by Anton Shilov
07/14/2006 | 05:33 AM
Sources with knowledge of the matter told X-bit labs that no technology, which is widely referred to as “reverse-Hyper-Threading” is available in Advanced Micro Devices modern central processing units (CPUs). The information about appropriate technologies is claimed to be “misunderstanding”.
One source close to AMD said that he has not heard of any technology that could boost single-threaded applications performance on the dual-core chips by combining the resources of the two execution engines. Another one, polled by The Inquirer web-site, also said that the technology does not exist. The source which informed an X-bit labs affiliate about the technology could not be reached for verification.
The first mention of the “reverse-HT” technology is dated 9th of April and belongs to X86-secret web-site. The latter reported that AMD’s processors code-named K10 will have a technology that would boost performance of single-threaded programs, particularly games, on several processing engines. The web-site described the tech as “emulation [of a] single virtual processor with two (or several) physical processors” and claimed that the capability was to be enabled in a processor code-named K10, citing an unnamed source inside AMD.
Official AMD did not comment on the news-story.
In fact, there is a huge misunderstanding with AMD’s processor code-names in the industry. Some observers and analysts believe that the K-generation code-names reflect a product, not a micro-architectural change. Following that logic, the K8 is the original single-core AMD64 processor, K9 is a dual-core chip and K10 may be either dual-core chip with new memory controller, or the quad-core processor. The logic seems to be quite incorrect, in fact, even though such vision was caused by claimes made by AMD itself.
Back in October, 2003, it was said by AMD’s Fred Weber that the K9 processor would appear in 2005. He then clarified the information in an interview with a web-site:
“What you see as we move forward is larger caches, multiple cores on a die […] Opteron, eighth generation processor, was really designed for 130nm moving to 90nm, our ninth generation is going to be designed for 90nm moving to 65nm, taking advantage of those additional transistors to make even higher performance,” said AMD’s chief technology officer back in 2003.
In 2005, however, AMD delivered its dual-core K8 micro-architecture chips. Moreover, even in early 2004 the company said that it would release dual-core processors in 2005, but did not attributed any micro-architecture number to the project. Right after the dual-core K8 release some sources indicated that this was actually the project K9 (following the claims made by Mr. Weber 1.5 years ago, as the chips were made using 90nm process technology). However, AMD itself refers improved version of its dual-core chips as K8L, which means that it does not consider the current AMD Opteron or AMD Athlon 64 X2 chips as K9, at least now.
The code-named K8L chips are to debut in 2H 2007 or 1H 2008, which, will mean that the K8 micro-architecture will be afloat for five long years. It is unclear, what will succeed the K8L, but it is obvious that the latter would live for at least a year or even two. It is logical to assume that after the project K8L, the K9 finally emerges. Given that microprocessor micro-architectures live for five-six years, the successor of the K9 – presumably, K10 – should emerge sometimes in 2014 – 2015, nearly a decade from now.
AMD knows the peculiarities of its microprocessors due to be released in a decade. Moreover, a decade after the first dual-core x86 chips appearance virtually 99% of applications should take advantage of hardware multi-threading, which means that there is no need for any reverse-HT capabilities in the K10. At the same time, even if AMD’s chips in AM2 form-factor are mistakenly referred to as “K10”, the reverse-HT capability, according to present information, is still not available.