Finally, some good news from AMD

[Regs]

http://www.dailytech.com/article.aspx?newsid=3797

"With the news of AMD's DDR2 Opteron launch, AMD managed to squeeze in one tidbit of information definitely newsworthy: quad-core Opterons have been taped out. AMD's Executive Vice President Henri Richard had previously dubbed these native quad-core design as the K8L architecture.

The company's press release claims "AMD plans to deliver to customers in mid-2007 native Quad-Core AMD Opteron processors that incorporate four processor cores on a single die of silicon." For a little historical perspective, AMD's dual-core Opteron was taped out in June 2004, and then officially introduced in late April, 2005."

Thanks Kris.

 

[Furen]

I wonder if the dual-core K8L has taped out yet. Considering that quad-core takes a hell of a lot more die space (and I dont plan on buying quad-core for a while yet, lol) I'd guess that AMD is more interested in updating its dual-core series first.

 

[tatteredpotato]

At least it wont be till 2008

And since the k8l design is so modular wouldnt it be fairly simple to just remove two cores? I realize im probably oversimplifying the process, but it wouldnt require a complete redesign or anything.

 

[Pabster]

Don't be so sure. Quad-core takes more die space, but it also will bring in more $. And AMD needs it.

 

[xenolith]

I don't know about this... AMD better hope software developers kick it in gear in developing a significant performance bump in multi-thread apps. I mean how long have we waited for a mainstream 64-bit OS?

 

[TuxDave]

Now comes the next fun part of CPU design. Silicon debug.

 

[dexvx]

So right now its mid August. That means the absolute soonest they can come to production is around December. Correct me if I'm wrong.

 

[gobucks]

blah. what's the point of quad core right now? dual core is nice for multitasking and what not, but even so it's extremely hard to keep 2 cores busy as it is, much less 4 cores.

sure quad core will be nice eventually, but for the foreseeable future it's just gonna be expensive to make and to buy for little in the way of extra performance.

Also, 4 cores on 1 chip is gonna eat up a lot of power and generate a lot of heat, so clockspeeds will have to be reduced in order to keep things running smoothly. plus, considering how even with dual-core, overclocking is often limited by the weaker core, i imagine it will be even more so with quad core because you're odds of getting at least one slower core is higher.

 

[xtremejack]

CPU validation cycles are much longer. Usually a year from tapeout to production, not to mention the amount of trouble it would take to ramp these parts.

 

[dmens]

Originally posted by: dexvx
So right now its mid August. That means the absolute soonest they can come to production is around December. Correct me if I'm wrong.

15 months on a major redesign.

 

[Markfw]

well, if I don't have my opterons upgraded to dual-core by next summer, I can go quad ! Just paid off the house, so I need to pay a few other bill I robbed the money from first. (like property taxes...)

 

[Keysplayr]

Originally posted by: dexvx
So right now its mid August. That means the absolute soonest they can come to production is around December. Correct me if I'm wrong.

Usually a year from tapeout to availability. Unless they cut corners, which I have heard is not a good thing to do in CPU design.

 

[RichUK]

Originally posted by: keysplayr2003

Originally posted by: dexvx
So right now its mid August. That means the absolute soonest they can come to production is around December. Correct me if I'm wrong.

Usually a year from tapeout to availability. Unless they cut corners, which I have heard is not a good thing to do in CPU design.

but but but .. i thought they usually cut four corners

 

[OBCT]

but but but .. i thought they usually cut four corners

No, with this new design the chips will be circular.

 

[Viditor]

Originally posted by: dexvx
So right now its mid August. That means the absolute soonest they can come to production is around December. Correct me if I'm wrong.

You aren't wrong, and probably too conservative. My guess is that they will release when they said they would (mid-2007). Going from tapeout to shipping is usually ~9 months to 1 year.

 

[dmens]

Originally posted by: Viditor

Originally posted by: dexvx
So right now its mid August. That means the absolute soonest they can come to production is around December. Correct me if I'm wrong.

You aren't wrong, and probably too conservative. My guess is that they will release when they said they would (mid-2007). Going from tapeout to shipping is usually ~9 months to 1 year.

9-12 months on shrinks. 15 months on a major redesign is considered fast.

 

[arcas]

How is AMD dealing with memory bandwidth for 4 cores? Am I correct in assuming that system memory will effectively hang off one of the cores' memory controller while the other 3 cores are "fed" by the first over HT?

I recall seeing some benchmarks comparing the two types of dual Opteron motherboards: those where each CPU has its own local memory and those where only one has local memory. For synthetic tests, there was a noticible difference but not so much for real-world tests. I'm wondering if the same will be true of 4-way or 4-core setup? Will a memory bottleneck become apparent?

 

[Keysplayr]

It would be pretty neat if each core had it's own DIMM (or whatever) slot(s).

 

[ncage]

Originally posted by: xenolith
I don't know about this... AMD better hope software developers kick it in gear in developing a significant performance bump in multi-thread apps. I mean how long have we waited for a mainstream 64-bit OS?

Do MP coding is harder than you think. I am a software developer and you can run into so many bugs with MP coding. Some of these guts only show up ever 100,000 to 1,000,000 times you run through this section so it can be next to impossible to find bugs. I think we will gradually move toward a more Multi-core approach but there is only so much you can do in another core. THere are limitations. This is why going with highly number of cores really doesn' tmake that much sense. Sure dualcore was a boon but we are going to hit the law of deminishing returns. now of course all this power can be used on server type applications like Database servers, Web servers , ect but on desktop there is only so much we can do.

ncage

 

[Pabster]

Originally posted by: gobucks
Also, 4 cores on 1 chip is gonna eat up a lot of power and generate a lot of heat, so clockspeeds will have to be reduced in order to keep things running smoothly. plus, considering how even with dual-core, overclocking is often limited by the weaker core, i imagine it will be even more so with quad core because you're odds of getting at least one slower core is higher.

Yes, 4 cores on a single die is gonna consume some juice. But if they're produced on 45nm process technology...

 

[Viditor]

Originally posted by: Pabster

Originally posted by: gobucks
Also, 4 cores on 1 chip is gonna eat up a lot of power and generate a lot of heat, so clockspeeds will have to be reduced in order to keep things running smoothly. plus, considering how even with dual-core, overclocking is often limited by the weaker core, i imagine it will be even more so with quad core because you're odds of getting at least one slower core is higher.

Yes, 4 cores on a single die is gonna consume some juice. But if they're produced on 45nm process technology...

AMD has already stated that the quad core will have exactly the same TDP as their current dual cores...

 

[TuxDave]

Originally posted by: Pabster

Originally posted by: gobucks
Also, 4 cores on 1 chip is gonna eat up a lot of power and generate a lot of heat, so clockspeeds will have to be reduced in order to keep things running smoothly. plus, considering how even with dual-core, overclocking is often limited by the weaker core, i imagine it will be even more so with quad core because you're odds of getting at least one slower core is higher.

Yes, 4 cores on a single die is gonna consume some juice. But if they're produced on 45nm process technology...

If we're talking about the quad core that's planned for release mid next year, it's going to be on 65nm unless AMD did some voodoo in the fabs.

 

[Pabster]

Originally posted by: Viditor
AMD has already stated that the quad core will have exactly the same TDP as their current dual cores...

I'm actually referring to Intel with Kentsfield. 65nm Kentsfield samples are already out and I have no doubt Intel will ramp up 45nm well before AMD.

And before anyone spouts off about me being a fanboy... I was spot on about Intel having 65nm up and running before AMD, despite the bevy of criticism I took.

 

[Furen]

Originally posted by: arcas
How is AMD dealing with memory bandwidth for 4 cores? Am I correct in assuming that system memory will effectively hang off one of the cores' memory controller while the other 3 cores are "fed" by the first over HT?

I recall seeing some benchmarks comparing the two types of dual Opteron motherboards: those where each CPU has its own local memory and those where only one has local memory. For synthetic tests, there was a noticible difference but not so much for real-world tests. I'm wondering if the same will be true of 4-way or 4-core setup? Will a memory bottleneck become apparent?

The whole thing will only have one memory controller which is connected to the Crossbar. The Execution cores will also be connected to the crossbar so they would request to use the memory controller as needed. L3 cache should lower contention for the memory controller and, because it will be shared, keep core to core traffic off the crossbar (not that this should make a huge difference in single-socket systems). Here's how I think about it:


Core 1->|
Core 2->|L-3|----> |Crossbar|->Memory Controller/HT links to other CPUs/Northbridge
Core 3->|
Core 4->|

(Had to simplify it a lot 'cause this damn thing erases extra spaces)

Each core having its own L1 and L2 (I dont believe I missed anything too important except the SRI). I do wonder if L3 will be partitioned like in Core (each core gets a segment of cache assigned to itself) or if it'll be a single pool that any core can draw from.

 

[Viditor]

Originally posted by: Pabster

Originally posted by: Viditor
AMD has already stated that the quad core will have exactly the same TDP as their current dual cores...

I'm actually referring to Intel with Kentsfield. 65nm Kentsfield samples are already out and I have no doubt Intel will ramp up 45nm well before AMD.

And before anyone spouts off about me being a fanboy... I was spot on about Intel having 65nm up and running before AMD, despite the bevy of criticism I took.

Agreed on the 45nm ramp...Intel is predicting to have 45nm ~6months before AMD, though AMD IS closing the process gap.