Slashdot Mirror
Intel Launches Pentium Extreme Edition 955
BSG Man writes "Intel's 3.46 GHz Pentium Extreme Edition 955 dual-core processor launches today, and
HotHardware has a full review with benchmarks on Intel's new i975X Express
based D975XBX motherboard. This processor is based on Intel's 65nm (or .065 micron) Presler core with 2MB of full speed, on-die L2 cache dedicated to each core, for
a whopping 4MB of total L2 cache. As expected,
the new Pentium Extreme Edition 955 scores well in encoding, desktop
business and a few professional rendering tests but overall it's given a run for
its money by AMD's Athlon 64 X2 4800+ dual-core processor, especially in gaming
scenarios."
Pentium Processor Extreme 955
Price: $1,112.37 - $1,393.49
AMD Athlon 64 X2 4800+ 2.4GHz, Toledo, Dual-Core, 2x1MB L2 Cache, Socket 939, 64-bit Processor
Price: $780.74 - $1,185.00
More run, less money, it would appear.
SCO employee? Check out the bounty
The Nvidia version 80.0 and above drivers are multithreaded, so they take decent advantage of dual core / SMT chips.
A couple of games, I know Quake4 for one, have been benchmarked with the dual core offerings edging out the fastest single core products. How much of that is related to the multithreaded Nvidia drivers and how much is from threading in the games themsleves I don't know. (The effect is not on all games when running the det 80s, so I'd assume Q4 has at least some usefull (beyond file I/O and netcode) threading).
Heat production and the subsequent power needed may be two of the reasons they are not now going to the higher speed processors. Intel using a smaller die size though may result in higher speeds again soon.
How about Quake 4 and Call of Duty 2?
They already have, just not on dual-core parts. Expect major reduction in clock speeds on Intel CPUs once they migrate away from Netburst-based chips towards more efficient designs such as Yonah, Merom, and Conroe.
Conroe is the chip to watch.
Just a short lesson in processor speed:
The throughput of a processor is related to the number of pipeline stages (think of a laundry room; you have three "stages"- washer, dryer, folding table. You can have three "loads" ("instructions") in the laundry room ("pipeline") at one time). When you shorten the clock time (i.e., increase the 'speed' to a higher number of Hz), you usually do this by adding more stages to the pipeline. This results in a longer pipeline (a given instruction takes more cycles to complete, AKA longer latency), but also gives you greater throughput (some instruction finishes every N seconds).
Recently, additions such as dual core changed all of this. Instead of fscking with the pipeline (in general), you just add a second pipeline. You double your throughput without affecting latency or timing. Other changes, such as reducing the number of pipeline stages while maintaining the same clock cycle, also result in an increase in speed. As well as greater on-die L1/L2 cache, since it significantly speeds up memory accesses.
This is why Intel is trying to get away from the association of GHz == performance, and why AMD a long time ago started using numbers (e.g., 4800) instead of clock speed.
So, the bottom line is that a 3.5GHz processor is not faster than a 3.1GHz processor or a 100Hz processor (well, probably the last one) because it has a higher clock speed; the clock speed is one symptom of some of the techniques used to increase performance. It's a lot like looking at a car and determining it's max speed by its MPG; sure, all other things equal (aerodynamics, etc.), a lower fuel efficiency means a higher powered engine means top speed- but this doesn't work if you're comparing a Motorcycle to a Hummer.
b) The whole reason why AMD outpaced Intel, was because they didn't go for raw Ghz, but instead used the existing power more efficiently, with consequently less wasted energy (ie heat)
c) Clock speed is related to the speed of light, or how far an electron travels in a given time. The higher the clock speed, the quicker an electron moves, and the closer each transistor has to be to be to each other in order not to waste the advantage of the faster clock speed. There are limitations as to how "close" you can get the transistors before quantum tunneling occurs and you get leakage between channels. More here.
Did you steal that UID, or have you been under a rock for the last 10 years ?
First, the fastest overclocked Intel processor was something like 7 GHz, so it is evident that the electronics are capable of substantially better performance.
Second, the new Intel chips are hyperthreaded and multicore...
The biggest limitation is moving data around, not the CPU itself. Adding HyperTransport, DMA, etc, to the CPU itself would be a Good Thing, as would doubling the width of the data bus.
Executing all possible paths is not efficient when combined with hyperthreading, as you're wasting processing elements. Probabalistic branching (where you only follow one branch, but it's the most likely) would seem more efficient and would free up more elements for better threading.
I don't believe registers are ping-pong buffered, but it would save having to wait on writes if you still need to do reads because of the difference in timing from multithread execution.
Nobody uploads microcode to CPUs, but everybody runs code that would be efficient if run internally on the CPU. It would be good if the OS could upload atomic architecture-specific hardware operations into the CPU as pseudo instructions. Save having to hunt through physical memory for common tasks that will likely fall out of cache if you rely on that.
Processor overheating is a big problem and keeps the speed down. Processor casing simply isn't optimal for keeping the internals cool. It wouldn't be hard to improve the heat transfer from the chip surface to the casing surface.
Processors aren't made from optimal materials. If you're using silicon, for example, you want something that is single isotope, stressed and allowed to crystalize slowly. It's substantially cheaper to produce flawed silicon wafers, but they will never perform as well.
Along with this, I've learned that the reason aluminium is the most popular for CPU interconnects and copper is second, with silver unused, is problems with silver being too reactive and copper being only just managable. If they could find a way to prevent the silver from reacting with the rest of the CPU - should be possible - then you'd improve speeds there, too.
Electron leakage is a problem, as it also imposes a speed limit. Not sure how you'd prevent it, but there might be ways to limit the problem. Electrons have spin. It is certainly possible to polarize something by spin, and it is certainly possible to filter by polarization. There MAY, therefore, be ways to limit the impact of leakage and therefore ways to bypass the speed limit such problems would otherwise cause.
It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
ATI Catalyst 5.12 also added support for multiple cores to improve performance as well. I think it's the same thing as the nVidia driver situation.
All your base are belong to Wii.
955 is the name of the chipset (north+south bridge) on the mobo. It's slightly higher speed when compared to the similar 945 chipset.