First of all, I never said Xenon is faster than a tri-core Phenom. Re-read that thread, especially my latest post. As for 16 threads not being significantly > than 4 threads, in the PC space this is true. Do you know of any meaningful workloads outside of server and HPC realm that scale that well?
Compression/decompression and encoding/decoding scale well. So does pretty much any type of 3d rendering while gaming is starting to take advantage too.
But the fact that multicore scaling is limited today means little for what it will be like in a few of years when Nehalem level CPU's enter mainstream - particularly in relation to games. Software is still playing catch up but it won't be long before software is generally catering for multiple cores.
I would expect SMT to make more of an impact than the additional cores (to utilize the idle execution resources of each core), and only if implemented well, and even then only in corner cases of extremely thread parallel workloads.
That doesn't make much sense. Code must be multithreaded to take advantage of extra threads and your always going to get more performance from running a thread on a dedicated core than running that same thread on the second hardware thread of an already utilised core.
I figured we were discussing useful performance, but it's clear you're just discussing available execution resources, in which case there's no doubt Nehalem will be superior to Penryn by a wide margin.
Well its both. Usefull performance as you call it should be a fair bit higher but total available power is obviously much higher. Iits not as though that power is locked away though. Certain code types will use it today, more will use it when Nehalem is on the market and many more will use it 2-4 years from then. Anyone planning to keep the CPU for a few years will see the benefit of those cores far more so than what your proposing they would today.
Anyway, it was you that was claiming 6 threads would give a significant performance boost over 3 in the Phenom thread. So if nothing else an 8 Core Nehalem can dedicate a core per Xenon thread to ported games while Penryn does not have sufficient threads to do so - just like the Phenom. You
were saying thats a big advantage weren't you? The same could also be said for ported Cell code.
Physics, while very parallelizable, do not put a heavy strain on any modern performance-class processor, whether it be specialized hardware (PhysX), multi-core GP CPUs, GPUs, or multi-core in-order console processors.
That doesn't make much sense. Physics can scale to your available resources. PS3 will likely be doing things with physics that Xenon simply can't pull off. Similarly their are feats claimed for the Ageia PPU that are not possible on current CPU's. Nehalem should be able to breeze through these kinds of things making "PPU only" like physics and full PS3 level physics possible on the desktop (if its not already so).
Unless Intel is able to get the Havok engine to auto-parallelize physics and scale effects with core count, I doubt we'll see much difference (short of SSEx optimizations and optimizations specifically for Core MPUs).
Why would Intel not be able to get Havok to scale with core count? Like you said, its a pretty trivial task and its obviously in their best interest to make it happen. In fact they have already demonstrated their interest in doing so with the ice fighters demo and thats before they bought Havok. Havok will scale with multicores and Intel will encourage devs to take advantage of that, ala Alan Wake. You can count on that.
First of all, where'd you get the crystal ball? You don't know how well Shanghai will compare to Penryn, nor Nehalem.
Secondly, Nehalem is not a native 8-core design, it is also 2x quad-core dies on a package.
No, Nehalem is a native 8 core design, check it out.
And Shanghai is little more than a die shrink of Barcelona with extra L3. If you think thats going to allow it to seriously compete with Penryn then your optimisitic IMO. By the time its launched Penryn could easily be clocking over 3.6Ghz so Shanghai will have to be clocking at something similar to have a chance of competing, unless its IPC is significantly higher than Barcelona that is. And how can you expect two of them to compete with Nehalem assuming that Nehalem is going to give pretty substantial improvements over Penryn on a per core basis and be native octo core?
%á#P> </P>
AMD hasn't even attempted to create a product for the HPC space? What do you call Opteron? When was the last time you checked the Top 500 supercomputer list? Do you not remember who pioneered stream computing in the first place? I'll give you a hint, it's not Nvidia.
I call Opteron a server chip just like everyone else does. Opteron is clearly not aimed at the same markets as Cell and Larrebee. It might be offered to those markets as AMD's only option but its certainly not designed to be a serious competitor to them in their specialised fields, i.e, high performance stream computing.
I don't think anyone outside of MS/Nintendo/Sony know precisely when their next-gen consoles will launch, so it's tough to justify any statement about the maturity of any future PC product's viability as a console part when the availability of said PC part is also just a timeframe and not a hard number. I'm certainly not going to argue against using a Sandy Bridge or Larrabee derivative in a console from a performance standpoint, but I think it's fair to assume such a derivative would have to be rather cut-down in order to meet price/performance targets of a relatively low-cost consumer electronic product. Remember, we're talking about a part of the build cost of an entire machine that costs as much as only that component does in the PC space.
I agree, however as you say the timefrmes are important here. If Sandy Bridge is a year old when the next gen consoles launch then a low end/low power version of it would be conceivable. Not that it matters, it was just an off hand comment on my part, not something I want a detailed and ultimatly fruitless debate over.
