I suspect after this episode any chance of a Biostar sponsor or whoever is affiliated with Biostar has gone out the window.

just figured I'd make mention it's not a great idea to bite the hand that could be feeding you in the future.

other than that it was an ok broadcast good job.... just surprised you decided to openly bash a potential sponsor.

Audio Needs normalized, lots of peaks and valleys making it less enjoyable than normal, only 8 minutes in!

Hey guys,

No problems with my filesize, because I handle it all through iTunes. Weird sound effects at ~15 min mark, and ~15:43 mark. Also, at roughly ~20:20, Mr. Rogers rings a doorbell or somethin...

Love you too.

ssid.

Scott - Thank you for the decently elaborate piece on my question.

------------------------------------

#4 - Prototyped
1. There really isnt anything remotely like a pricematch here in Sweden with regards to matching clockspeed. But both the 9450+- 1 model and the 8500 are really the sweetspots in each of its release trains. While the much more expensive E8600 is actually the same as the Q9300. While the 8400 is about half the price. So beyond what the extra chip/cores inside the package costs , i consider the E8500 and Q9450 equal in terms of sweetspot.
3. Good writeup. I've had some of my own speculations go in a somewhat parallel way to some of yours.

------------------------------------

I ended up with the Q9450 after some more reading. And due to a good bit of multitasking, all the cores see use, and windows seems to do a decent job of keeping them all used and everything is really snappy. And getting it up to 3.2GHz was a breeze with the Thermaltake Ultima 90, even with a slow 18db papst fan. Prime95 x64 runs all the tests solidly. Seems todays CPU's have somewhat of a 20% headroom on aircooling nowdays.

Although it seems i got bitten by some oddity p45 memory bug with my Asus p5q3(ddr3), which have me running the memory at 1333 instead of 1600 as it is speced. Which means that even if prime95 runs as a champ, as soon as i test 3dmark in the OC-state at 1600mem, it crashes intermediately. So im still fiddling with that and also voltage on the northbridge, etc. A googling on some forums gave a bunch of problems with the p5q3 in different bioses, etc. Seems to be a problem only with the p45 based boards sadly enough.But some reports about the new 130x bios seems to be decent so far. So thats the next thing I'm fiddling with. But on another windows install. Trying an upgrade to 1201 actually hosed the Vista x64 install and forced me to do a repair after having stepped down again.

A little disappointing that the Tech Report podcast doesn't want to talk about the detailed technology in Intel's new CPU. Then again I love the detail, don't know if that's what the average listener wants.

Re: dual-core vs quad-core:

1. There's no quad-core processor from Intel at the same price of the E8400 (~$175 from an e-tailer) currently. The cheapest is the Q6600 at ~$200.
2. The Core i7 processors that Intel is about to launch in October are native quad-core at 45 nm (and Nehalem processors can be built from native dual core up to native 8-core depending on market segment). However, if you're talking about 8-core single-socket desktop processors, that's probably going to have to wait for (or perhaps past) 32 nm Westmere, yeah.
3. Scott's right in that you don't generally "optimize" for multiple cores the same way you might use compiler optimizations to target software compilation for a new processor, or use compiler autovectorization to use SIMD instructions like SSE (and in order to use SIMD instructions effectively you often have to hand-code anyway since this is hard for compilers to do automatically in the general case). Parallelizing apps generally involves human thought and effort. One of the issues is that not every technique used by software is parallelizable; when there are serial dependencies in such a way that you cannot start to process the next part before the first part is processed, you essentially have to process in lock-step.

Fortunately very many problems are solvable in parallel. Unfortunately, it isn't always simple for the average programmer to figure out how to parallelize the technique the program uses to solve its problem, especially on computationally intensive techniques ("algorithms"), and today, even if the programmer does figure it out (the hard part), the parallel nature of execution makes certain kinds of bugs likely to creep in (inconsistent reads, unisolated processing, lost updates, corruption of memory contents, other race conditions, deadlocks, livelocks, etc.) that take a lot of time and effort to solve (due to the fact that a consistent, repeatable scenario that can reproduce the bug frequently cannot be constructed when code runs concurrently), with the result that many programming teams don't bother to parallelize their software.

Parallel programming is by no means a new field. We've had multiprocessors for 40 years now, even if integration of multiple processors onto a single IC is relatively new. The concepts, programming techniques and so on are identical. The issue is that computer scientists have yet to figure out easier methods for programmers to suss out how to break down solutions to problems in such a way that the pieces can be run in parallel, and still provide correct, complete results. There are already people coming up with domain-specific languages that make it easier for programmers to express parallelism in their programs less painfully so that the bugs resulting from parallel execution can be reduced.

So I expect that eventually, one of the two issues limiting the availability of parallelized software (the part where parallelism is not simple to express, resulting in difficult-to-detect-and-eliminate bugs) will be minimized. And perhaps researchers at universities and large research organizations can come up with easier methods that programmers can use to discover ways to parallelize more workloads. Once these happen, parallel programming and maintenance of parallelized software will become easier and cheaper, making more parallelized software available. Once that happens, increasing numbers of cores and logical processors will show real benefit.

Until these things happen, I expect that for desktop-style workloads, it will be most cost-effective and performant to pick processors that, as Damage mentioned, represent the smallest native computing unit, which at least for next year will mean quad-core processors, and until Core i7 is out, dual-core processors.

Now when it comes to when quad-core processors will be at least as fast as dual-core processors on single- or dual-threaded workloads at similar prices, I suspect that once quad-core units are what Intel tries to sell hardest, the dual-core processors will slip into the budget-only range at low clock speeds and the highest clock speeds will only be available on quad-core processors. At this point, it will be more advantageous to use a quad-core processor, even if your CPU-bound application is single-threaded. We saw this happen with single cores when native dual-core processors became the norm. We'll see it happen again when quad-core processors become the norm and the manufacturers decide to position dual-core processors as the budget-level alternatives rather than the mainstream processors.

Re: Core i7 being high-end, yeah, X58 does seem like an "Extreme" class chipset, and X58 boards will probably cost quite a bit, but the processors aren't known to be (at least so far) all really expensive. The cheapest one (Core i7 800, 2.66 GHz, still triple channel, 4.8 GT/s QuickPath) is rumored to be $284, which really isn't all that much. [1]

[1] http://en.wikipedia.org/wiki/Intel_Core_i7

Re: LucidLogix's chip, I'm guessing that with the possibility of SLI certification for Intel's X58, they just lost a big chunk of their market. Now NVIDIA is expected to build LGA 1160 chipsets [2] for the mainstream Nehalem (not Core i7 [3]) processors, to which I wonder whether they'll extend SLI licensing, given that they want their own silicon to compete in this market. If they don't LucidLogix will be able to market their Hydra to the entire market again. (I don't consider AMD and NVIDIA chipsets to sell enough to form a compelling market by themselves.)

[2] http://www.bit-tech.net/news/2008/08/29/nvidia-has-a-qpi-license/1
[3] http://blogs.intel.com/technology/2008/08/getting_to_the_core_intel...

Re: Turbo Mode, it'll clock up to two speed grades higher when the other cores are idle. This tells me that Intel must be validating these processor cores to two speed grades higher (i.e. not just iTAT/Orthos/Prime stable, which isn't nearly as rigorous as Intel's use of validation pins on the processor), which suggests to me that that's the bare minimum "overclocking" headroom, since the processor must run correctly at those speeds.

I'm hoping that motherboards will provide for the ability to clock up to at least the validated clock rates for people who don't particularly care about the power consumption.

Great podcast guys!

Nice to see you commented on Lucid's technology when i emailed you about it 3 weeks ago ;]

looking forward to next week and the Case reviews.

"I knew him before he was who he is."

Name dropper! :-)

But seriously, cool, very cool. Small world. Given that previous relationship, I'm surprised he's not given you/TR any shout-outs or exclusive interviews lately.