Seminar reportonITANIUM:The 64-bit microprocessor from intelITANIUM
History:
In 1994, Hewlett-Packard and Intel Corporation agreed to jointly design EPIC (Explicitly Parallel Instruction Computing), a post-RISC and post-IA-32 technology. Using EPIC concepts, HP and Intel® then jointly defined Itanium’s 64-bit Itanium Processor Family (IPF) architecture, the basis of Intel’s future, high-performance microprocessor family a broad range of technical and commercial applications at 1.0GHz. The starting place for the team was a comprehensive understanding of the capabilities of the .18um bulk technology transistors and interconnects with a view toward exploiting these capabilities to the fullest. The Itanium processor began shipping in end-user pilot systems in late 2000. Intel intends to follow Itanium with additional processors in the Itanium family: McKinley, Madison and Deerfield in late 2002.
Need For Itanium:
Internet commerce and large database applications are dealing with ever- increasing quantities of data, and demands placed on both server and workstation resources are increasing correspondingly. One demand is for more memory than the 4 GB provided by today’s 32-bit computer architectures. Itanium’s ability to address a flat 64-bit memory address space in the millions of gigabytes has been the focus of attention. Beyond very large memory (VLM) support, however, other traits, including a new Explicitly Parallel Instruction Computing (EPIC) design philosophy that will handle parallel processing differently than previous architectures, speculation, predication, large register files, a register stack and advanced branch architecture. IA-64 also provides an enhanced system architecture supporting fast interrupt response and a flexible, large virtual address mode. The 64-bit addressing enabled by the Intel Itanium architecture will help overcome the scalability barriers and awkward, maintenance-intensive partitioning directory schemes of current directory services on 32-bit platforms. , Intel has been assiduous in providing backward compatibility with 32-bit binaries (IA-32), from the x86 families. The Itanium has a complex, bleeding edge, forward looking processor family that holds promise for huge gains in processing power
The How’s & Wows of Itanium:
Itanium defined a new architecture that provides a unique combination of innovative features, which overcomes the performance limitations of traditional architectures. The IA-64 architecture is based on innovative techniques/features such as Explicit parallelism, Parallelism, and Predication and Speculation, resulting in superior Instruction Level Parallelism (ILP) and increased instructions per cycle (IPC) to address the current and future requirements of these demanding Internet, high end server, and workstation applications. In addition, the IA-64 architecture provides headroom and scalability for continued future growth.
We take a look at all the features that make it possible, that are the bricks and mortar of the Itanium; and then we shall see the architecture that emerges.
àEPIC:
EPIC stands for Explicitly Parallel Instruction Computing–a new design philosophy going beyond the RISC and CISC processors that are available today. EPIC technology enables greater instruction level parallelism than previous processor architectures, supporting higher levels of performance in targeted application segments. The Itanium architecture is based on EPIC technology. EPIC is based on a unique combination of innovative features such as predication, speculation and explicit parallelism enabling world-class performance for the high-end enterprise class of computing
The EPIC architecture uses complex instruction wording that, in addition to the basic instruction, contains information on how to run the instruction in parallel. EPIC instructions are put together by the compiler into a threesome called a "bundle." Bundles instructions are sent to the CPU together other instructions. The "bundles," or their parts, are put together in an "instruction group" with other instructions. The IA-64 utilizes 128 bit bundles, organized as a template plus three IA-64 instructions. The template contains information provided by the compiler to the processor. The template contains the following information:
1. Which instructions can be executed in parallel, and which instructions must be executed serially.
2. Information relating to parallelism in respect to it's neighboring bundles.
3. Maps instruction slots to execution types
Instructions in a bundle do not affect each other with the data they are working on, so they can run together without getting in each other's way. There is no limit to the size of an instruction group, and an instruction group can begin or end in the middle of a bundle. The instructions are actually bundled and grouped together when software is compiled. This simplifies the process of running multiple instructions at once on an Itanium CPU, allowing it to make greater use of multiple execution units without having to rely on complex on-die logic to determine what operations can run in parallel. The Itanium will still use on-die logic to improve upon instruction level parallelism, but EPIC instructions, at the minimum, provide a parallel blueprint for the Itanium processor. Of course, for this reason, compiler technology and programming algorithms will have a massive impact on Itanium performance. The compiler adds branch hints, register stack and rotation, data and control speculation, and memory hints into EPIC instructions. Thus, EPIC makes processing faster, much faster.
EPIC Pipelining: The parallel execution core of EPIC can have upto 10 pipelined stages. The first-generation Itanium processor is able to issue up to six EPIC instructions in parallel every clock cycle. The next generation Itanium might yield 20 instructions per cycle, though not consistently, but the potential is there and proper coding and compiling should yield efficient usage of the CPU.
àPredication:
Predication is a compiling technique used in the Itanium that optimizes or removes branching code by working it so that much of the code runs in
parallel. By designing software and compilers to rework branches into parallel code with fewer or no branches, and by running this code on a "wide" processor that can process this code in parallel, which the Itanium is intended to be able to do, the number of cycles it takes to complete a task drops.
What predication does is minimize the time it takes to run if-then-else situation and uses processor width to run both the 'then' and 'else' in parallel. When the 'if' branch is determined, the incorrect branch's result is discarded. By removing branches and making code more parallel, predication reduces the number of cycles it takes to complete a task while making better use of a wide processor. Also, there are less branch mispredicts. Branch mispredicts necessitate that the pipeline be flushed, a very cycle expensive procedure, so by lessening the number of branches, predication can greatly reduce wasted processor time.
àSpeculation:
The latency of memory is a big performance bottleneck in today's systems. IA-64 architecture employs a technique called as speculation to initiate loads from memory earlier in the instruction stream—even before a branch. Data speculation thus, is caching and calling for data that may be needed or may be changed before it is needed, so that, in the case that the data is needed and it has not changed, the CPU does not have to take a latency impact from calling for the data. The processor, with the help of compiled instructions, looks ahead, anticipates what information it may need, and then brings it to cache or into the processor. This helps hide memory latency. Thus speculation increases instruction level parallelismand reduces the impact of memory latency resulting in handsome performance gains.
Control speculation is a feature that runs deep within IA-64. Data while being loaded might be erroneous. So, here data values are associated with a bit which tells whether an error was generated when the data was loaded or not. This error ("Nat") bit follows the data through arithmetical operations, moves, and conditionals until the data is checked through a check instruction. This means that with IA-64, the compiler can aggressively load data well ahead of time (speculation) with out paying unnecessary exception penalties. A final benefit of "Nats" is that it provides for structured error handling as is found in C and C++. Because the compiler can schedule the "check" instruction whenever it wants, it is possible to safely defer errors until the best possible time to recover from them. The architectural support for structured error handling in IA-64 is an important feature promoting system reliability and performance
àCache:
When a processor is waiting for data or instructions, time is wasting. The longer it takes for data and instructions to get to the CPU, the worse it gets. When data and instructions are in cache, the processor can grab them much quicker than when having to go to slow main memory. Not only is cache latency much lower than DRAM latency, the bandwidth is much higher.
There are some trick programming techniques in use out there to keep often-used data and instructions in cache and they are not the kind of techniques you learn in your high school BASIC course. Still, the easiest way to keep data and instructions in cache is to have a lot of cache to keep them in. Intel knew that when they designed the Itanium.
The Itanium has three levels of cache. L1 and L2 are on-die while L3 is on cartridge. According to Intel, the L3 cache weighs in at 2MB or 4MB of four-way set associative cache on two or four 1MB chips. IDC reports that the L2 cache size is 96k in size, and the L1 cache, which does not deal with floating point data, has a 16KB integer data and a 16KB instruction cache.
The 294.8 million transistors of (4MB) level three cache runs at the full processor speed, giving 12.8GBps of memory bandwidth at 800MHz. With 2MB or 4MB of L3 cache on the Itanium, the chances of the required data and instructions being in cache are quite good, bus traffic can be reduced, and performance increases. With six pipelines hungry for instructions and data, the Itanium needs all the cache it can get. Caching is made effective through data speculation and cache hints. Data speculation has been explained above. Cache hints are two-bit markers for memory loads set by the compiler that help the CPU find data in cache. This improves the speed of retrieving data from cache.
èThe Overall Architecture:
The IA-64 architecture is based on innovative techniques/features such as Explicit parallelism, and Predication and Speculation, resulting in superior Instruction Level Parallelism (ILP) and increased instructions per cycle (IPC) to address the current and future requirements of these demanding Internet, high end server, and workstation applications. In addition, the IA-64 architecture provides headroom and scalability for continued future growth. The Itanium was not designed for small systems, it is intended for 1 to 4000 processor workstations and servers. The first-generation Itanium processor is able to issue up to six EPIC instructions in parallel every clock cycle.
The six issue (two bundle) scheduler disperses instructions into nine functional slots, two integer slots, two floating-point slots, two memory slots, and three branch slots, giving a total of nine dispersal slots. Proper compiler design should be able to handle most situations without overloading any type of functional slot. In addition to pure clock speed boosts, futures Itanium CPUs are sure to have more functional units (FMACs, ALUs).
The Itanium contains four pipelined FMAC units (Floating-point Multiply Add Calculator). The primary two are each capable of processing two single-precision, two double precision, or two double-extended-precision floating-point operations per clock. That yields up to 3.2GFLOPS of highly precise floating point processing. There are an additional two FMACs tuned for 3D applications. They are each capable of processing up to two single-precision floating-point operations per clock. That yields another 3.2GFLOPS of single-precision processing power. All together, the Itanium has a theoretical max of 6.4GLOPS of single-precision floating point processing power. There are four pipelined ALUs (Arithmetic Logic Unit) in the original Itanium. Each can process one integer calculation per cycle. They can also process MMX type instructions.
The Itanium comes with 128 floating point and 128 integer registers. When processing up to 20 operations in a single clock, the registers give plenty of room for data inside the processor. The registers also have the ability to rotate. Rotating registers allows the processor to perform an operation on multiple software accessible registers in turn efficient parallel execution is achieved for integer operations with the six 1-cycle integer and six 2-cycle multi-media units which accomplish full symmetric bypassing with each other and the L1D cache in combination with a 20 ported, 128 X 65b register file. Virtually any
instruction can be predicated off with a prior compare instruction. The integer units and register file are 4mm X 1.9mm including hardware support for IA-32 code. The dual 82b FMAC units have a 4-cycle latency and are fully bypassed with each other. In combination with the 14 ported, 128 X 82b register file and other miscellaneous FP support, these units are 9mm X 2.2mm
There are several Itanium features designed to help with hardware scalability: a full-CPU-speed Level 2 bus, a large L3 cache, deferred-transaction support and flexible page sizes. The full-CPU-speed Level 3 bus provides quick communication between CPUs. The large L2 cache reduces inter-CPU bus traffic by keeping data close to the CPU that needs it. Deferred-transaction support can stop one CPU from getting in the way of another.
Flexible page sizes, from 4KB to 256MB, give the Itanium family the flexibility to access small amounts of memory in small chunks and massive amounts of memory in massive chunks without the pre-validated, 4 port 16KB L1D cache [1] is tightly coupled to the integer units to achieve the half cycle load. As a result, the less latency sensitive FPU directly interfaces to the L2D cache [1] with 4 82b load ports (6 cycle latency) and 2 82b store ports. The 3MB, 12 cycle latency L3 cache [1] is implemented with 135 separate "subarrays" that enable high density and the ability to conform to the irregular shape of the processor core with flexible subarray placement. Each level of on-chip cache has matched bandwidths at 32GB/s across the hierarchy he overhead of smaller page sizes.
The front-end instruction fetch pipe stages are de-coupled from the backend stages via an 8-bundle queue. The same pre-validated, single-ended cache technology [1] that enables the 1/2 cycle latency L1D cache is leveraged to improve instruction fetch and branch restore. Each set of 6 instructions (2 bundles) stored in the 16KB L1 cache is accompanied by branch target address and branch prediction information. The data is read out of the cache in the first half-cycle. In the next half-cycle, the prediction information is examined and if the branch to which the stored target corresponds is predicted taken; this address is input to the instruction pointer mux for the next instruction fetch.
The Itanium has extensive error handling capabilities. It features ECC and parity error checking on most processor caches and busses. The processor also has the capability to kill an application or thread that has experienced a machine error without having to reboot. A major link in the food delivery system for the Itanium is the system bus. The Itanium uses a 2.1GBps multi-drop system bus to keep well fed with data and instructions.
The first generation of Itanium systems, using the 460GX chipset, is expandable with up to 64GB of memory. Generations beyond that are able to take more memory. Higher end Itanium systems designed by the likes of SGI, IBM and HP should eventually be able to take far more than 64GB.
Specifications:
Here are the Itanium specifications to summarize things.
Physical Characteristics:
- 25.4M transistors
- .18micron CMOS process
- 6 metal layers
- C4 (flip-chip) assembly technology
- 1012-pad organic land grid array
- 733MHz and 800MHz initial release clock speeds
Instruction Dispersal:
- 2 bundle dispersal windows
- 3 instructions per bundle
- 9 function unit slots
- 2 integer slots
- 2 floating point slots
- 2 memory slots
- 3 branch slots
- Maximum of 6 instructions issued each cycle
Floating Point Units:
- 2 extended and double precision FMACs (Floating-point Multiply Add Calculators)
- 4 double or single precision operations per clock maximum
- 3.2 GFLOPS of peak double precision floating point performance at 800MHz
- 2 additional single precision FMACs
- 4 single precision operations per clock maximum
- 6.4 GFLOPS of peak single precision floating point performance total at 800MHz
Integer and Branch Units:
- 4 single cycle integer ALUs
- 4 MMX units
- 3 branch units
Compatibility:
The Itanium is fully x86 compatible in hardware. Applications and operating systems can run without any changes. A decoder internal to the CPU decodes x86 instructions into EPIC instructions, than dynamically schedules them to run with increased parallelism. While the Itanium is compatible with x86 software, it is not expected to do it quickly. Compatibility is being included to ease the transition from x86 code to EPIC code.
Most of the compilers that are used to create all the programs have already been written for the IA-64 architecture, and others are in pursuit. So writing programs for the Itanium is not a permanent headache, though it may seem troublesome for the time being, but that’s what the 31/2” floppy diskette seemed at first.
Fig. Seamless Architecture allows Full Itanium Performance on IA32system Functions.
Applications:
The new IA-64 architecture finds its applications in various fields. This processor has a tremendous potential in terms of speed and performance. The inherently scalable nature of the architecture makes it very compelling for the high-end server and workstation market segments. The Itanium was not designed for small systems, it is intended for 1 to 4000 processor workstations and servers.
The biggest, toughest computing challenges in the world are tackled—and very often solved through high performance computing (HPC). Such diverse and life-essential research areas like meteorological modeling, automotive crash test simulations, human genome mapping, and nuclear blast modeling are all part of HPC. Solutions built on open standards-based Intel® platforms provide supercomputing capabilities at significant cost savings for cutting-edge scientific, research, industry, and enterprise HPC applications.
We look at some of its applications, which due to its innovative architecture make it the best bet.
Business Intelligence
Itanium’s 64-bit VLM support directly benefits multi-terabyte data warehousing and data mining, while other Itanium architecture techniques — such as predication — improve parallel transaction code execution even in the face of unpredictable control flows. Itanium’s floating-point performance is also significant for software performing complex numerical analyses of large data sets, as is the ability to explicitly specify code execution order in well understood algorithms.
Technical and Scientific
Performance is a primary driver in electronic design automation (EDA), mechanical design automation (MDA), digital content creation (DCC), financial services, and scientific application purchases. Itanium’s floating-point features, large memory addressability for large data sets, and increased parallelism for complex processes combine to deliver new performance and scalability benefits to a market that is already highly receptive to the price/performance delivered by Intel architectures.
Security
The performance of encryption and decryption operations limits the scope of security system deployment because encrypting all network traffic from a client or a server requires 10 to 20 times the processor resources of unencrypted traffic. At a minimum, Itanium security performance is likely to benefit those e-Business applications that make some use of security protocols but are not dedicated to security operations.
Directory Services
The IA-64 architecture and its first microprocessor implementation, the Intel® Itanium™ processor, provide capabilities that enhance the performance and scalability of directory services. The 64-bit addressing enabled by the Intel Itanium architecture will help overcome the scalability barriers and awkward, maintenance-intensive partitioning directory schemes of current directory services on 32-bit platforms.
Conclusion:
First processor with Itanium architecture. Developed, manufactured, and marketed by Intel® .In 1994, Hewlett-Packard and Intel Corporation agreed to jointly design EPIC (Explicitly Parallel Instruction Computing), a post-RISC and post-IA-32 technology. Using EPIC concepts, HP and Intel® then jointly defined Itanium’s 64-bit Itanium Processor Family (IPF) architecture, the basis of Intel’s future, high-performance microprocessor family a broad range of technical and commercial applications at 1.0GHz. The easiest way to keep data and instructions in cache is to have a lot of cache to keep them in. Intel® knew that when they designed the Itanium.
The Itanium has a complex, bleeding edge, forward looking processor family that holds promise for huge gains in processing power. The processor uses the entirely new EPIC architecture that has the potential to deliver large improvements in processor parallelism. It is all about speed, and the Itanium has the paper pedigree to deliver it. If Intel can deliver, expect to see blood in the enterprise server water.
GLOSSARY:
Architecture—Implementation of an instruction set for a processing method (for example, PA-RISC, IA-32, MIPS, UltraSPARC).
Branch—A branch in the program that precedes a decision about what path to take (for example, IF… THEN… ELSE).
CISC (Complex Instruction Set Computing)—Processing method with a complex set of machine instructions of variable length. Up to now has been the opposite of RISC.
Compiler—A software tool that converts the instructions of a higher-level programming language that are the language of the microprocessor.
EPIC (Explicit Parallel Instruction Computing)—Processing method, jointly developed by Hewlett-Packard and Intel, that enhances and replaces processing according to the CISC and RISC procedures.
Explicit parallelism—The ability of the compiler to directly inform the processor of the independent nature of operations.
IA-32 (Intel 32-bit architecture)—The instruction set that forms the basis for the broad spectrum of Intel processors for notebooks, PCs, workstations, and servers. Executed in the Intel Pentium® processor, for example.
IA-64 or Itanium (Intel 64-bit architecture)—The Intel 64-bit architecture that implements EPIC concepts. It provides full IA-32 and PA-RISC compatibility. Now known as IPF (Intel Processor Family) architecture.
Implicit parallelism—Found in conventional microprocessor architectures, this requires the Compiler to create sequential machine code that can interact with the processor.
ISA (Instruction Set Architecture)—The operating instructions that tell a chip how to perform software functions and direct operations within the microprocessor. HP and Intel jointly developed a new 64-bit ISA architecture, known as the IPF architecture, which integrates technical concepts of the EPIC technology.
IPF (Itanium Processor Family) architecture—Processor architecture that implements the EPIC processing principle.
Itanium— First processor with Itanium architecture. Developed, manufactured, and marketed by Intel.
Latency, latency period, memory latency—The time the processor waits for the completion of loading instructions to get data from memory.
Merced™ processor—Early code name of the first processor from Intel in the Itanium family.
Mispredict—A wrong decision regarding which path to take.
Parallelism—The ability to execute multiple instructions at the same time. This is the opposite of sequential processing, one instruction after the other.
Predication—A technical concept that contributes to increasing overall performance by the removal of branches and associated mispredicts.
Processor—Semiconductor chip whose components process machine instructions of a specific architecture.
RISC (Reduced Instruction Set Computing)— Processing method with a reduced set of machine instructions of the same length (for example, 32 bits).
