Intel Demos Core 2 Extreme QX9650 Quad-Core At IDF

MojoKid writes "Intel demonstrated a dual socket gaming rig at IDF this week, based on their Skulltrail platform with the X38 chipset. The interesting thing about this machine wasn't just that it had 45nm quad-core CPUs in its sockets, as well as PCI Express 2.0 capable slots, but also that it was running a pair of NVIDIA graphics cards in SLI. That's right, SLI on an Intel chipset. No word whether or not X38 would officially be supported with SLI just yet. In fact, NVIDIA representatives noted Intel was buying NVIDIA nForce 100 SLI Southbridges just for this one Intel motherboard model."

Comments

SSE4 is overrated

[Silverlancer]

The SSE4 ESA SAD instruction (for an exhaustive SAD motion search) has been touted constantly as the "new big thing" with DivX benchmarks showing 100% speed improvements, etc, etc. This is generally bullshit. The DivX encoder was specifically modified so as to use the exhaustive motion search in its normal encoding, basically contriving the test to work faster with SSE4. Talk to anyone working on an encoder and they'll tell you the same--not only is an exhaustive motion search generally useful, but there are equivalent algorithms, such as successive elimination (SEA) that are actually faster implemented in software than SSE4 is implemented in hardware with mathematically equivalent results. The main developer of x264, for example, has stated that SSE4 offers basically no useful instructions whatsoever.

The chip as a whole, on the other hand, is quite awesome, but I think its important to correct this bit of common misinformation.

Re:SSE4 is overrated

[Silverlancer]

Slight correction, "not only is an exhaustive motion search generally useful" > "not only is an exhaustive motion search not generally useful".

Re:SSE4 is overrated

[samkass]

I assume you're complaining about Intel documents such as this one that show a 1.6x to 3.8x speedup for certain HDTV encoding operations. There are still some likely reasons someone might pick SAD for encoding. One reason might be patent coverage-- the better the algorithm, the more likely it is to be patented these days. Providing a baseline fast algorithm that's part of all future Intel hardware and can be used without patent problems might be nice. Also, just because something can be done "faster in software" doesn't mean it's really faster. If it doesn't use the SIMD pipeline, it's occupying more of the int or fp pipelines and thus may have fewer potential parallelism opportunities and/or stall the pipeline more often. With SAD offloaded onto the SIMD pipeline one could do other things with the other pipelines-- perhaps even a combination of both algorithms running in parallel (assuming sufficient instruction dispatch speed) to get even more speedup.

I think there's a lot of as-yet unrealized potential here, and it will be interesting to see where it leads. It won't be the end-all be-all of encoding, but it will be another arrow in the quiver.

all this and the kitchen sink too...

[downix]

What are the chances a real machine will ever ship with these features? I highly doubt. They pulled every stop for a demonstration, and now the beancounters will start knocking features off to try and save a buck. End result, customers will likely never see such a rig in mass production, which is a shame, for it would make me seriously consider Intel again.

when i was a kid...

[kesuki]

we thought 640k was enough to run games in and 16 colors was good enough for anyone! now we have to have 2-4 graphic cards, and 8 processing cores??? and probabbly 8 gigs of ram!!! dosent anyone think of how many watts these gaming rigs use anymore? i mean wow... pulling 49 amps over the 12 v rail... you might as well sell them with a dc generator and solid copper power rails.

seriously add in liquid cooling and cold cathodes and a 52" HDTV and youre talking over 3 killowatts of power draw... Im glad i play blizzard games, not only to people play them for a decade after theyre made, the initial launches try to have a configuration setting that will lower the bar and let less impressive systems play too.

sure their engines might not be so impressive that youd need quad 100 pixel pipeline cards... that themselves have 2 GB of ram on them.. or a system with 4x processors with 8 gb or ram... but i think the gaming industry has gone too far ever since they realized there was a market for $600 gaming cards..

Re:when i was a kid...

[kesuki]

actually about this technology trickling down to the mid and low end im going to have to take point with... with ati releasing 2 pixel pipeline cards as the low end market things arent getting better on the low end.. and the mid range cards are still as expensive as they were 18 months ago. they have been really careful not to bump up the mid range very much, you still pay over $120 for a 12 pixel pipeline card, and those are using reject 16 pix pipe chips. card vendors really arent trying to go for economy of scale. but theyre trying to maximize their bottom lines.. and as long as they keep charging mucho dinero for any chip with more than 4 pixel pipes its just going to be more of the same in the mid range... not that long back i remember their being $40 graphic cards in best buy, but last time i was in their they sold nothing under $80.

by the time 100 pixel pipeline cards become affordable ill probabbly be an old man... at least the way the vendors are dragging their toes at lowering the cost of performance cards. before the latest generation of ati card the lowest number of pixel pipes they had in a card was 8, this generation they sell defect 4 pipe cards as 2 pipeline cards, and nvidia isnt any better at lowering the cost of decent graphic cards.

Re:Fans -- Compressors!

[Technician]

What's a little fan noise and power consumption for a few fans compared to a triple cascaded refrigerated rig like this overclock demo.

http://www.custompc.co.uk/news/601310/idf-556ghz-penryn-breaks-three-benchmark-records-in-two-minutes.html

That is 5.56 Ghz demonstrated at the IDF. It ran and broke 3 CPU speed records in under 2 minutes. This is extreme! Don't ask how much power the cascaded refrigeration system uses. ;-)

"Worth then proceeded to show off the fruit of his labours by claiming he could break three world benchmark records in just two minutes. This included SuperPi 1M, where he beat Team Japan's previous record, and he then went on to run AquaMark where his score of 273,000 trashed the previous record of 267,000. Finally, he then ran the 32-bit version of CineBench 10, where a score just shy of 20,000 (he didn't reveal the actual score, unfortunately) again clinched a world record for a quad-core CPU."

Re:Super duper deca-core X8789FDS extreme

[plover]

Punk. I see your Z80 and raise you a 4004.

I was a youth BEFORE the invention of the microprocessor. I did my growing up years on a CDC Cyber 72 mainframe, but we were all pretty excited when the 4004 came out. The 4004 was actually the first microprocessor I ever owned: in a Mattel Electronics Football game. Before that, I had an ENIAC -- which was just six really smart switches I could wire to do all sorts of cool tasks. And when I was really little, I had a Dr. Nym, which was a marble-and-gravity cascading flip-flop game.

Now get off my lawn, you damn kids!