This is a question I bring up because I spent some 2 hours on Ebay checking out potential motherboards for my new PC, a backup PC that will serve as my main PC when I am on my Easter and Christmas breaks from university and perhaps stand in for the actual main PC should it break (if my Graphics Card or CPU become silicon crackers through an OC'ing accident or a fall, I'll probably have to wait 2 months for the money to get a comparable replacement). One of the specimens I came across was this, an Lga771 Dual CPU eATX motherboard for £30; accompanying CPUs were similarly astonishing at £7 for two and Israel can provide a suitable stack of memory at 8GB for £25. Better still, spend £33 more and you get 8 cores with 16MB total L2 cache :)
To put this into perspective, £40 will get you a bog-standard Lga775 board that supports Core 2 Duo CPUs (sure, you can go as low as £15 if you fancy the Intel 915 Chipset THAT much), I spent £8 on a Pentium E2160 that didn't play ball with the ASUS P5SD2-TM/S I had because of it's obscure SiS chipset from the depths of Netburst, and I hold onto a stick of 2GB Corsair Dominator which has a bluescreen-churning brother simply because to get the same amount of DDR2 for £25 these days makes you supernaturally fortunate.
What this signals to me is that Dual CPU motherboards are becoming less relevant to the common man - Why would an enthusiast get an eATX monster-board that requires a corresponding cage to put it in (looks like a deal? It comes with everything but all the things that make it a computer case as opposed to bent metal and plastic... thanks lads) when a single CPU socket can play host to 8 CPU cores, even if the individual cores are not much cop? There is also the fact that the cores arms race between Intel and AMD has juddered to a halt, largely due to the fact that programming an existing single-threaded game engine for parallel computing is a task comparable to writing an entire new DirectX Generation.
So why were these things desirable to begin with? I suspect that back in the days when Pentium 4 was "winning" the Gigahertz war while simultaneously being tailed by Athlon XP at every turn (so much so that AMD called a 2.2GHz CPU a "3200+"), trying to edit a word document while a game was idling in the background could be described as flaky at best and rage-worthy at worst; what we needed, we thought, was a second CPU to handle the word document or perhaps something more risque like running video encoding alongside Bf1942 or maybe just open piles of word documents without reducing the windows desktop to a single-digit frame rate.
Then along came Athlon 64 which roundly trounced Pentium 4, followed by Athlon 64 X2 2 years after in 2005; Netburst's final stand was with the Pentium D, the Presler models practically on fire with a TDP of 130 watts. Two CPU cores in one package was a big deal in 2005 simply because it became so much less stressful to Alt-Tab Half life 2 to look up a walkthough owing to the second core being available for the browser; as was said before, parallel coding for consumer apps is a significant stumbling point even today. In 2006, Core 2 Duo from Intel was released, finally putting that crispy-pink burnt old Netburst dog to rest and ushering in the "multicore war", a war which saw AMD, now on the losing side, stuffing as many cores as was humanly possible into an ageing architecture, resulting in this virtually meaningless overkill today.
Intel, meanwhile, saw no reason to give the enthusiast more than six cores and the consumer more than four; even the server motherboards of today don't get more than 16 physical cores from them. And herein lies the answer - CPU cores simply outstripped the ability of reasonably-sized coding houses to take advantage of them, why trouble ourselves to code for more than 4 cores when the user is probably going to use the 4th core to surf the net while our program runs anyway? Why take on such a gargantuan task within a 1 year release cycle? In the aftermath of all this, we enter Core ix generation (I did my best to delineate the x from the i!), and TDP now becomes the battleground with Ivy Bridge i7s at 77w from Sandy Bridge at 95w (Haswell moves some of the voltage regulation circuitry on-chip, hence the 84w TDP).
The one customer base that stands no chance of relinquishing these gems, however, is computing-based research; Organizations that not only stuff as much CPU and RAM as they can into a case only 44.5mm high, but daisy chain hundreds of the damn things and pack them into a practically refridgerated room that reaches 30 Degrees C in spite of that, all to simulate a mouse brain in real time. A fascinating phenomenon in relation to this is that for 8GB of RAM, error rates due to cosmic radiation is around once per 9 years non-ECC, 45 years for ECC memory; remember that as memory increases, the error rate increases with half as many years between errors for double the memory - the upshot is that based on these statistics, even WITH ECC, a server farm totalling 400TB of memory will have a single-bit error every 7 hours and 53 minutes (1 hour and 35 minutes without).
When PCs have arguably reached the point where CPUs are considered must-have at the same level as network cards, it is easy to forget just how desirable dual CPU boards once were - they once commanded massive attention from performance freaks and made a colossal difference in performance to everything imaginable; the idea of multi-tasking as fluidly as we frequently do now on an 11" netbook was a pipe dream worthy of comparison to interstellar travel. Since server farms are compulsively upgrading to expand their research capability, there will be a steady flow of decommissioned server motherboards for us to play with, and if that isn't a dozen of fun, tell me what is!
No comments:
Post a Comment