After we posted now infamous story about Larrabee, something interesting happened. Instead of closing down, we received a lot of contacts from people involved in the project and actually have an even better look inside the project than before. It looks like old saying "the truth will set you free" is true after all. Following our story about EVGA and XFX, we got contacted by companies that weren’t selected and they opened up about Intel’s selection process and the situation got quite interesting.
For starters, we learned that recently, Intel’s again changed its attitude towards Larrabee. In the past, individuals inside the company were bullish and claimed that they’ll demolish AMD and nVidia. Few quarters passed and it looked to us that the company is taking a piss at their own roadmaps; delaying each and every ASIC into the future, followed by departure of people involved with the matter. But one of high qualities that Paul Otellini has is clearing up the matter at hand. After paying the price for his predecessors and cleaning up senior management, instituting his replacement [Sean Maloney i.e. "No Baloney", as one source told us], few quarters ago Paul ordered to clean up all the projects at hand. This decision also brought up project LRB. Transistor-heavy, humongous-die, underperforming ASIC was taken apart and each and every segment was analyzed in a bid to see why the real world part didn’t deliver what simulations promised. Again, if we would count out co-funded developments such as Intel’s Itanic – Itanium architecture [since it was co-funded by HP], this is single largest project investment in Intel’s history.
During regular quarterly meetings with large OEMs, Larrabee was recently brought out and Intel’s representatives stated that the part will surprise the competition with performance. Among other things, Intel re-iterated that the part is on track for introduction in very late first half of 2010, which would pitch the part in time for 30th anniversary of Computex Taipei 2010 [June 1-5]. After checking with our own sources, we were told that these dates might work for professional parts and accelerator cards but that mainstream application compatibility is still a long way out.
If Larrabee indeed comes to market in time for commercial cycle, AMD FireStream and Tesla C2070 are easy targets; AMD’s passive attitude as far as FireStream line-up is considered is exactly the opposite of what Intel allegedly offers, while nVidia’s long-delayed Tesla C2050 won’t arrive to market before 2Q 2010 and C2070 is delayed until third quarter 2010.
The AIB partner list is also taking shape, although it was flat-out denied by Intel’s Larrabee spokeperson – our sources claim that EVGA and XFX indeed signed with Intel for Larrabee. The list is nowhere as close to being closed. What Intel demands from AIBs is global approach to the market and ability to clear up any warranty claims, as well as carry in some cost of development. Interestingly though, we haven’t heard about ASUS and Larrabee for quite some time.
Update #1, December 3, 2009 12:54 GMT - We have posted a story disclosing Larrabee’s SGEMM performance [BLAS calculation, 4K by 4K matrix] and compared it with nVidia GT200, ATI RV770 GPUs etc.