Official Improvements of Piledriver Cores.

[Silverforce11]

How can you seriously call Ivy a failure?

Let's see.. it can't beat SB for raw performance due to its terrible OC scaling and huge thermal problems. 90C with water cooling and a mild OC is nuts. it's also priced higher.

They can't beat their previous gen on a bigger node with huge spending for 22nm and trigate.. if that aint a failure, you have to be a pretty optimistic person.

Why would you pick IvB over SB if its IPC gains are 3-5% (really outside of benchmarks, who is going to notice 3-5%??? its a joke) yet it falls way short compared to OC potential and needing a much beefier cooling setup??

 

[pelov]

If you want to compare architectures objectively you need to run the CPUs featuring them at the same clock speed. We're not comparing clock speeds, we're comparing architectures.

That's what you do to compare IPC in specific workloads. That's not CPU architecture...

CPU architectures aren't determine core-by-core at specific workloads but rather as a whole. If you mean specifically "core design with respect to IPC" then yea, you'd have a case here but it would ignore ISAs, clock speeds (which are also part of its architecture...) and the addition of those other cores, cache sizes, on-die GPUs, etc. When you say "architecture" you're referring to the entirety of the processor and not just its IPC in X workload. If that were the case then the Power6 and Power7 would have horrible "architectures" because they relied on clock speeds and threading. Of course, this doesn't make any sense as those chips were designed for specific purposes and offered fantastic performance (as well as hopped above Moore's curve).

IPC is quite clearly important on the desktop, though. We've learned that after an epic fail from each x86 manufacturer after their clock speed goals ended in utter failure so I can understand the vehement dismissal -- admittedly I too cringe every time I see clock speeds become a major and overwhelming goal of an architecture. But that's just it... there's a massive difference between the two. CPU architecture assumes both clock speed and IPC, along with the rest of the CPU. Comparing CPU architectures amounts to an in-depth hardware review with benchmarks.

 

[frozentundra123456]

Let's see.. it can't beat SB for raw performance due to its terrible OC scaling and huge thermal problems. 90C with water cooling and a mild OC is nuts. it's also priced higher.

They can't beat their previous gen on a bigger node with huge spending for 22nm and trigate.. if that aint a failure, you have to be a pretty optimistic person.

Why would you pick IvB over SB if its IPC gains are 3-5% (really outside of benchmarks, who is going to notice 3-5%??? its a joke) yet it falls way short compared to OC potential and needing a much beefier cooling setup??

You are only looking at it from the very limited view of overclocking. Ivy would only be a "failure" if it were a step back. It may not be the kind of advancement that SB was, but it is hardly a "failure" for the 99% of the people that dont overclock. And I dont think you could possibly argue that it is not an improvement in the mobile space due to the much improved graphics and lower power usage.

 

[Lepton87]

You are only looking at it from the very limited view of overclocking. Ivy would only be a "failure" if it were a step back. It may not be the kind of advancement that SB was, but it is hardly a "failure" for the 99% of the people that dont overclock. And I dont think you could possibly argue that it is not an improvement in the mobile space due to the much improved graphics and lower power usage.

IB is a mild failure only in the eyes of overclockers, when we disregard OC it's a solid chip. I'm still terribly disappointed because I bought 4 threaded SB because I wanted to buy 8 threaded IB with better IPC and HIGHER OC headroom later on, but with the way things are it does not make a lot of sense.

 

[Hatisherrif]

+1 for pelov / LOL_Wut_Axel discredited for being a fanboy

Though I am not the type of guy to be the inquisitor and point "fanboy" at everyone whose opinion I somehow deem invalid, I must say that your level of ignorance has amazed me. There are people like you on every forum on the internet, but god damn it, stop, chill and cease spreading the hate. Stop your hatetul remarks against AMD engineers and stop your hateful remarks against fellow forum members who disagree with you.

Apart from that, I must say one or two things more. I am not as "pro" in many of these things like you guys, I don't really know much about specific elements of architectural design that can make a CPU perform better or worse, but I know when to give someone credibility and someone posting fishy evidence from a clearly biased point of view cannot be regarded as a tech pro because his behavior is unprofessional. I have owned many AMD and Intel CPUs and for most of the time, I liked them both. What decided me as an AMD supporter in the end is the fact that it never proved itself to be a blackmailing, moneymaking giant who will stop at nothing to disable the competition in any way possible.

Also, some of you like to point out AMD's lies and I think we can all agree on that: AMD's marketing team is a bunch of noobs, and I wouldn't insult them if they weren't so deceitful in their info and utterly ineffective, bashing the competitor in an uncivilized manner. But that's it for AMD.

Intel, on the other hand, may not lie to you, but will always employ the good old Machiavelli: the ends justify the means. I think this says everything.

In the end, I will always be true about my position - morally, I'm leaning towards AMD in regard to past experiences. CPU performance-wise I will ALWAYS be objective. Nobody deserves to get a worse product because of my ideas!

And please stop the C2Q argument. To sum it up:

Phenom II is architecturally superior
Core2 came out years earlier
Core2 has slightly higher IPC (45nm)
Phenom II 955 was significantly cheaper than the competitive Quad and even performed better than Q9650 in several key areas
gaming performance was better with the Phenom
Intel's quads were still better optimized for programs 99% of people use, like Microsoft programs (Office etc).

 

[Abwx]

You mix up INT cores and issue ports. Core 2 for example got 3 SSE issue ports per core, but can only use 2 at a time. And be careful looking at macroscaled diagrams. The one you link is actually terrible.

Quite right about the pic , still , C2 has less parralelized exe ressources,
as you also point it.

 

[LOL_Wut_Axel]

Hi,

You forget to note that clock speed is limited a whole lot by microarchitectural decisions. So what you really want to look at is performance per thread per watt.

No because, again, we are comparing CPU architecture and nothing else. The point is to analyze raw speed, and you won't be able to do that if you introduce other variables like different clock speeds because you're not putting each architecture on equal ground. If you want compare CPU speed, instead of architecture performamce, then it's fine to compare at different clock speeds.

Sent from my VM670 using Tapatalk 2

 

[Olikan]

No because, again, we are comparing CPU architecture and nothing else. The point is to analyze raw speed, and you won't be able to do that if you introduce other variables like different clock speeds because you're not putting each architecture on equal ground. If you want compare CPU speed, instead of architecture performamce, then it's fine to compare at different clock speeds.

but clock speed is a part of cpu arquitecture ¬¬

 

[nismotigerwvu]

Clearly the FSB was't very limited if it allowed the higher-end Core 2 Quad Q9*50 to make short work of the Phenom II X4 in multi-threaded, and that's very well documented.

The R&D argument is an old and tired excuse. AMD's CPU engineers suck, plain and simple, because the GPU engineers have zero problems making excellent products, even though they have a lot less money than NVIDIA.

And yes, the team that made the netbook version of Atom was stupid. The only design goal was to make the processor as cheap to manufacture as possible and to get the lowest power consumption, actual efficiency and performance be damned. The Atom team has switched their focus to making mobile SoCs, and their first attempt has been a very good one. Atom is moving forward when it comes to the mobile (smartphone, tablet) market and not the netbook market.

Oh of course, all these guys are idiots, I mean it must be a small miracle when they wake up in the morning and don't trip over the coffee table. Also, since you clearly are so much smarter than these guys and that budgets don't matter, where's the benchmarks of the super fast CPU you made then? You do realize that proper trolling requires at least a little subtlety right? Taking the wreckage approach just gets you ignored.

 

[pelov]

but clock speed is a part of cpu arquitecture ¬¬

He's misusing the term "architecture." To him it represents IPC in general (which itself doesn't make too much sense) rather than "core design" or actual CPU architecture. I think he's confusing "core architecture" with "CPU architecture" (which is used on marketing slides as "architecture" referring to core architecture), but even then both have IPC included in their terms and their specificity would depend on the uncore, and to confuse it even further that too has a say in core architecture. He's not ignoring clock speeds but also cache, IMC, on-die GPUs, perf-per-watt, die size, transistor density, and essentially everything else that goes into CPU architecture as these all have a say in performance.

To clarify a bit, let's take AMD's Piledriver:

Piledriver is the core architecture but is featured in both Trinity and Vishera. They're essentially the same design except that Trinity is an APU and Vishera is a desktop CPU, lacking the on-die GPU, while Trinity lacks the 8MB L3 cache that Vishera has. Both are still sharing the same core architecture (or more accurately modular core design) therefore AMD refers to both as Piledriver on their slides.

IPC, though, would differ on the workload despite the same core architecture as the cache plays a major role (remember that IPC varies depending on the workload). Furthermore, Vishera is supposedly getting a couple of other goodies on top of the Trinity Piledriver module so that too may have an impact on IPC despite the fact that they still share the same core architecture.

He's got valid points that I agree with, for example Bulldozer being an architectural screwup because of the chase for clock speeds (and the other being the CMT approach which he hasn't), but he's woefully confusing his terms and assuming that it's far less complicated than it really is.

And to twist things even more, AMD initially designed Bulldozer because they knew they'd never catch Intel in IPC due to fab advantage. So even fabs have a massive impact on IPC and CPU architectures. Talk about complicated 😛

 

[LOL_Wut_Axel]

No, just as comparing clocks is not comparing CPU architectures, comparing IPC *alone* is also not comparing CPU architectures, not in the sense in which you're using it - to point out that engineers suck.

If IBM came out tomorrow with a CPU that beats IB by 2-3x in every known benchmark while working at 30GHz, would you also point out how far behind Intel they are and how much their CPU architecture sucks and hence their engineers must suck too (it's worse than P3!!)? Completely oblivious of real world performance? Sorry, but that would be laughable. CPU engineers develop CPU architectures to deliver performance in real workloads, not to maximize IPC...

Yeah... yes, it is. We're comparing the raw speed of the architecture. The only way you'll get that is by comparing the speed of each at the same clock speed. Want to compare a Phenom II X4 955 at 4GHz and say that it's faster than a Core 2 Quad Q6600 at 3.2 GHz and say the Phenom ia faster? That's fine, but it's false to say the underlying architecture is faster. Just because the architecture speed, which is measured in IPC, is higher, doesn't necesarrily mean that the CPU it's powering will automatically be faster. However, having higher IPC contributes a lot to performance in all workloads and nowadays both Intel and AMD achieve near parity in terms of clock speeds.

Sent from my VM670 using Tapatalk 2

 

[Silverforce11]

You are only looking at it from the very limited view of overclocking. Ivy would only be a "failure" if it were a step back. It may not be the kind of advancement that SB was, but it is hardly a "failure" for the 99% of the people that dont overclock.

If that's the case, the K version is an utter failure.. while the non-K version, meh (almost identical performance, slightly less power use, more expensive!). Fair enough?

 

[Hatisherrif]

Yes, I've owned Phenom II X4 955. It takes the Q6600 for a walk in the park and then rapes it. Sorry for that...

Architecturally, Phenom has L3 cache, IMC and loads of other advantages that can be witnessed in many applications, even against Q9x5x. In gaming, slaughter.

 

[Makaveli]

Why would you pick IvB over SB if its IPC gains are 3-5% (really outside of benchmarks, who is going to notice 3-5%??? its a joke)

Wouldn't this be the same as the very small oc headroom SB has over IB like 4.5 vs 5.0?

You won't notice this outside of benchmarks?

The only valid argument is heat but if IB can do 4.5 at stock vcore even that is not an issue.

The only people complaining are those that can't get to that magical 5.0 number. Everyone else will just buy IB because it is better than SB minus abit of headroom overclocking not a huge deal my friend!

And the fools saying IB is the next Bulldozer are even dumber than they sound.

 

[LOL_Wut_Axel]

Oh of course, all these guys are idiots, I mean it must be a small miracle when they wake up in the morning and don't trip over the coffee table. Also, since you clearly are so much smarter than these guys and that budgets don't matter, where's the benchmarks of the super fast CPU you made then? You do realize that proper trolling requires at least a little subtlety right? Taking the wreckage approach just gets you ignored.

Yet again missing the point. The line about AMD's GPU engineers being excellent must have gone over your head "accidentally", right?

Again, R&D argument: excuse. AMD didn't have any problems making great CPUs from 2002 to 2006, and AMD hasn't had problems making excellent GPUs for 4 years now.

Sent from my VM670 using Tapatalk 2

 

[SlowSpyder]

If you want to compare architectures objectively you need to run the CPUs featuring them at the same clock speed. We're not comparing clock speeds, we're comparing architectures.

Whether an 8150 can run at 4.5GHz or not doesn't change the fact that the architecture, clock for clock, is slower than something that came out 6 years ago from Intel. It's simply a way to point out how bad AMD's CPU engineers are and how far behind they are in comparison to Intel's.

Want to compare CPU to CPU? Okay, fine: compare with varying clock speeds that represents what you'll get out of each. If you want to compare architectures, it needs to be clock-for-clock because we're measuring Instructions Per Cycle and not Instructions Per Second. It'd be completely stupid to say one architecture is faster to the other when you're not putting all of them on equal ground so you can compare their raw speed.

Phenom DID scale better with highly threaded apps compared to C2Q's. But, the better scaling was not enough to neccessarily make it faster. But you could see that the AMD architecture did scale better. The FSB was generally fast enough, but HT gained AMD a few percent.

And I get comparing architecture to architecture, I have no horse in this race, I don't see what comparing these CPU's matters. But I think we have to look at more than just the architecture of the cores, because the cores have to be fed to perform.

Lastly, performance is derived from the IPC and clock speed. If Bulldozer ran at 8GHz none of us would care about the IPC, we would just know it's way faster than Sandy Bridge. Unfortunately for AMD, I think they counted too much on clock speed, then when they missed their targets things looked that much worse. Piledriver needs some IPC improvements in my opinion, as it'll not likely hit the clockspeds needed to be competitive. But, if they can increase the clockspeeds 10+% and the IPC 10+%, it may end up being a decent enough part. I just hope they do not do it at the expense of power consumption.

 

[LOL_Wut_Axel]

Yes, I've owned Phenom II X4 955. It takes the Q6600 for a walk in the park and then rapes it. Sorry for that...

Architecturally, Phenom has L3 cache, IMC and loads of other advantages that can be witnessed in many applications, even against Q9x5x. In gaming, slaughter.

LOL, whatever you say.

The evidence keeps invalidating your claims, but keep raving on about how your Phenom II "rapes" Core 2 Quads when it comes to gaming--it doesn't.

Speaking of cache, maybe you forgot Kentsfield had 8MB of L2 cache and Yorkfield (Q9*50) had 12MB of it.

 

[LOL_Wut_Axel]

Phenom DID scale better with highly threaded apps compared to C2Q's. But, the better scaling was not enough to neccessarily make it faster. But you could see that the AMD architecture did scale better. The FSB was generally fast enough, but HT gained AMD a few percent.

And I get comparing architecture to architecture, I have no horse in this race, I don't see what comparing these CPU's matters. But I think we have to look at more than just the architecture of the cores, because the cores have to be fed to perform.

Lastly, performance is derived from the IPC and clock speed. If Bulldozer ran at 8GHz none of us would care about the IPC, we would just know it's way faster than Sandy Bridge. Unfortunately for AMD, I think they counted too much on clock speed, then when they missed their targets things looked that much worse. Piledriver needs some IPC improvements in my opinion, as it'll not likely hit the clockspeds needed to be competitive. But, if they can increase the clockspeeds 10+% and the IPC 10+%, it may end up being a decent enough part. I just hope they do not do it at the expense of power consumption.

CPU performance is derived from that, yes. I'm comparing one aspect: IPC, or architecture raw speed. Nothing else. I made that very clear from the beginning.

Unfortunately, AMD can't defy nature, so they should've been smart enough from the beginning to know that making CPUs focused only on high clock speeds wasn't gonna get them anywhere. They didn't learn from Intel and the Pentium 4, and now it's come back to haunt them. They also forgot Intel has their own foundries which are much more advanced than GF's. It seems like their engineers completely forgot about the fact that power consumption increases exponentially as you raise clock speed, thinking they'd be able to achieve stock clock speeds of 5GHz.

Instead, they can only compete with Intel when it comes to clock speeds, meaning they weren't even able to achieve their high clocks goal--the average overclock of an unlocked Sandy/Ivy Bridge and Bulldozer is 4.5GHz, and the base clock speeds are nearly identical with an FX-8150 being 3.6GHz and a 3770K 3.5GHz. That's why their engineers are idiots in comparison to Intel's: they ignored fundamental mistakes Intel made way back at the end of 2000, while Intel used this harsh lesson to improve by unprecedented amounts. At the same time Intel was moving forwards, AMD moved backwards.

 

[Rvenger]

+1 for pelov / LOL_Wut_Axel discredited for being a fanboy


And please stop the C2Q argument. To sum it up:

Phenom II is architecturally superior
Core2 came out years earlier

First of all, lets not get this thread locked with jabs like this.


If I am correct, all of the CPUs before BD are based on the original A64 architecture. So saying the Phenom II is architecturally superior doesn't make much sense.

A64 - 2003
Conroe - 2006

 

[Abwx]

It seems like their engineers completely forgot about the fact that power consumption increases exponentially as you raise clock speed, thinking they'd be able to achieve stock clock speeds of 5GHz.

Stop trolling and go buy yourself a brain rather than displaying
your complete lack of understanding in EE.

Any second rate graduate know that power comsumption increase
linearly with frequency while it increase as a square law with
increased supply voltage.

Increasing both parameters would yield a cubic law but certainly
not an exponential law that is only relevant for leakages.....

 

[Hatisherrif]

CPU performance is derived from that, yes. I'm comparing one aspect: IPC, or architecture raw speed. Nothing else. I made that very clear from the beginning.

Unfortunately, AMD can't defy nature, so they should've been smart enough from the beginning to know that making CPUs focused only on high clock speeds wasn't gonna get them anywhere. They didn't learn from Intel and the Pentium 4, and now it's come back to haunt them. They also forgot Intel has their own foundries which are much more advanced than GF's. It seems like their engineers completely forgot about the fact that power consumption increases exponentially as you raise clock speed, thinking they'd be able to achieve stock clock speeds of 5GHz.

Instead, they can only compete with Intel when it comes to clock speeds, meaning they weren't even able to achieve their high clocks goal--the average overclock of an unlocked Sandy/Ivy Bridge and Bulldozer is 4.5GHz, and the base clock speeds are nearly identical with an FX-8150 being 3.6GHz and a 3770K 3.5GHz. That's why their engineers are idiots in comparison to Intel's: they ignored fundamental mistakes Intel made way back at the end of 2000, while Intel used this harsh lesson to improve by unprecedented amounts. At the same time Intel was moving forwards, AMD moved backwards.

So where is it? Where is the CPU you designed and/or fabricated? I am dying to see it, since nobody at AMD knows what they're doing.

 

[LOL_Wut_Axel]

Stop trolling and go buy yourself a brain rather than displaying
your complete lack of understanding in EE.

Any second rate graduate know that power comsumption increase
linearly with frequency while it increase as a square law with
increased supply voltage.

Increasing both parameters would yield a cubic law but certainly
not an exponential law that is only relevant for leakages.....

Except you forgot the simple fact that you need higher voltage with higher clock speed, so yes, power consumption will rise exponentially, because higher voltage raises power consumption exponentially. That's why Intel was never able to reach 7-10GHz like they wanted to on the Pentium 4 and instead abandoned the idea altogether.

Even if we had a theoretical CPU that could run at the same voltage at both 3GHz and 6GHz, since power consumption would rise linearly that means the 6GHz CPU would have 2x higher absolute power consumption. You need to increase performance within a given thermal and power target, which is why your argument makes no sense. What you're essentially arguing is that AMD shouldn't have any problems reaching 7GHz stock on Bulldozer since apparently voltage and power targets don't exist or don't matter.

Your line about me not understanding EE is hilarious, because that's exactly what you've just displayed in your comment.

 

[dma0991]

So where is it? Where is the CPU you designed and/or fabricated? I am dying to see it, since nobody at AMD knows what they're doing.

I agree. Even though AMD might not be doing so well, I can be damn sure that I can't do what AMD engineers are paid to do. It is not my field of expertise, therefore I limit my questioning on this matter to a certain degree that I am capable of understanding.

 

[Olikan]

Piledriver is the core architecture but is featured in both Trinity and Vishera. They're essentially the same design except that Trinity is an APU and Vishera is a desktop CPU, lacking the on-die GPU, while Trinity lacks the 8MB L3 cache that Vishera has. Both are still sharing the same core architecture (or more accurately modular core design) therefore AMD refers to both as Piledriver on their slides.

actually no, there is more changes...

the biggest one, is that vishera will be able to do 4 64bit Movs per clock, while trinity will do 2 64bit Movs per clock

Movs are very important in x86 code, the IPC gain could be huge if there is no bottleneck (but it will)

AMD is going this route....
Bulldozer -> Piledriver v1 + GPU -> Piledriver v2 + No GPU -> Steamroller v1 + GPU -> Steamroller v2 + No GPU -> etc.

 

[pelov]

If you're going to quote me to tell me I'm wrong then you may want to read the following paragraph, otherwise you might end up making a bit of a fool of yourself.

Piledriver is the core architecture but is featured in both Trinity and Vishera. They're essentially the same design except that Trinity is an APU and Vishera is a desktop CPU, lacking the on-die GPU, while Trinity lacks the 8MB L3 cache that Vishera has. Both are still sharing the same core architecture (or more accurately modular core design) therefore AMD refers to both as Piledriver on their slides.

IPC, though, would differ on the workload despite the same core architecture as the cache plays a major role (remember that IPC varies depending on the workload). Furthermore, Vishera is supposedly getting a couple of other goodies on top of the Trinity Piledriver module so that too may have an impact on IPC despite the fact that they still share the same core architecture.