AMD Bulldozer Finally!

[kitch9]

I know fine well Piledriver will be an improvement, but it clocking much higher? You pulled that out of no where.
Who the hell suggested Piledriver wouldn't be an improvement?

Piledriver is touted 20% faster than Zambezi, which is probably a mixture of clocks and IPC.
And IPC is still damn important, it inherently increases performance across the board.

But no way in hell can anyone at this point in time come out with a blanket statement of "IT'LL CLOCK MUCH HIGHER".

I'm not saying it won't clock better, it probably will clock higher, and Piledriver CPU's will come out higher clocked as standard, but they could top out around the same, or marginally higher than Zambezi is now.

Zambezi to Piledriver will be like Agena to Deneb, higher clocked as standard, much better IPC, higher clocking. That's exactly the way I see it. But don't forget that Ivy will also launch the same as that, although to a lower scale.


EDIT : Also, undeniable proof that AMD allowed lies http://www.xtremesystems.org/forums...nd-Bulldozer&p=4531177&viewfull=1#post4531177

JF-AMD is no small fish, he'd have damn well known how it would have performed.

Bulldozer is designed to facilitate higher clocks than it currently has. It's only manufacturing that's lowering current clocks.

This will be fixed, and then the designed higher clocks can be implemented.

 

[mmj_uk]

Bulldozer is designed to facilitate higher clocks than it currently has. It's only manufacturing that's lowering current clocks.

This will be fixed, and then the designed higher clocks can be implemented.

The problem is in its current form it's going to need to reach 6ghz+ just to match Sandy Bridge on a per core basis, will that be feasible in the coming years and what will Intel's IPC be like by that time?

If AMD are going to focus purely on multithreaded performance then don't forget that Intel can release a £200-300 6-core at the drop of a hat if they need to for an instant 50% performance boost and still dominate on IPC.

 

[Martini1991]

Bulldozer is designed to facilitate higher clocks than it currently has. It's only manufacturing that's lowering current clocks.

This will be fixed, and then the designed higher clocks can be implemented.

And I'm sure Intel could ramp out higher clocked parts if they so wished.
Judge it on what it is right now, because that's what's known.

Piledriver, for all we know, could just be a higher clocked version of what we've got, which frankly isn't good enough.

Right now, Intel are outclocking and outperforming AMD clock for clock. That's how it is. AMD need to massively outclock Intel if they want to go the "High clock speed" way. Given that as it is, the FX4 is struggling to outperform a 955, yet having a clock advantage, that's an absolutely failure when it replaces it price wise.

 

[kitch9]

And I'm sure Intel could ramp out higher clocked parts if they so wished.
Judge it on what it is right now, because that's what's known.

Piledriver, for all we know, could just be a higher clocked version of what we've got, which frankly isn't good enough.

Right now, Intel are outclocking and outperforming AMD clock for clock. That's how it is. AMD need to massively outclock Intel if they want to go the "High clock speed" way. Given that as it is, the FX4 is struggling to outperform a 955, yet having a clock advantage, that's an absolutely failure when it replaces it price wise.

Intel don't seem to be easily adding cores, I'm not convinced we'll see an octo core for mainstream money from them anytime soon.

If AMD can get power consumption down by circa 15-20%, clocks up 10% and IPC up 10-15% whether through software or hardware tweaks they are back in the game. I know you don't want to hear that as you appear to enjoy spanking Intels monkey, but the first two in that list will be solved as manufacturing process naturally improves, I would suggest IPC would naturally improve as manufacturing improves too as they can be more adventurous with transistor placement.

The design is very scalable, and they have a design where they can easily add more cores as time progresses, once the kinks are worked out it should be ok.

 

[AgedIdiot]

once the kinks are worked out it should be ok.

But not until - and given 'history' , I ,for one, will not take it as a fact that all will be sorted until it is built,delivered and independently tested to have been so.

 

[kitch9]

But not until - and given 'history' , I ,for one, will not take it as a fact that all will be sorted until it is built,delivered and independently tested to have been so.

Well history does dictate that both manufacturers do reasonably well at sorting their problems.

The bulk of Bulldozers problems are manufacturing, and that usually improves as a process matures, we have to hope it does as AMD need competitive chips at the £200 mark for competitions sake.

 

[gareth170]

it's all if's and but's, we all know and can say what needs to be done, at the end of the day we have no idea what amd will do unless u work for them

 

[mmj_uk]

Intel don't seem to be easily adding cores, I'm not convinced we'll see an octo core for mainstream money from them anytime soon.

That's not because Intel are having difficulties, it's because AMD can't even beat their quads so they are keeping their (untouchable) hex prices high.

 

[kitch9]

it's all if's and but's, we all know and can say what needs to be done, at the end of the day we have no idea what amd will do unless u work for them

They've got two choices:

1: Get the tech to work as originally intended, and carve its place out in the market.

2: Go out of business.

I have a good idea what AMD are trying to do, and we all want to hope they get it to work.

 

[buchtis]

Will these improve with Windows 8 then? Or rather, will Windows 8 make these chips looks better in the public eye, performance wise?

 

[SeeKified]

Will these improve with Windows 8 then? Or rather, will Windows 8 make these chips looks better in the public eye, performance wise?

Doubtful. Whatever optimisations Microsoft can make with Windows 8 won't be game-changing.

Even if Piledriver turns out to manage higher speeds and perform slightly better clock for clocks, that in turn might mean that power consumption will get even higher. Meanwhile, Ivy Bridge has a rated TDP of 77W and will still likely perform better than an architecture that's already thrashing AMD's counterparts.

Hate to say it, but it ain't looking good for them.

 

[CAT-THE-FIFTH]

Doubtful. Whatever optimisations Microsoft can make with Windows 8 won't be game-changing.

Even if Piledriver turns out to manage higher speeds and perform slightly better clock for clocks, that in turn might mean that power consumption will get even higher. Meanwhile, Ivy Bridge has a rated TDP of 77W and will still likely perform better than an architecture that's already thrashing AMD's counterparts.

Hate to say it, but it ain't looking good for them.

Upto 20% improvement in some scientific applications when lightly threaded in this article:

http://techreport.com/articles.x/21865

Upto 12% with WoW:

http://www.tomshardware.co.uk/fx-8150-zambezi-bulldozer-990fx,review-32295-23.html

So it is better news for whatever CPUs AMD has at the end of next year but not so good ATM. Not a huge increase but the power consumption improvements look a bit more promising.

Also why would Piledriver have increased power consumption over Bulldozer??

Even the C2 to C3 stepping change of the Phenom II X4 lead to a drop in power consumption with the same CPU. The C2 stepping Phenom II X4 965 was 140W originally and the C3 stepping version was 125W.

Even the Phenom II X6 has around the same power consumption as a Phenom II X4 and it has two extra cores and 20% extra transistors.

Piledriver is a tweaked version of Bulldozer primarily to improve power consumption and most likely the performance of the FlexFP unit which is underwhelming. This is the CPU core which will find its way into the laptop and desktop APUs AMD are introducing next year. I would be surprised if the high end Piledriver CPUs were bigger than the Bulldozer CPUs they replaced with the same core count.

 

[mmj_uk]

Even the Phenom II X6 has around the same power consumption as a Phenom II X4 and it has two extra cores and 20% extra transistors.

But they achieved that mainly by lowering the general clockspeed a few hundred mhz as well as the introduction of turbo core which underclocked the 3 idle cores to 800mhz whilst boosting the 3 others.

If you compared an X4 @3.6ghz to an X6 @3.6ghz with turbo core disabled the X6 TDP would obviously be a lot higher, BD already has this turbo mechanism so there's no quick and easy way to reduce TDP this time.

 

[CAT-THE-FIFTH]

But they achieved that mainly by lowering the general clockspeed a few hundred mhz as well as the introduction of turbo core which underclocked the 3 idle cores to 800mhz whilst boosting the 3 others.

If you compared an X4 @3.6ghz to an X6 @3.6ghz with turbo core disabled the X6 TDP would obviously be a lot higher, BD already has this turbo mechanism so there's no quick and easy way to reduce TDP this time.

But a 3.2GHZ Phenom II X4 955BE at the same clockspeeds has the same power consumption as a 3.2GHZ Phenom II X6 1090T. You do realise the the highest clocked Phenom II X4 at the time was the 3.4GHZ Phenom II X4 965. Adding 20% more transistors for a 5.9% decrease in base clockspeed at the same TDP indicates the process had improved massively by then.

You are making the assumption that:
1.)A high end 8 core Piledriver based CPU will be bigger than a 8 core Bulldozer based one. I assume you think they are going to plonk more L3 cache but then again this might not be the case.

One of the main reasons for the poorer performance of Bulldozer is the FlexFP unit. Improving that alone would lead to an IPC improvement in games as these are FP heavy operations AFAIK.

2.)Processes don't improve over time. Both AMD and Intel release faster CPUs on the same process meaning performance per watt does increase.

Even the C2 to C3 stepping change for the Phenom II X4 965 alone dropped TDP by 15W.

So why would Piledriver have greater power consumption than Bulldozer?? By the time it is introduced it will be at least 9 months to year after AMD first started to use the GF 32NM process.

 

[Martini1991]

Intel don't seem to be easily adding cores, I'm not convinced we'll see an octo core for mainstream money from them anytime soon.

If AMD can get power consumption down by circa 15-20%, clocks up 10% and IPC up 10-15% whether through software or hardware tweaks they are back in the game. I know you don't want to hear that as you appear to enjoy spanking Intels monkey, but the first two in that list will be solved as manufacturing process naturally improves, I would suggest IPC would naturally improve as manufacturing improves too as they can be more adventurous with transistor placement.

The design is very scalable, and they have a design where they can easily add more cores as time progresses, once the kinks are worked out it should be ok.

If AMD can get clocks up 10%, power consumption down and IPC up, then AMD's chip will be a good chip.
I'm no fanboy. AMD's Zambezi is pathetic though. Should I act like "Oh, nice try AMD"?

No.. We need competition. Now.

Also, where the hell did I mention about Intel and cores? Although, Intel were the first to release a hexcore in the enthusiast arena, the 980x, and that destroys AMD's hexcores and their 8 core CPU's. Intel don't have any reason to take hexcores mainstream yet.

You seem to be the one with the bias here, be that against Intel, or for AMD, I couldn't care less.




Edit : Will people please start saying Zambezi? As Piledriver is part of the bulldozer stuff.

I can't see Piledriver having a larger power consumption than Zambezi.

 

[Gashman]

sure they have problems, but the design of the processor is fundamentally sound for the most part. also what the hell is with all the 'doesn't look good for them...' talk? has anyone noticed that they thrive in some areas of the market, and apparently Bulldozer is looking good in the server segment as well as it was expected to, seriously a bump in the road and everyone is basically assuming they will disappear overnight.

half of the problem is nothing to do with them either, half of it is entirely down to Global Foundries and their somewhat sketchy 32NM process. look at it objectively for once please, Llano is doing well, Zacate is doing well, Trinity will do well, Bulldozer is doing fair job in server market, how is this a bad time for AMD then? we are seeing around the internet as we speak, optimisations being discovered, tinkering with Bulldozer is allowing people to slowly uncover hidden performance, so still think everyone is jumping the gun entirely.

also the whole Bulldozer architecture is lost in translation on here, it isn't and has never been intended to combat six core Intel parts. the primary opponent for Bulldozer was the four core 2500K and the four core (Hyper-Threading) 2600K. the whole purpose from the start was to have two 'part' cores combating a single core, where it would loose out slightly in single-threaded performance but overturn in multi-threaded performance. the whole point was to create an architecture that had proper execution resources for a second thread within a singular core structure, so each module is more like a double threaded core (like physical Hyper-threading) than a true (full resource) core. know people complain endlessly about the semantics of it all but that is just where the architecture lies, its not a core vs. core affair, it is module vs. core, sorry to anyone who can't see that that was AMD intention from day one.

but at the end of the day I still agree with all the other posts, that the processor does need to improve to be considered a success, but to be a success in AMD eyes all it needs to do is properly compete with the 2500K and 2600K, we all just need to take a step back and wait to see what AMD do next with this situation, already heard on a couple of sites that they are pushing forward the release of Trinity.


Edit: also Piledriver is the successor to Bulldozer, it isn't Bulldozer it is Piledriver, the core codename won't be Piledriver, just like Bulldozers is 'Zambezi', the codename for Piledriver is 'Vishera' I believe. so don't see the problem in Piledriver/Bulldozer, or else you should start calling Piledriver Vishera instead...

 

[Martini1991]

Why's it called BD2 etc?
I thought Vishera was the platform, just like Zambezi is the Scorpius platform. Hence why they're called "piledriver cores"?
Codenames are retarded, as you can see.
If we can't compare core for core, what should we compare as? Software doesn't care what it is, it sees a module as two cores/threads, not 1.

 

[kitch9]

Why's it called BD2 etc?
I thought Vishera was the platform, just like Zambezi is the Scorpius platform. Hence why they're called "piledriver cores"?
Codenames are retarded, as you can see.
If we can't compare core for core, what should we compare as? Software doesn't care what it is, it sees a module as two cores/threads, not 1.

You compare price v performance, and if you are really smart you try to compare price v performance in your own circumstances.

At the minute BD provides poor price v performance in most circumstances, they need to find an extra 20-25% and all of a sudden they are good price v performance, and I would probably buy one.

Only an idiot would use IPC as the only way of measuring overall performance on a multi-core (Module.) chip.

 

[sunama]

At the minute BD provides poor price v performance in most circumstances, they need to find an extra 20-25% and all of a sudden they are good price v performance, and I would probably buy one.

Indeed. BUT, improvements are unlikely to be seen until next year. This means that the only way to improve price/performance is to reduce price. This won't happen until there is enough supply in the retail channels.

I just checked OCUK and they dont have any 8150 or 8120s in stock. This means that at current price levels, they can clear all their stock. While this is the case, a retailer (and AMD) would be insane to reduce prices.

Only an idiot would use IPC as the only way of measuring overall performance on a multi-core (Module.) chip.

I agree. With the advent of multiple cores/threads, we have to move away from IPC, from a single core.

 

[ATIorNvidia]

Bulldozer is designed to facilitate higher clocks than it currently has. It's only manufacturing that's lowering current clocks.

This will be fixed, and then the designed higher clocks can be implemented.

Exactly.

sure they have problems, but the design of the processor is fundamentally sound for the most part. also what the hell is with all the 'doesn't look good for them...' talk? has anyone noticed that they thrive in some areas of the market, and apparently Bulldozer is looking good in the server segment as well as it was expected to, seriously a bump in the road and everyone is basically assuming they will disappear overnight.

half of the problem is nothing to do with them either, half of it is entirely down to Global Foundries and their somewhat sketchy 32NM process. look at it objectively for once please, Llano is doing well, Zacate is doing well, Trinity will do well, Bulldozer is doing fair job in server market, how is this a bad time for AMD then? we are seeing around the internet as we speak, optimisations being discovered, tinkering with Bulldozer is allowing people to slowly uncover hidden performance, so still think everyone is jumping the gun entirely.

also the whole Bulldozer architecture is lost in translation on here, it isn't and has never been intended to combat six core Intel parts. the primary opponent for Bulldozer was the four core 2500K and the four core (Hyper-Threading) 2600K. the whole purpose from the start was to have two 'part' cores combating a single core, where it would loose out slightly in single-threaded performance but overturn in multi-threaded performance. the whole point was to create an architecture that had proper execution resources for a second thread within a singular core structure, so each module is more like a double threaded core (like physical Hyper-threading) than a true (full resource) core. know people complain endlessly about the semantics of it all but that is just where the architecture lies, its not a core vs. core affair, it is module vs. core, sorry to anyone who can't see that that was AMD intention from day one.

but at the end of the day I still agree with all the other posts, that the processor does need to improve to be considered a success, but to be a success in AMD eyes all it needs to do is properly compete with the 2500K and 2600K, we all just need to take a step back and wait to see what AMD do next with this situation, already heard on a couple of sites that they are pushing forward the release of Trinity.


Edit: also Piledriver is the successor to Bulldozer, it isn't Bulldozer it is Piledriver, the core codename won't be Piledriver, just like Bulldozers is 'Zambezi', the codename for Piledriver is 'Vishera' I believe. so don't see the problem in Piledriver/Bulldozer, or else you should start calling Piledriver Vishera instead...

Fully agree with this, too. It's a shame.... there's a lot of daft people in this thread that will not be able to grasp this, as they look at everything from a consumer/enthusiast's viewpoint.