Piledriver not coming until 2013 now?

[Ferzerp]

Another words, if Piledriver can get close to Sandybridge in single threaded performance and still manage the same level of multithreaded performance that Bulldozer has, I will buy one for sure

If I understand you correctly, what it seems you are saying is that you don't care for one company or the other. What matters most to you is performance?

I think you will find that that is the view of the overwhelming majority of us.

 

[blckgrffn]

I can honestly say there is no way PD will gain 30-40% IPC. You can quote me on that.

Done!

I agree, sounds like 5-7% IPC with reductions in power usage and clock speed increases is how PD is getting its 10-15% increase in performance.

Really, that in itself is going to be pretty admirable, IMHO. A true Tock.

If I understand you correctly, what it seems you are saying is that you don't care for one company or the other. What matters most to you is performance?

I think you will find that that is the view of the overwhelming majority of us.

I think that gets diluted around here when we starting talking performance/watt, performance/$, etc.

If it was cut and dry what would we argue about? And if we didn't argue ("discuss") it would be a pretty boring forum

 

[AtenRa]

Well, what im saying is that the majority of the gamers are playing in 1680x1050 or 1920x1080/1200. In those resolutions with filters enabled the majority of DX-11 games (and a lot DX-9/10) are GPU bound and BD will perform close or the same as SB.

I know and i have said it before that there are games that really are faster with Intel Core architecture but those games are beginning to be the minority as more and more games are coded in DX-11.

About PD, yes if they will keep the higher scaling with 10-15% higher single thread performance then PD could be really fast in multithreaded apps. But they will really need to lower power usage a lot too.

 

[Rvenger]

If I understand you correctly, what it seems you are saying is that you don't care for one company or the other. What matters most to you is performance?

I think you will find that that is the view of the overwhelming majority of us.

I agree with you to an extent Ferzerp.

Performance matters to me but I am willing to take a 10% hit to support AMD if they can manage to at least match Intel's previous generation. I have been an AMD supporter ever since I started building systems and I would hate to see them exit the desktop CPU market completely. If the performance is a 10% hit I can usually make that up with overclocking anyways.

As a former hardware reviewer, I have to take an unbiased approach no matter who I like more. Its like being a referee but sometimes they make me wonder :sneaky:

 

[exar333]

I agree with you to an extent Ferzerp.

Performance matters to me but I am willing to take a 10% hit to support AMD if they can manage to at least match Intel's previous generation. I have been an AMD supporter ever since I started building systems and I would hate to see them exit the desktop CPU market completely. If the performance is a 10% hit I can usually make that up with overclocking anyways.

As a former hardware reviewer, I have to take an unbiased approach no matter who I like more. Its like being a referee but sometimes they make me wonder :sneaky:

Within 10% of what? IPC, perf/watt, total processing power? Thats a very vague statement. Intels 'last gen' was Nehalem, and I doubt PD will match that. In 2 months, the 'last gen' will be SB, are you referring to this? As this will be the 'last gen' when PD is (presumably) released.

I support AMD too, and want them to be competitive, but also want to be realistic here as well.

 

[Chiropteran]

I can honestly say there is no way PD will gain 30-40% IPC. You can quote me on that.

IPC is not the whole story. With sufficiently better clock frequency it becomes less relevant.

 

[exar333]

IPC is not the whole story. With sufficiently better clock frequency it becomes less relevant.

Thats exactly why I asked the question.

Haven't heard if PD will OC any better than BD.

 

[Idontcare]

I was hopeful that AMD and Glofo would see a repeat of 45nm in that they initially launched some nice x4 PhII SKUs at reasonable clocks and reasonable power consumption, held some OC'ing demo's to showcase capability, and then worked and tweaked the 45nm process to the point that they not only raised clockspeeds while keeping power at the same level but they managed to increase the core count 50% in the same step. I was hopeful this would be the case for 32nm as well.

Now, however, it looks to me like AMD is all but abandoning GloFo's 32nm as quickly as possible. They cancelled the 10core/20core opterons, and they are doing what they can to build confidence in shareholders by way of saying their future is 28nm and it is coming as quickly as possible.

If true, I think it takes exceptional courage on behalf of the executive team to make these kinds of decisions and attempt to navigate a ship the size of AMD through a minefield of potentially bad ideas that could doom the company without much lead time.

There's a lot of moxy and spirit in them yet.

 

[Rvenger]

Within 10% of what? IPC, perf/watt, total processing power? Thats a very vague statement. Intels 'last gen' was Nehalem, and I doubt PD will match that. In 2 months, the 'last gen' will be SB, are you referring to this? As this will be the 'last gen' when PD is (presumably) released.

I support AMD too, and want them to be competitive, but also want to be realistic here as well.

10% in single threaded performance. Once Piledriver releases SB will be last gen Intel. Sorry for the lack of clarification.

 

[janas19]

I don't think AMD can compete with Intel on a head-on IPC basis anymore. It seems like AMD will use it's APUs to give more price and performance options to a consumer market. It's still Intel vs AMD, but not on a clock to clock basis.

 

[Idontcare]

I don't think AMD can compete with Intel on a head-on IPC basis anymore. It seems like AMD will use it's APUs to give more price and performance options to a consumer market. It's still Intel vs AMD, but not on a clock to clock basis.

Sadly I all too well remember when Cyrix started their initial stumble and fall decline. It was very much kinda like this.

It started with a less than stellar microarchitecture coupled with a less than stellar IBM foundry process tech. And they never recovered.

 

[Rvenger]

Sadly I all too well remember when Cyrix started their initial stumble and fall decline. It was very much kinda like this.

It started with a less than stellar microarchitecture coupled with a less than stellar IBM foundry process tech. And they never recovered.

Then VIA bought Cyrix in 1999, introduced the VIA C3 processor, it was pretty cool that you didn't have to use a heatsink with those CPUs but they were pretty slow, especially when it came to gaming, AMD K6-2 would smoke it. Then after the first generation of the VIA CPUs, they disappeared. VIA seems to be a pretty silent company as of late.

 

[janas19]

Sadly I all too well remember when Cyrix started their initial stumble and fall decline. It was very much kinda like this.

It started with a less than stellar microarchitecture coupled with a less than stellar IBM foundry process tech. And they never recovered.

Here's to hoping AMD rebounds!

:thumbup:

 

[Torquemada]

They also indicated that L1 cache was back up to 64 bytes, which is a good thing, it seems like any software that has been optimized for AMD or recent Intel is going to be tuned for that cache size. I am hoping that bit stays accurate.

(You mean 64 kB not 64 B. Or you meant cache line length?) Not sure it'd be so good. They're going for the "cloud" server market (funny how many ex-hippies are in Marketing nowadays) which would indicate focus on 2P if not 4P, and inclusive all the way is easier for cache coherency, and 32 kB is easier for inclusive. 64+64 kB exclusive L1 was sweet when total cache was in the kB range but I suspect the latency of L1 has become more important, nowadays when L3 can be bicus dicus anyway.

That said, I'm pretty much clueless about the software tuning you mention. I do understand plenty of it has to do with avoiding L1 trashing. But I thought SB has 32 kB L1I and 32 kB L1D?

It'll be interesting to see if PD introduces any differences in the cache system (or the apparently blamed memory controller). I'm no engineer but just finished reading through Agner Fog's excellent architectural analysis of recent architectures. He seems to point out the front-end decoder as a big potential bottleneck in BD. (Which SB-doing-HT sidesteps by the uop cache and other sauce.) So I want to see what AMD has done with PD when it arrives.

[Regardless of the above, cheers all. fr1st ps0t]

 

[Torquemada]

If you're going to be reductive, you could at least go all they way and say Sandy Bridge is just a "tweaked" Pentium Pro, because it is just as accurate as what you are saying right now.

Can I?

Sandy Bridge is just a tweaked Pentium Pro.

That said, for that range of values for "tweak", I wouldn't want for example my own nose tweaked. It would be interesting to see though. I'd expect physical dislocation beyond the same room.

 

[AtenRa]

If im not mistaken, PD will only double the ITLB from 32 entries to 64 entries
Both L1 Icache (64kb) and L1 Dcache (16kb) will remain the same as in BD.

Edit: ITLB not P

 

[CHADBOGA]

Well I see the int.. i mean anantech community has managed to get polyzp's thread locked, I guess this one deserves the same fate. Seriously OP, why are you trolling this forum? Why are you posting false information and what's with the sad face in the title, when you are clearly posting misinformation? Why do you post here and why don't you just leave nobody wants to read your flamebait trolling, especially when you know full well it's not true. This thread serves no purpose and should be locked.

No need to be flippin out.

 

[janas19]

No need to be flippin out.

I am disturbed by your avatar

 

[CHADBOGA]

I am disturbed by your avatar

http://i1196.photobucket.com/albums/aa420/JimRaynor1975/WorldsSexiestMan.jpg

 

[Torquemada]

If im not mistaken, PD will only double the ITLP from 32 entries to 64 entries
Both L1 Icache (64kb) and L1 Dcache (16kb) will remain the same as in BD.

ITLB?

 

[Maximilian]

http://i1196.photobucket.com/albums/aa420/JimRaynor1975/WorldsSexiestMan.jpg

What meme is this?

 

[Torquemada]

Thank you, Maximilian.

 

[AtenRa]

ITLB?

ehmm yes, TLB

 

[BD231]

lol, Chadboga rules.

 

[Rvenger]

http://i1196.photobucket.com/albums/aa420/JimRaynor1975/WorldsSexiestMan.jpg

http://i749.photobucket.com/albums/xx132/rpisano01/Will_Ferrell_Old_School_Streaking.jpg


:hmm:

Let's stick to the topic, guys. If you must post pictures, please don't inline them
-ViRGE