Changkyu Kim

Bio: Changkyu Kim is an academic researcher from Intel. The author has contributed to research in topics: Cache & Cache pollution. The author has an hindex of 33, co-authored 67 publications receiving 6705 citations. Previous affiliations of Changkyu Kim include Sony Broadcast & Professional Research Laboratories & Google.

Debunking the 100X GPU vs. CPU myth: an evaluation of throughput computing on CPU and GPU

Abstract: Recent advances in computing have led to an explosion in the amount of data being generated. Processing the ever-growing data in a timely manner has made throughput computing an important aspect for emerging applications. Our analysis of a set of important throughput computing kernels shows that there is an ample amount of parallelism in these kernels which makes them suitable for today's multi-core CPUs and GPUs. In the past few years there have been many studies claiming GPUs deliver substantial speedups (between 10X and 1000X) over multi-core CPUs on these kernels. To understand where such large performance difference comes from, we perform a rigorous performance analysis and find that after applying optimizations appropriate for both CPUs and GPUs the performance gap between an Nvidia GTX280 processor and the Intel Core i7-960 processor narrows to only 2.5x on average. In this paper, we discuss optimization techniques for both CPU and GPU, analyze what architecture features contributed to performance differences between the two architectures, and recommend a set of architectural features which provide significant improvement in architectural efficiency for throughput kernels.

An adaptive, non-uniform cache structure for wire-delay dominated on-chip caches

Abstract: Growing wire delays will force substantive changes in the designs of large caches. Traditional cache architectures assume that each level in the cache hierarchy has a single, uniform access time. Increases in on-chip communication delays will make the hit time of large on-chip caches a function of a line's physical location within the cache. Consequently, cache access times will become a continuum of latencies rather than a single discrete latency. This non-uniformity can be exploited to provide faster access to cache lines in the portions of the cache that reside closer to the processor. In this paper, we evaluate a series of cache designs that provides fast hits to multi-megabyte cache memories. We first propose physical designs for these Non-Uniform Cache Architectures (NUCAs). We extend these physical designs with logical policies that allow important data to migrate toward the processor within the same level of the cache. We show that, for multi-megabyte level-two caches, an adaptive, dynamic NUCA design achieves 1.5 times the IPC of a Uniform Cache Architecture of any size, outperforms the best static NUCA scheme by 11%, outperforms the best three-level hierarchy--while using less silicon area--by 13%, and comes within 13% of an ideal minimal hit latency solution.

Exploiting ILP, TLP, and DLP with the polymorphous TRIPS architecture

Abstract: This paper describes the polymorphous TRIPS architecture which can be configured for different granularities and types of parallelism. TRIPS contains mechanisms that enable the processing cores and the on-chip memory system to be configured and combined in different modes for instruction, data, or thread-level parallelism. To adapt to small and large-grain concurrency, the TRIPS architecture contains four out-of-order, 16-wide-issue Grid Processor cores, which can be partitioned when easily extractable fine-grained parallelism exists. This approach to polymorphism provides better performance across a wide range of application types than an approach in which many small processors are aggregated to run workloads with irregular parallelism. Our results show that high performance can be obtained in each of the three modes--ILP, TLP, and DLP-demonstrating the viability of the polymorphous coarse-grained approach for future microprocessors.

ClearPath: highly parallel collision avoidance for multi-agent simulation

Abstract: We present a new local collision avoidance algorithm between multiple agents for real-time simulations. Our approach extends the notion of velocity obstacles from robotics and formulates the conditions for collision free navigation as a quadratic optimization problem. We use a discrete optimization method to efficiently compute the motion of each agent. This resulting algorithm can be parallelized by exploiting data-parallelism and thread-level parallelism. The overall approach, ClearPath, is general and can robustly handle dense scenarios with tens or hundreds of thousands of heterogeneous agents in a few milli-seconds. As compared to prior collision avoidance algorithms, we observe more than an order of magnitude performance improvement.

FAST: fast architecture sensitive tree search on modern CPUs and GPUs

Abstract: In-memory tree structured index search is a fundamental database operation. Modern processors provide tremendous computing power by integrating multiple cores, each with wide vector units. There has been much work to exploit modern processor architectures for database primitives like scan, sort, join and aggregation. However, unlike other primitives, tree search presents significant challenges due to irregular and unpredictable data accesses in tree traversal. In this paper, we present FAST, an extremely fast architecture sensitive layout of the index tree. FAST is a binary tree logically organized to optimize for architecture features like page size, cache line size, and SIMD width of the underlying hardware. FAST eliminates impact of memory latency, and exploits thread-level and datalevel parallelism on both CPUs and GPUs to achieve 50 million (CPU) and 85 million (GPU) queries per second, 5X (CPU) and 1.7X (GPU) faster than the best previously reported performance on the same architectures. FAST supports efficient bulk updates by rebuilding index trees in less than 0.1 seconds for datasets as large as 64Mkeys and naturally integrates compression techniques, overcoming the memory bandwidth bottleneck and achieving a 6X performance improvement over uncompressed index search for large keys on CPUs.

Second-generation PLINK: rising to the challenge of larger and richer datasets

Abstract: Background: PLINK 1 is a widely used open-source C/C++ toolset for genome-wide association studies (GWAS) and research in population genetics. However, the steady accumulation of data from imputation and whole-genome sequencing studies has exposed a strong need for faster and scalable implementations of key functions, such as logistic regression, linkage disequilibrium estimation, and genomic distance evaluation. In addition, GWAS and population-genetic data now frequently contain genotype likelihoods, phase information, and/or multiallelic variants, none of which can be represented by PLINK 1’s primary data format. Findings: To address these issues, we are developing a second-generation codebase for PLINK. The first major release from this codebase, PLINK 1.9, introduces extensive use of bit-level parallelism, O √ n -time/constant-space Hardy-Weinberg equilibrium and Fisher’s exact tests, and many other algorithmic improvements. In combination, these changes accelerate most operations by 1-4 orders of magnitude, and allow the program to handle datasets too large to fit in RAM. We have also developed an extension to the data format which adds low-overhead support for genotype likelihoods, phase, multiallelic variants, and reference vs. alternate alleles, which is the basis of our planned second release (PLINK 2.0). Conclusions: The second-generation versions of PLINK will offer dramatic improvements in performance and compatibility. For the first time, users without access to high-end computing resources can perform several essential analyses of the feature-rich and very large genetic datasets coming into use.

Second-generation PLINK: rising to the challenge of larger and richer datasets

Abstract: PLINK 1 is a widely used open-source C/C++ toolset for genome-wide association studies (GWAS) and research in population genetics. However, the steady accumulation of data from imputation and whole-genome sequencing studies has exposed a strong need for even faster and more scalable implementations of key functions. In addition, GWAS and population-genetic data now frequently contain probabilistic calls, phase information, and/or multiallelic variants, none of which can be represented by PLINK 1's primary data format. To address these issues, we are developing a second-generation codebase for PLINK. The first major release from this codebase, PLINK 1.9, introduces extensive use of bit-level parallelism, O(sqrt(n))-time/constant-space Hardy-Weinberg equilibrium and Fisher's exact tests, and many other algorithmic improvements. In combination, these changes accelerate most operations by 1-4 orders of magnitude, and allow the program to handle datasets too large to fit in RAM. This will be followed by PLINK 2.0, which will introduce (a) a new data format capable of efficiently representing probabilities, phase, and multiallelic variants, and (b) extensions of many functions to account for the new types of information. The second-generation versions of PLINK will offer dramatic improvements in performance and compatibility. For the first time, users without access to high-end computing resources can perform several essential analyses of the feature-rich and very large genetic datasets coming into use.

Dark Silicon and the End of Multicore Scaling

Abstract: A key question for the microprocessor research and design community is whether scaling multicores will provide the performance and value needed to scale down many more technology generations. To provide a quantitative answer to this question, a comprehensive study that projects the speedup potential of future multicores and examines the underutilization of integration capacity-dark silicon-is timely and crucial.

Reciprocal n-Body Collision Avoidance

Abstract: In this paper, we present a formal approach to reciprocal n-body collision avoidance, where multiple mobile robots need to avoid collisions with each other while moving in a common workspace In our formulation, each robot acts fully independently, and does not communicate with other robots Based on the definition of velocity obstacles [5], we derive sufficient conditions for collision-free motion by reducing the problem to solving a low-dimensional linear program We test our approach on several dense and complex simulation scenarios involving thousands of robots and compute collision-free actions for all of them in only a few milliseconds To the best of our knowledge, this method is the first that can guarantee local collision-free motion for a large number of robots in a cluttered workspace

Dark silicon and the end of multicore scaling

Abstract: Since 2005, processor designers have increased core counts to exploit Moore's Law scaling, rather than focusing on single-core performance. The failure of Dennard scaling, to which the shift to multicore parts is partially a response, may soon limit multicore scaling just as single-core scaling has been curtailed. This paper models multicore scaling limits by combining device scaling, single-core scaling, and multicore scaling to measure the speedup potential for a set of parallel workloads for the next five technology generations. For device scaling, we use both the ITRS projections and a set of more conservative device scaling parameters. To model single-core scaling, we combine measurements from over 150 processors to derive Pareto-optimal frontiers for area/performance and power/performance. Finally, to model multicore scaling, we build a detailed performance model of upper-bound performance and lower-bound core power. The multicore designs we study include single-threaded CPU-like and massively threaded GPU-like multicore chip organizations with symmetric, asymmetric, dynamic, and composed topologies. The study shows that regardless of chip organization and topology, multicore scaling is power limited to a degree not widely appreciated by the computing community. Even at 22 nm (just one year from now), 21% of a fixed-size chip must be powered off, and at 8 nm, this number grows to more than 50%. Through 2024, only 7.9x average speedup is possible across commonly used parallel workloads, leaving a nearly 24-fold gap from a target of doubled performance per generation.