AMD K10 Architecture - technical explanation by InsideHW

Status
Not open for further replies.

Wildside

Hellfire!
Messages
3,038
Location
Riverside, CA
InsideHW - AMD K10 Architecture

a huge, huge, explanation about the upgrade from the old K8 architecture to K10. I read some of it and i would like to say i am impressed on how AMD is doing, even though Intel is like 2-3 times bigger i dont care. AMD is going for things different then Intel n different is good in my opinion. For having good performance for cheap prices is strong with AMD, it's just that i wish AMD can do more then what they have so far. The biggest change i want to see in AMD overall is more advertising; letting the customers know that AMD are there, AMD is cheaper if u have a budget limit. The only advertising ive seen is the internet, that's it. Magazines............ehh just a logo in pre-built machines with a price tag is all.

i dont know the whole history of AMD, but i rather have them stay in the game rather than pay even higher prices from Intel any day, aka monoploy like with Dell.

any comments about AMD u guys like to make? What about AMD has done so far? That link up there, there is an arrow so u can skip pages all the way to the end if u want, i just wanted to post the huge article here so everyone can read about AMD n understand them a little be better with there upgrade to Phenom.
 
Amd has done plenty for the market in previous years and even now.

And right now, theyre holding on to the htpc market with the 780G chipset.

They've also introduced some low power 45w dual cores...aimed at the htpc users im sure.
 
Amd has done plenty for the market in previous years and even now.

And right now, theyre holding on to the htpc market with the 780G chipset.

They've also introduced some low power 45w dual cores...aimed at the htpc users im sure.

i like the idea of Hybrid Crossfire, even Hybrid SLI, it makes it so low-budget gamers can enjoy video games in better quailty.
 
i like the idea of Hybrid Crossfire, even Hybrid SLI, it makes it so low-budget gamers can enjoy video games in better quailty.

I think hybrid graphics would have been nice if they were available 6 months ago where the only gpu's for around $100 were the HD2600XT 512mb and 8600gt 512mb. Now that the 8800gs 384mb and HD3850 256mb are available at that price there is less incentive to go with a hybrid system.

I have found K10 to be disappointing compared to K8's success.They need to move on to a new architecture soon so they can get back to being competition for Intel.
 
After having researched a lot about K10 beforehand, I can say I am impressed with how AMD was able to pull off what they did, even if it is a bit behind C2Q.

I think Intel believed that before Phenom was released, that AMD couldn't pull it off
K10 is a completely revamped architecture, much more complex than that of K8. And, it's a native quad core, and it's being manufactured on AMD's 65nm SOI process

I think the reason Intel hasn't got a monolithic design so far, is because it is a real challenge to do so.

I have heard that Nehalem is having some difficulty, specifically in getting clock speeds up, and being able to manufacture them with good yields.

Like K10, it is also a much more complex design; though AMD's 65nm quad's are only about 8% larger than Nehalem at 45nm, so that might indicate it being more complex than Phenom.

Don't quote me on this, but I suspect that Nehalem might not be all that great, at least on first release.
Intel has been a bit quiet on how Nehalem should perform; other than details such as having an IMC etc..
It may be that Intel just wants more of their Core 2's to sell before the release, but it could also mean that performance isn't as good as they'd hoped it would be.

Also, I have heard a rumor - and this is just a rumor so far, so don't shoot me - but Intel might be planning on locking all but their high-end Nehalem chips from overclocking
Intel stops new Nehalem from overclocking - PCSTATS Forum
 
After having researched a lot about K10 beforehand, I can say I am impressed with how AMD was able to pull off what they did, even if it is a bit behind C2Q.

I think Intel believed that before Phenom was released, that AMD couldn't pull it off
K10 is a completely revamped architecture, much more complex than that of K8. And, it's a native quad core, and it's being manufactured on AMD's 65nm SOI process

I think the reason Intel hasn't got a monolithic design so far, is because it is a real challenge to do so.

I have heard that Nehalem is having some difficulty, specifically in getting clock speeds up, and being able to manufacture them with good yields.

Like K10, it is also a much more complex design; though AMD's 65nm quad's are only about 8% larger than Nehalem at 45nm, so that might indicate it being more complex than Phenom.

Don't quote me on this, but I suspect that Nehalem might not be all that great, at least on first release.
Intel has been a bit quiet on how Nehalem should perform; other than details such as having an IMC etc..
It may be that Intel just wants more of their Core 2's to sell before the release, but it could also mean that performance isn't as good as they'd hoped it would be.

Also, I have heard a rumor - and this is just a rumor so far, so don't shoot me - but Intel might be planning on locking all but their high-end Nehalem chips from overclocking
Intel stops new Nehalem from overclocking - PCSTATS Forum

that's crazy! Locking up clock speed? That hurts everybody who OCs.

as for K10, Apok, think of it this way: it's just two Athlon 64 x2's duct taped together, that's it. Sure it increases performance, but it just isnt enough anymore. I really think that the 128-bit memory controller is too small even though the numbers of the data transfer rate is so high in that article. It should be like 256-bit, then we might see impressive results man. I mean seriously, at 256-bit bandwidth that would be impressive performance. I just cant wait for the 45nm to come out, even if it is a die shrink with more transistors, nothing else.
 
as for K10, Apok, think of it this way: it's just two Athlon 64 x2's duct taped together, that's it.
It's a lot more than that.
They've upgraded almost every aspect of the core itself, the cache, and the memory controller.
It is based on K8, but it's got a lot of architectural advantages that K8 doesn't have.
That's why K10 can perform very close to a Core 2 clock for clock (and sometimes exceed it in games, especially if clocked over 3GHZ)

I really think that the 128-bit memory controller is too small even though the numbers of the data transfer rate is so high in that article. It should be like 256-bit, then we might see impressive results man. I mean seriously, at 256-bit bandwidth that would be impressive performance. I just cant wait for the 45nm to come out, even if it is a die shrink with more transistors, nothing else.
I don't think that will really give them enough advantage to justify the increase in transistors, therefore power consumption, size and consequently cost to manufacture.

I mean, having fast access to memory is one thing. But they also have to be able to decode all that data fast enough to be executed by the ALU/FPU. And then the ALU/FPU would need to be fast enough to be able to process the data that's being sent to it.

I think one of the weaknesses is actually to do with the prefetchers and branch predictors not being as accurate as they should be, so the ALU's are empty more than they should be.
 
well if the memory controller is fine, then the next thing i would want an upgrade is the amount of cache that is put into the Phenom, it's just too low.
 
Also, I have heard a rumor - and this is just a rumor so far, so don't shoot me - but Intel might be planning on locking all but their high-end Nehalem chips from overclocking
Intel stops new Nehalem from overclocking - PCSTATS Forum

Wouldn't be surprised, the *******s....

Maybe my next build will be an AMD if they come back against these new intels with OC abilities that can surpass that of a locked nehalem chip.
 
Status
Not open for further replies.
Back
Top Bottom