The R8000 is a microprocessor chipset developed by MIPS Technologies, Inc. (MTI), Toshiba , and Weitek . It was the first implementation of the MIPS IV instruction set architecture . The R8000 is also known as the TFP , for Tremendous Floating-Point , its name during development.
60-510: Development of the R8000 started in the early 1990s at Silicon Graphics , Inc. (SGI). The R8000 was specifically designed to provide the performance of circa 1990s supercomputers with a microprocessor instead of a central processing unit (CPU) built from many discrete components such as gate arrays . At the time, the performance of traditional supercomputers was not advancing as rapidly as reduced instruction set computer (RISC) microprocessors. It
120-525: A Re-IPO , and fully divested itself in 2000. In the late 1990s, when much of the industry expected the Itanium to replace both CISC and RISC architectures in non-embedded computers, SGI announced their intent to phase out MIPS in their systems. Development of new MIPS microprocessors stopped, and the existing R12000 design was extended multiple times until 2003 to provide existing customers more time to migrate to Itanium. In August 2006, SGI announced
180-427: A cache miss , the data must be loaded from the streaming cache with an eight-cycle penalty. The cache is virtually indexed, physically tagged , direct mapped , has a 32-byte line size and uses a write-through with allocate protocol. If the loads hit in the data cache, the result is written to the integer register file in stage five. The R8010 executed floating-point instructions provided by an instruction queue on
240-560: A 2400 to a 2400T. The 2500 and 2500T had a larger chassis, a standard 6' 19" EIA rack with space at the bottom for two SMD disk drives weighing approximately 68 kg each. The non-Turbo models used the Multibus for the CPU to communicate with the floating point accelerator, while the Turbos added a ribbon cable dedicated for this. 60 Hz monitors were used for the 2000 series. The height of
300-418: A 32-byte line size. Instruction decoding and register reads occur during stage two, and branch instructions are resolved as well, leading to a one-cycle branch mispredict penalty. Load and store instructions begin execution in stage three, and integer instructions in stage four. Integer execution was delayed until stage four so that integer instructions which use the result of a load as an operand may be issued in
360-431: A broad range of MIPS-based workstations and servers during the 1990s, running SGI's version of UNIX System V, now called IRIX . These included the massive Onyx visualization systems, the size of refrigerators and capable of supporting up to 64 processors while managing up to three streams of high resolution, fully realized 3D graphics. In October 1991, MIPS announced the first commercially available 64-bit microprocessor,
420-524: A computer graphics process that turns text and images into pixels to be displayed on screens. Affected devices include Apple iPhone, HTC EVO4G, LG Thrill, Research in Motion Torch, Samsung Galaxy S and Galaxy S II, and Sony Xperia Play smartphones. SGI's first generation products, starting with the IRIS (Integrated Raster Imaging System) 1000 series of high-performance graphics terminals, were based on
480-723: A deal with Nintendo to develop the Reality Coprocessor (RCP) GPU used in the Nintendo 64 (N64) video game console. The deal was signed in early 1993, and it was later made public in August of that year. The console itself was later released in 1996. The RCP was developed by SGI's Nintendo Operations department, led by engineer Dr. Wei Yen . In 1997, twenty SGI employees, led by Yen, left SGI and founded ArtX (later acquired by ATI Technologies in 2000). In 1998, SGI relinquished some ownership of MIPS Technologies, Inc in
540-516: A delisting notification from NASDAQ , as its market value had been below the minimum $ 35 million requirement for 10 consecutive trading days, and also did not meet NASDAQ's alternative requirements of a minimum stockholders' equity of $ 2.5 million or annual net income from continuing operations of $ 500,000 or more. On April 1, 2009, SGI filed for Chapter 11 again, and announced that it would sell substantially all of its assets to Rackable Systems for $ 25 million. The sale, ultimately for $ 42.5 million,
600-503: A floating-point throughput of 300 million instructions per second, a SPECfp92 rating of 310, and a more modest SPECint92 rating of 108. The chip set consisted of the R8000 microprocessor, the R8010 floating-point unit, two Tag RAMs, and the streaming cache. The R8000 is superscalar , capable of issuing up to four instructions per cycle, and executes instructions in program order. It has a five-stage integer pipeline . The R8000 controlled
660-444: A group of seven graduate students and research staff from Stanford University: Kurt Akeley , David J. Brown , Tom Davis , Rocky Rhodes, Marc Hannah , Herb Kuta , and Mark Grossman ; along with Abbey Silverstone and a few others. Ed McCracken was CEO of Silicon Graphics from 1984 to 1997. During those years, SGI grew from annual revenues of $ 5.4 million to $ 3.7 billion. The addition of 3D graphic capabilities to PCs , and
SECTION 10
#1732780416489720-602: A load queue, a store queue, and two identical floating-point units. All instructions except for divide and square-root are pipelined. The R8010 implements an iterative division and square-root algorithm that uses the multiplier for a key part, requiring the pipeline to be stalled the unit for the duration of the operation. Arithmetic instructions except for compares have a four-cycle latency. Single and double precision divides have latencies of 14 and 20 cycles, respectively; and single and double precision square-roots have latencies of 14 and 23 cycles, respectively. The streaming cache
780-621: A new line of credit. SGI announced it was postponing its scheduled annual December stockholders meeting until March 2006. It proposed a reverse stock split to deal with the de-listing from the New York Stock Exchange. In January 2006, SGI hired Dennis McKenna as its new CEO and chairman of the board of directors. Mr. McKenna succeeded Robert Bishop , who remained vice chairman of the board of directors. On May 8, 2006, SGI announced that it had filed for Chapter 11 bankruptcy protection for itself and U.S. subsidiaries as part of
840-427: A number of initiatives were taken which are considered to have accelerated the corporate decline. One such initiative was trying to sell workstations running Windows NT called Visual Workstations in addition to workstations running IRIX , the company's version of UNIX . This put the company in even more direct competition with the likes of Dell, making it more difficult to justify a price premium. The product line
900-716: A plan to reduce debt by $ 250 million. Two days later, the U.S. Bankruptcy Court approved its first day motions and its use of a $ 70 million financing facility provided by a group of its bondholders. Foreign subsidiaries were unaffected. On September 6, 2006, SGI announced the end of development for the MIPS/IRIX line and the IRIX operating system. Production would end on December 29 and the last orders would be fulfilled by March 2007. Support for these products would end after December 2013. SGI emerged from bankruptcy protection on October 17, 2006. Its stock symbol at that point, SGID ,
960-464: A tilt/swivel base; 19" 30 Hz interlaced and a 15" 60 Hz non-interlaced (with tilt/swivel base) were also available. The IRIS 3130 and its smaller siblings were impressive for the time, being complete UNIX workstations. The 3130 was powerful enough to support a complete 3D animation and rendering package without mainframe support. With large capacity hard drives by standards of the day (two 300 MB drives), streaming tape and Ethernet, it could be
1020-490: Is an external 1 to 16 MB cache that serves as the R8000's L2 unified cache and the R8010's L1 data cache. It operates at the same clock rate as the R8000 and is built from commodity synchronous static RAMs . This scheme was used to attain sustained floating point performance, which requires frequent access to data. A small low-latency primary cache would not contain enough data and frequently miss, necessitating long latency refiles that reduce performance. The streaming cache
1080-457: Is two-way interleaved . It has two independent banks , each containing data from even or odd addresses. It can therefore perform two reads, two writes, or a read and a write every cycle, provided that the two accesses are to separate banks. Each bank is accessed via two 64-bit unidirectional buses, one for reads, and the other for writes. This scheme was used to avoid bus turnover , which is required by bidirectional buses. By avoiding bus turnover,
1140-599: The MIPS architecture and able to run Windows NT and SCO UNIX . The group produced the Advanced RISC Computing (ARC) specification, but began to unravel little more than a year after its formation. For eight consecutive years (1995–2002), all films nominated for an Academy Award for Distinguished Achievement in Visual Effects were created on Silicon Graphics computer systems. The technology
1200-554: The Motorola 68000 family of microprocessors . The later IRIS 2000 and 3000 models developed into full UNIX workstations . The first entries in the 1000 series (models 1000 and 1200, introduced in 1984) were graphics terminals, peripherals to be connected to a general-purpose computer such as a Digital Equipment Corporation VAX , to provide graphical raster display abilities. They used 8 MHz Motorola 68000 CPUs with 768 kB of RAM and had no disk drives . They booted over
1260-467: The Multibus standard. Later 1000-series machines, the 1400 and 1500, ran at 10 MHz and had 1.5 MB of RAM. The 1400 had a 72 MB ST-506 disk drive, while the 1500 had a 474 MB SMD-based disk drive with a Xylogics 450 disk controller. They may have used the PM2 CPU and PM2M1 RAM board from the 2000 series. The usual monitor for the 1000 series ran at 30 Hz interlaced . Six beta-test units of
SECTION 20
#17327804164891320-515: The New York Stock Exchange because its common stock had fallen below the minimum share price for listing on the exchange. SGI's market capitalization dwindled from a peak of over seven billion dollars in 1995 to just $ 120 million at the time of delisting. In February 2006, SGI noted that it could run out of cash by the end of the year. In mid-2005, SGI hired Alix Partners to advise it on returning to profitability and received
1380-503: The R4000 . SGI used the R4000 in its Crimson workstation. IRIX 6.2 was the first fully 64-bit IRIX release, including 64-bit pointers. To secure the supply of future generations of MIPS microprocessors (the 64-bit R4000 ), SGI acquired the company in 1992 for $ 333 million and renamed it as MIPS Technologies Inc., a wholly owned subsidiary of SGI. In 1993, Silicon Graphics (SGI) signed
1440-534: The 1400 workstation were produced, and the first production unit (SGI's first commercial computer) was shipped to Carnegie-Mellon University's Electronic Imaging Laboratory in 1984. SGI rapidly developed its machines into workstations with its second product line — the IRIS 2000 series, first released in August 1985. SGI began using the UNIX System V operating system . There were five models in two product ranges,
1500-676: The 2000/2200/2300/2400/2500 range which used 68010 CPUs (the PM2 CPU module), and the later "Turbo" systems, the 2300T, 2400T and 2500T, which had 68020s (the IP2 CPU module). All used the Excelan EXOS/201 Ethernet card, the same graphics hardware (GF2 Frame Buffer, UC4 Update Controller, DC4 Display Controller, BP3 Bitplane). Their main differences were the CPU, RAM, and Weitek Floating Point Accelerator boards, disk controllers and disk drives (both ST-506 and SMD were available). These could be upgraded, for example from
1560-768: The Court entered an order granting the parties' agreed motion for dismissal and final judgment. In November 2011, GPHI filed another patent infringement lawsuit against Apple Inc. in Delaware involving more patents than their original patent infringement case against Apple last November, for alleged violation of U.S. patents 6,650,327 ('327), U.S. patent 6,816,145 ('145) and U.S. patent 5,717,881 ('881). In 2012, GPHI filed lawsuit against Apple, Sony, HTC Corp, LG Electronics Inc. and Samsung Electronics Co., Research in Motion Ltd. for allegedly violating patent relating to
1620-595: The District Court entered an order that permits AMD to pursue its invalidity affirmative defense at trial and does not permit SGI to accuse AMD's Radeon R700 series of graphics products of infringement in this case. On April 18, 2011, GPHI and AMD had entered into a confidential Settlement and License Agreement that resolved this litigation matter for an immaterial amount and that provides immunity under all GPHI patents for alleged infringement by AMD products, including components, software and designs. On April 26, 2011,
1680-411: The R8000. The queue decoupled the floating-point pipeline from the integer pipeline, implementing a limited form of out-of-order execution by allowing floating-point instructions to execute when possible after or before the integer instructions from the same group are issued. The pipelines were decoupled to help mitigate some of the streaming cache latency. It contained the floating-point register file,
1740-526: The SGI Pleiades , a TOP500 supercomputer at NASA Ames Research Center, in its portfolio. During Silicon Graphics Inc.'s second bankruptcy phase, it was renamed to Graphics Properties Holdings, Inc.(GPHI) in June 2009. In 2010, GPHI announced it had won a significant favorable ruling in its litigation with ATI Technologies and AMD in June 2010, following the patent lawsuit originally filed during
1800-589: The Silicon Graphics, Inc. era. Following the 2008 appeal by ATI over the validity of U.S. patent 6,650,327 ('327) and Silicon Graphics Inc's voluntary dismissal of the U.S. patent 6,885,376 ('376) patent from the lawsuit, the Federal Circuit upheld the jury verdict on the validity of GPHI's U.S. Patent No. 6,650,327, and furthermore found that AMD had lost its right to challenge patent validity in future proceedings. On January 31, 2011,
1860-503: The ability of clusters of Linux - and BSD -based PCs to take on many of the tasks of larger SGI servers, ate into SGI's core markets. The porting of Maya to Linux , Mac OS and Microsoft Windows further eroded the low end of SGI's product line. In response to challenges faced in the marketplace and a falling share price Ed McCracken was fired and SGI brought in Richard Belluzzo to replace him. Under Belluzzo's leadership
R8000 - Misplaced Pages Continue
1920-472: The base(register) + index(register) address style added in the MIPS IV ISA. The R8000 issues at most one integer store per cycle, and one final read port delivers the integer store data. Two register file write ports are used to write results from the two integer functional units. The R8000 issues two integer loads per cycle, and the other two write ports are used to write the results of integer loads to
1980-609: The bold move of allowing the resulting OpenGL API to be cheaply licensed by SGI's competitors, and set up an industry-wide consortium to maintain the OpenGL standard (the OpenGL Architecture Review Board). This meant that for the first time, fast, efficient, cross-platform graphics programs could be written. For over 20 years – until the introduction of the Vulkan API – OpenGL remained
2040-518: The cache can be read from in one cycle and then written to in the next cycle without an intervening cycle for turnover, resulting in improved performance. The streaming cache's tags are contained on two Tag RAM chips, one for each bank. Both chips contain identical data. Each chip contains 1.189 Mbit of cache tags implemented by four-transistor SRAM cells. The chips are implemented in a 0.7 μm BiCMOS process with two levels of polysilicon and two levels of aluminium interconnect . BiCMOS circuitry
2100-569: The centerpiece of an animation operation. The line was formally discontinued in November 1989, with about 3500 systems shipped of all 2000 and 3000 models combined. With the introduction of the IRIS 4D series, SGI switched to MIPS microprocessors. These machines were more powerful and came with powerful on-board floating-point capability. As 3D graphics became more popular in television and film during this time, these systems were responsible for establishing much of SGI's reputation. SGI produced
2160-415: The chip set and executed integer instructions. It contained the integer execution units, integer register file , primary caches and hardware for instruction fetch, branch prediction the translation lookaside buffers (TLBs). In stage one, four instructions are fetched from the instruction cache. The instruction cache is 16 kB large, direct-mapped , virtually tagged and virtually indexed , and has
2220-410: The cycle after the load. Results are written to the integer register file in stage five. The integer register file has nine read ports and four write ports. Four read ports supply operands to the two integer execution units (the branch unit was considered part of an integer unit). Another four read ports supply operands to the two address generators. Four ports are needed, rather than two, because of
2280-505: The end of production for MIPS/IRIX systems, and by the end of the year MIPS/IRIX products were no longer generally available from SGI. Until the second generation Onyx Reality Engine machines, SGI offered access to its high performance 3D graphics subsystems through a proprietary API known as IRIS Graphics Library ( IRIS GL ). As more features were added over the years, IRIS GL became harder to maintain and more cumbersome to use. In 1992, SGI decided to clean up and reform IRIS GL and made
2340-403: The machines using Motorola CPUs was reached with the IRIS 3000 series (models 3010/3020/3030 and 3110/3115/3120/3130, the 30s both being full-size rack machines). They used the same graphics subsystem and Ethernet as the 2000s, but could also use up to 12 "geometry engines", the first widespread use of hardware graphics accelerators. The standard monitor was a 19" 60 Hz non-interlaced unit with
2400-593: The market. In the mid-2000s the company repositioned itself as a supercomputer vendor, a move that also failed. On April 1, 2009, SGI filed for Chapter 11 bankruptcy protection and announced that it would sell substantially all of its assets to Rackable Systems, a deal finalized on May 11, 2009, with Rackable assuming the name Silicon Graphics International . The remnants of Silicon Graphics, Inc. became Graphics Properties Holdings, Inc. James H. Clark left his position as an electrical engineering associate professor at Stanford University to found SGI in 1982 along with
2460-514: The mid to late-1990s, the rapidly improving performance of commodity Wintel machines began to erode SGI's stronghold in the 3D market. The porting of Maya to other platforms was a major event in this process. SGI made several attempts to address this, including a disastrous move from their existing MIPS platforms to the Intel Itanium , as well as introducing their own Linux -based Intel IA-32 based workstations and servers that failed in
R8000 - Misplaced Pages Continue
2520-470: The multiply-divide unit, which is not pipelined. As a result, the latency for a multiply instruction is four cycles for 32-bit operands and six cycles for 64-bit. The latency for a divide instruction depends on the number of significant digits in the result and thus it varies from 21 to 73 cycles. Loads and stores begin execution in stage three. The R8000 has two address generation units (AGUs) that calculate virtual address for loads and stores. In stage four,
2580-723: The network (via an Excelan EXOS/101 Ethernet card) from their controlling computer. They used the "PM1" CPU board, which was a variant of the board that was used in Stanford University 's SUN workstation and later in the Sun-1 workstation from Sun Microsystems . The graphics system was composed of the GF1 frame buffer , the UC3 "Update Controller", DC3 "Display Controller", and the BP2 bitplane. The 1000-series machines were designed around
2640-675: The newer Amphitheatre Parkway headquarters was sold to Google (which had already subleased and moved into the facility in 2003). Both of these locations were award-winning designs by Studios Architecture . In April 2008, SGI re-entered the visualization market with the SGI Virtu range of visualization servers and workstations, which were re-badged systems from BOXX Technologies based on Intel Xeon or AMD Opteron processors and Nvidia Quadro graphics chipsets, running Red Hat Enterprise Linux , SUSE Linux Enterprise Server or Windows Compute Cluster Server . In December 2008, SGI received
2700-516: The only real-time 3D graphics standard to be portable across a variety of operating systems. SGI was part of the Advanced Computing Environment initiative, formed in the early 1990s with 20 other companies, including Compaq , Digital Equipment Corporation , MIPS Computer Systems , Groupe Bull , Siemens , NEC , NeTpower , Microsoft and Santa Cruz Operation . Its intent was to introduce workstations based on
2760-420: The register file. The level 1 data cache was organized as two redundant arrays, each of which had one read port and one write port. Integer stores were written to both arrays. Two loads could be processed in parallel, one on each array. Integer functional units consisted of two integer units, a shift unit, a multiply-divide unit, and two address generator units. Multiply and divide instructions are executed in
2820-572: The two chips contained 3.43 million transistors. Both were fabricated by Toshiba in their VHMOSIII process, a 0.7 μm, triple-layer metal complementary metal–oxide–semiconductor (CMOS) process. Both are packaged in 591-pin ceramic pin grid array (CPGA) packages. Both chips used a 3.3 V power supply, and the R8000 dissipated 13 W at 75 MHz. Silicon Graphics Silicon Graphics, Inc. (stylized as SiliconGraphics before 1999, later rebranded SGI , historically known as Silicon Graphics Computer Systems or SGCS )
2880-686: The usual way of implementing set-associative caches. Access to the streaming cache is pipelined to mitigate some of the latency. The pipeline has five stages: in stage one, addresses are sent to the Tag RAMs, which are accessed in stage two. Stage three is for the signals from the Tag RAMs to propagate to the SSRAMs. In stage four, the SSRAMs are accessed and data is returned to the R8000 or R8010 in stage five. The R8000 contained 2.6 million transistors and measured 17.34 mm by 17.30 mm (299.98 mm). The R8010 contained 830,000 transistors. In total,
2940-407: The virtual addresses are translated to physical addresses by a dual-ported TLB that contains 384 entries and is three-way set associative. The 16 kB data cache is accessed in the same cycle. It is dual-ported, and is accessed via two 64-bit buses. It can service two loads or one load and one store per cycle. The cache is not protected by parity or by error correcting code (ECC). In the event of
3000-563: Was acquiring MIPS Computer Systems, which became a subsidiary of SGI called MIPS Technologies, Inc. (MTI) in mid-1992. Development of the R8000 was transferred to MTI, where it continued. The R8000 was expected to be introduced in 1993, but it was delayed until mid-1994. The first R8000, a 75 MHz part, was introduced on 7 June 1994. It was priced at US$ 2,500 at the time. In mid-1995, a 90 MHz part appeared in systems from SGI. The R8000's high cost and narrow market (technical and scientific computing) restricted its market share, and although it
3060-640: Was also used in commercials for a host of companies. An SGI Crimson system with the fsn three-dimensional file system navigator appeared in the 1993 movie Jurassic Park . In the movie Twister , protagonists can be seen using an SGI laptop computer; however, the unit shown was not an actual working computer, but rather a fake laptop shell built around an SGI Corona LCD flat screen display. The 1995 film Congo also features an SGI laptop computer being used by Dr. Ross ( Laura Linney ) to communicate via satellite to TraviCom HQ. Address generation unit Too Many Requests If you report this error to
SECTION 50
#17327804164893120-666: Was an American high-performance computing manufacturer, producing computer hardware and software . Founded in Mountain View, California , in November 1981 by James Clark , its initial market was 3D graphics computer workstations , but its products, strategies and market positions developed significantly over time. Early systems were based on the Geometry Engine that Clark and Marc Hannah had developed at Stanford University , and were derived from Clark's broader background in computer graphics . The Geometry Engine
3180-632: Was canceled, and new stock was issued on the NASDAQ exchange under the symbol SGIC . This new stock was distributed to the company's creditors, and the SGID common stockholders were left with worthless shares. At the end of that year, the company moved its headquarters from Mountain View to Sunnyvale . Its earlier North Shoreline headquarters is now occupied by the Computer History Museum ;
3240-565: Was finalized on May 11, 2009; at the same time, Rackable announced their adoption of "Silicon Graphics International" as their global name and brand. The Bankruptcy Court scheduled continuing proceedings and hearings for June 3 and 24, 2009, and July 22, 2009. After the Rackable acquisition, Vizworld magazine published a series of six articles that chronicle the downfall of SGI. Hewlett Packard Enterprise acquired Silicon Graphics International in November 2016, which allowed HPE to place
3300-518: Was popular in its intended market, it was largely replaced with the cheaper and generally better performing R10000 introduced January 1996. Users of the R8000 were SGI, who used it in their Power Indigo2 workstation , Power Challenge server, Power ChallengeArray cluster and Power Onyx visualization system. In the November 1994 TOP500 list, 50 systems out of 500 used the R8000. The highest ranked R8000-based systems were four Power Challenges at positions 154 to 157. Each had 18 R8000s. SGI claimed
3360-533: Was predicted that RISC microprocessors would eventually match the performance of more expensive and larger supercomputers at a fraction of the cost and size, making computers with this level of performance more accessible and enabling deskside workstations and servers to replace supercomputers in many situations. First details of the R8000 emerged in April 1992 in an announcement by MIPS Computer Systems detailing future MIPS microprocessors. In March 1992, SGI announced it
3420-486: Was the first very-large-scale integration (VLSI) implementation of a geometry pipeline , specialized hardware that accelerated the "inner-loop" geometric computations needed to display three-dimensional images. For much of its history, the company focused on 3D imaging and was a major supplier of both hardware and software in this market. Silicon Graphics reincorporated as a Delaware corporation in January 1990. Through
3480-458: Was unchanged. At the same time, SGI announced a new logo consisting of only the letters "sgi" in a proprietary font called "SGI", created by branding and design consulting firm Landor Associates , in collaboration with designer Joe Stitzlein. SGI continued to use the "Silicon Graphics" name for its workstation product line, and later re-adopted the cube logo for some workstation models. In November 2005, SGI announced that it had been delisted from
3540-620: Was unsuccessful and abandoned a few years later. SGI's premature announcement of its migration from MIPS to Itanium and its abortive ventures into IA-32 architecture systems (the Visual Workstation line, the ex-Intergraph Zx10 range and the SGI 1000-series Linux servers) damaged SGI's credibility in the market. In 1999, in an attempt to clarify their current market position as more than a graphics company, Silicon Graphics Inc. changed its corporate identity to "SGI", although its legal name
3600-541: Was used in the decoders and combined sense amplifier and comparator portions of the chip to reduce cycle time. Each Tag RAM is 14.8 mm by 14.8 mm large, packaged in a 155-pin CPGA, and dissipates 3 W at 75 MHz. In addition to providing the cache tags, the Tag RAMs are responsible for the streaming cache being four-way set associative. To avoid high a pin count, the cache tags are four-way set associative and logic selects which set to access after lookup instead of
#488511