Monday, April 30, 2007
Benjamin Franklin, Cray's XT4 is ready for start
Alas, initially without AMD K10 inside.
It will have 19,344 compute CPUs, each with 2 GB
of memory per CPU. Quad core K10 could easily boost its FP performance triple. To 300TFlops peak. Seems like a clever plan for world's leadership in supercomputing.
The system will deliver sustained performance of at least 16 trillion calculations per second—with a theoretical peak speed of more than 100 teraflop/s—when running a suite of diverse scientific applications at scale. The system will have over 400 terabytes of high performance, parallel disk space. Initial installation began in 2006 and the full system is scheduled to go into production service in late summer 2007.
Then, 45nm upgrade in 2008 follows. That 4 core processor will consume no more than 95W. Thus 3 times faster, but no more power hungry. That path is extremely important for every supercomputer center. Upgrade processors and memory, and don't change anything else to triple its performance.
This year is the 300th anniversary of Franklin's birth. Close to 200 those cabinets would be needed in the end for its max performance configuration.
Friday, April 27, 2007
No more Athlon: AMD renames K10 to P H E N O M
Here is Google's translation:
Quote: › According to the PClab.pl site, next processors AMD general public based on new K10 architecture should be associated the official name of Phenom, which could be derived poured of it Phenom X2, Phenom X4 or Phenom FX. AMD could thus leave side the powerful Athlon mark. Remain to see now if the firm of Sunnyvale will be able to propose and to popularize this new mark on the segment of the performance with, in particular, a marketing plan with the height of its hopes of sale. For the moment, few details slipped out, as for the performances of these new processors, AMD simply communicated some figures (see AMD: Opteron 3 GHz and performances of Barcelona). It will be thus necessary still await the test officials of press to know if the performance of these new come be really phenomenal .
D'après le site PClab.pl, les prochains processeurs AMD grand public basés sur la nouvelle architecture K10 devraient être associés au nom officiel de Phenom, lequel pourrait être dérivé en versions Phenom X2, Phenom X4 ou encore Phenom FX. AMD pourrait donc laisser de côté la puissante marque Athlon. Reste à voir maintenant si la firme de Sunnyvale sera en mesure de mettre en avant et de populariser cette nouvelle marque sur le segment de la performance avec, notamment, un plan marketing à la hauteur de ses espérances de vente.
Pour le moment, peu de détails ont filtré quant aux performances de ces nouveaux processeurs, AMD a simplement communiqué quelques chiffres (voir AMD : Opteron 3 GHz et performances du Barcelona). Il faudra donc encore attendre les tests officiels de la presse pour savoir si les performances de ces nouveaux venus sont vraiment phenom...énales (oui, je vous l'accorde, elle était très facile celle-la !).
Thursday, April 26, 2007
The Inquirer changes name to The InTELquirer
After The Inquirer recently being sold to an unnamed owner, now turns that anonymous owner is pretty well known under the name INTEL. So, once distinquished journo became INTEL's offspring. Well, not the tragedy, however the real problem is that Intel is the most lying company on this planet. So the Inquirer incidentally became too. Nor a single word on AMD Barcelona performance? That is, beside being ridiculous, quite idiotic. Times when you could handle monopoly that way are gone. Today, the most known brand is Google, not the Microsoft, nor any other monopolist like Intel is. Mike ,you got amnesia, right? And wrong leading direction. On the Intel side, that seems like a pretty cheap way to boost lagging Merom performance. Compared (of course) to AMD Barcelona.
Monday, April 23, 2007
AMD Barcelona beats Xeon in SPECcpu2006, FP 50% , Integer 20%
However, please note at the same frequency remark. If correct, that means Intel is still better, right? Why is better? Because it is in 65nm and new architecture. Well, new K10 core Barcelona is too made in 65nm. Beside, AMD's advanced strained SSOI manufacturing process is better than 30 years old Intel's bulk silicon. Until 45nm ramp arrive in 1Q2008, or even than. And for the first time ever, AMD's weakest point second source fab, is luckily solved by the Barcelona launch time. :)
If even at this moment Opteron is better than Xeon in SPEC tests, let estimate Barcelona's performance improvements. Approximatively 50% -15% equals 35% max in FP.In SPEC integer 20% -2,5% peaks 17,5% . Quite decent K10's purely architectural improvements, IMHO.
Thus, Barcelona could lower its clock at 2,5 Ghz and still beat Intel's 3Ghz top Xeon. In all aspects: performance, reliability, power consumption, prices.
However, AMD fans think (not without substantiated benchies) that Intel should run much higher than 3 GHZ to match the 2.5 GHz K10 part. OMG
So, that is the real reason why Intel so much talks about Penryn. Much more than AMD on Barcelona, just around the corner. Because Intel has nothing else ready for market at this moment.
Yeah, What if the AMD Opteron™ processor never existed?
May is the month when the first Barcelona benchmarks should be released. In wild, of course. :)
Intel's response is given here. And here. Even now.Congrats.
Friday, April 20, 2007
AMD Barcelona started its 4 lane race !
Computex? What a breaking news. :) You want someting really new? AMD Bulldozer is a low power dual die 8 core MCM. Perhaps even in the third quarter this year! The source of AMD optimism? Like 2,8Ghz clock at the start with rev B dual core engineering sample..
K10 at 2.5 GHz beats Intel quad at 3.0+
The new K10 based processors will really make Intel run for its money, at least till the end of the year. Just as Intel stole K8's mojo with Merom aka Core 2 Duo and quad, AMD is ready to finally fight back. It took them almost a year to prepare for the big clash.
Barcelona / Agena FX will beat Intel Core 2 Quad at 3.0 GHz+. Intel knows it and it is scared. Intel also knows that it cannot go over 3.0 GHz with 65 nanometre, at least not at 120 W TDP. Some reports already claim that QX6850 with FSB 1333 and 3 GHZ clock is already pushing the envelope as intel's 120W TDP is actually 150W+ at least on AMD's TDP scale. TDP stands for Thermal Design Power.
Intel needs to go over 3.0 GHZ to 3.2 or more to match Barcelona / Agena parts at 2.5 GHz. Intel knows that it needs to go to 3.2+ GHz and it needs the 45 nm Yorkfield core to match Agena 2.5 GHz.
Finally AMD should be able to make 2.7 if not even more with its 65 nm process. Intel has a good fighting chance with next the generation Nehalem core in the second half of 2008.
Tuesday, April 17, 2007
Intel Development Forum: We have one mobile use more than AMD !
Finally, a revealing truth on bad hardware that protects the head of your investment.
Here is an application for processor's 100% dirty use.
Wednesday, April 11, 2007
AMD Turion X2, TL-66 2,3Ghz, DDR-2 800 Mhz , 35W
OK, X2 balls are obviously inside this "new" mobile chip kid on the block, but are they really all in the performance sense that might be expected by authorized Intel inclined hardware benchmarkers?
Seems nothing exciting in this case, except of course, low price tag for your street pleasure. AMD, save our souls with K10 ! Please.
Sun's Rock chip for 256TB of RAM address space
Total of 2395 pins per chip. It is actually a sea of pins. However, Sun's architecture design is different from AMD or Intel's. Cores are scheduled automatically, quite contrary to X86 cores. Why? You know, you can't learn an old dog the new tricks. :)
Monday, April 02, 2007
It's no wonder we need to upgrade our computers on a regular basis. Not only are chip companies regularly releasing ever-faster microprocessors, but new research has revealed that modern CPUs actually lose megahertz over time.
This startling conclusion follows a five-year research programme carried out by the Illyria University's Information Technology department.
"Five years ago, we activated a dozen new, freshly-purchased systems," said Computer Science Professor Asteio Artikolos. "Since then we have measured, on average, a 10-15 per cent reduction in the machines' clock speeds over that time. It's as if each machine's stock of megahertz were somehow leaking away."
The sad thing is that we all actually see this due to fact that all our application are mainly single threaded and only higher clock speed could give us illusion of higher productivity. (TM bad hardware week). But, I admit April fools idea is an original one. And the only one able to solve clock rise stall. :)
The top 100 April Fool's day hoaxes.
Here is my proposal for top 100 : Intel doesn't cheat and copy the others anymore !