Intel 45nm dual core 'Wolfdale' coming Q3 07

[harpoon84]

http://resources.vr-zone.com//newspics/Oct06/11/Wolfdale.png

Looks pretty impressive if true. 3.5 - 4GHz stock, can anyone smell 5GHz+ with a bit of overclocking?

6MB L2 and 1333FSB sounds nice as well, but what really impressed me is the projected 57W TDP.

PS. I hope current 965/975 mobos will support these chips, but knowing Intel's track record of changing VRM requirements constantly, I'm not holding my breath.

 

[Duvie]

I like the yorkfields for my own taste....I will take 4cores at 3ghz....

 

[harpoon84]

Originally posted by: Duvie
I like the yorkfields for my own taste....I will take 4cores at 3ghz....

Unlike you, I find QC overkill for my needs, being a gamer mainly. Unless you tell me ALL games in 07/08 will need QC to do 'special physics effects' ala Alan Wake, then I may be tempted, but I doubt that's gonna happen anytime soon.

 

[Lonyo]

3.5GHz sounds pretty reasonable, give that it'll have been 12 months + a process change since the original quad cores
And they had Kentsfield demo running at 3.73GHz reportedly.

If 57W is true (maybe at the lower end it will be) then AMD are going to be hard pressed even with K8L to show they have any competetive performance/watt chips, without resorting to horrifically expensive EE's.

 

[Duvie]

That potentially non OC'd yorkfield will be 160%+ faster then my quad 2ghz box and if I can OC it moderately It will be more then double the speed....with only one chip and likely cooler...

 

[Polish3d]

AMD still stuck at 2.9ghz in Q3 2007?

 

[Duvie]

Originally posted by: Frackal
AMD still stuck at 2.9ghz in Q3 2007?

not with 65nm parts and current speeds of 2.8ghz...I imagine they will be 3-3.2ghz...

 

[coldpower27]

Originally posted by: Frackal
AMD still stuck at 2.9ghz in Q3 2007?

Were talking about Rev.H parts at that speed which have improved iPC so things aren't as bad as it looks, remember we have learned that clockspeed isn't all there is to the equation.

 

[BrownTown]

But K8L will at best be equal to Core 2 in IPC, not better, so clock is the deciding factor there.

 

[coldpower27]

Originally posted by: BrownTown
But K8L will at best be equal to Core 2 in IPC, not better, so clock is the deciding factor there.

No, at this current time what we know is that Rev.H will bring IPC improvements to AMD, we don't know yet how much the exact improvement is yet. The exact improvement will be deciding factor in the end. That information won't be known till ES get out into the wild or AMD demos Rev.H parts.

 

[BrownTown]

we know what the improvement are already, and they pretty much just bring K8L up to the level that Core 2 is at now, maybe the IMC will help push it over in terms of IPC, but the two cores will be onyl a few % apart in terms of IPC.

 

[dmens]

Originally posted by: BrownTown
we know what the improvement are already, and they pretty much just bring K8L up to the level that Core 2 is at now, maybe the IMC will help push it over in terms of IPC, but the two cores will be onyl a few % apart in terms of IPC.

dunno, seems like k8l's improvements are almost entirely geared towards float/vector performance. so k8l will win in floating throughput, but what about generalized computing? given the thermal estimates, and if the fp bandwidth increase is accurate, that leaves very little to expand general machine depth.

float is a c2d weakness with little recourse, perhaps amd is targeting it.

 

[coldpower27]

Originally posted by: BrownTown
we know what the improvement are already, and they pretty much just bring K8L up to the level that Core 2 is at now, maybe the IMC will help push it over in terms of IPC, but the two cores will be onyl a few % apart in terms of IPC.

No we don't, we have estimates, but that is all they are. Rev.H could be better or worse then Core 2 Duo per clock we just won't know till the processor arrives.

Originally posted by: dmens
dunno, seems like k8l's improvements are almost entirely geared towards float/vector performance. so k8l will win in floating throughput, but what about generalized computing? given the thermal estimates, and if the fp bandwidth increase is accurate, that leaves very little to expand general machine depth.

float is a c2d weakness with little recourse, perhaps amd is targeting it.

I would agree it looks like Rev.H is focusing on floating point performance, which AMD has been strong at since the K7 architecture was released. I can't say for sure however what will happen on the Integer side of things though.

 

[Viditor]

Originally posted by: dmens

Originally posted by: BrownTown
we know what the improvement are already, and they pretty much just bring K8L up to the level that Core 2 is at now, maybe the IMC will help push it over in terms of IPC, but the two cores will be onyl a few % apart in terms of IPC.

dunno, seems like k8l's improvements are almost entirely geared towards float/vector performance. so k8l will win in floating throughput, but what about generalized computing? given the thermal estimates, and if the fp bandwidth increase is accurate, that leaves very little to expand general machine depth.

float is a c2d weakness with little recourse, perhaps amd is targeting it.

I think that when AMD designed the architecture for K8L (~3 years ago), they couldn't have known much about anything on C2D's weaknesses or strengths...
I agree that there is a strong increase for FP, in fact it appears that K8L will about double the internal throughput from K8. But the big question marks for me will be:

1. The new out of order load execution
2. The enhanced branch predicters
3. 32-bit fetch instead of the 16-bit fetch of the K8 and Conroe

Taken together, this could (hypothetically) be a much more powerful core than Conroe, but we just don't know how much of this is true and how well it will work. I've seen wild claims in both directions (from only a 5% increase up to a 50% increase), but I just don't have enough info on what is real...nor do I understand uA well enough to make an accurate judgement even if I had that info.
Personally, I tend to think (based more on intuition than anything else) that K8L's core will be about equivalent to C2D and that the reduced latency from HT and the ODMC will keep it at ~5-10% faster clock-for-clock.

 

[dmens]

you gotta stop lumping performance into a single comparison, it has to be divided into different workloads. the int vs fp/simd is the simplest bucketing, and already there seems to be a divergence of priorities between the two design teams.

imo, not a single thing i read on k8l will help much on the general computing side... which is interesting since speed demand in that type of workload has hardly plateaued.

 

[Viditor]

Originally posted by: dmens
you gotta stop lumping performance into a single comparison, it has to be divided into different workloads. the int vs fp/simd is the simplest bucketing, and already there seems to be a divergence of priorities between the two design teams.

imo, not a single thing i read on k8l will help much on the general computing side... which is interesting since speed demand in that type of workload has hardly plateaued.

I really am trying not to lump them together (because that is a fair point)...
But if you could be more specific?
For example, the 3 points I listed:

1. The new out of order load execution
2. The enhanced branch predicters
3. 32-bit fetch instead of the 16-bit fetch of the K8 and Conroe

Don't seem to me to be very FP specific...they actually seem to be more powerful in "the general computing side".
As I said, I'm no uA expert...so if you could point out the error of my ways, I would appreciate it. :thumbsup:

 

[bobsmith1492]

Isn't it funny that the Netburst, ostensibly the king of clocks, is already being beat by other processor lines? Well, not beat, but when it only hit 3.8 max and these are 3.5, it seems pretty obvious that someone neglected some things with the Netburst design... (thermal issues?)

If heat was the main problem with Netburst, then it seems Intel's latest focus on thermal efficiency will have good payoffs in top clocks as well as IPC (or whatever) efficiency.

 

[dmens]

i cannot assess the impact of out-of-order loads vs without because p6 always had it, but i dont think the improvement is literal. i doubt k7/k8 lacked any kind of sequencing on memory uops. branch predictors has been a dead end for years, won't get much perf out of +0.1% acc. on 98%, plus those things are heavily workload dependent, across the board improvements are very small in general. as for cache improvements, those things always have some tradeoff, like longer use latency. those caveats dont get any time in the marketing slides.

also, netburst is the king of clocks... consider process (130/90 vs 65/45), and fast clocking on certain circuits (marketing term is double-pumped, or something).

 

[bobsmith1492]

Double-pumped FPUs or whatnot? Anyway, I wonder what one could do on a 45 nm process... they should make some, just for old-times' sake.

 

[Viditor]

Originally posted by: dmens
i cannot assess the impact of out-of-order loads vs without because p6 always had it, but i dont think the improvement is literal. i doubt k7/k8 lacked any kind of sequencing on memory uops. branch predictors has been a dead end for years, won't get much perf out of +0.1% acc. on 98%, plus those things are heavily workload dependent, across the board improvements are very small in general. as for cache improvements, those things always have some tradeoff, like longer use latency. those caveats dont get any time in the marketing slides.

also, netburst is the king of clocks... consider process (130/90 vs 65/45), and fast clocking on certain circuits (marketing term is double-pumped, or something).

Actually, the out-of-order loads for memory is one of the big changes for Conroe (and IMHO one of the most effective)...I believe they call it "memory disambiguation".

Edit: BTW, I don't know if this has been posted (I've been on planes, trains and automobiles for the last week...), but Hans De Vries has a nice breakdown of the K8L die