pipeline design allow for higher switching speeds potentially mitigating the lower effective IPC across a general usage model. Usage models with less branching code should see a lower number of miss-predicts and pipeline stalls and therefore experience higher #throughput#.
Mentioning Pentium 4 and the perils of a long pipeline design in the same sentence should be avoided. There was a lot wrong with P4 but you will be very hard pressed to effectively pin it to the long pipeline principle. You can take the IBM Power processors as evidence that it can be used effectively.
Actually, Bakes' explanation was clearer than your first one, which I found slightly ambiguous.
But do continue your point with the P4's and IBM PowerPC's, Action_Parsnip. I'm genuinely interested.
It seems AMD says Bulldozer will still be released this summer:
+1 to that.
Remember the steaming turd that was the Pintium 4e with it's long pipeline?
This sucks... Not sure what to do anymore in terms of a system upgrade.
Wait till Sept... or build a system with a 6 core & AM3+ compatible board?
Separate names with a comma.