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

By Victor W. Lee, Changkyu Kim, Jatin Chhugani, Michael E. Deisher, Daehyun Kim, Anthony D. Nguyen, Nadathur Rajagopalan Satish, M. Smelyanskiy, Srinivas Chennupaty, Per Hammarlund and 2 more. International Symposium on Computer Architecture, 2010

854 citations recorded by the source corpus.

Subfields: Computer Networks and Communications, Hardware and Architecture

View Paper: https://www.semanticscholar.org/paper/94fd94b7ebcdd80e47706376aa0540cbeb009262

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Networks and Communications

In Hardware and Architecture

Citations in the corpus

Dark silicon and the end of multicore scaling (2011, methodology): “Our model does not account for specialized compute units, which contribute to the speedup in [21].”
Robotic Hand Pose Estimation Based on Stereo Vision and GPU-enabled Internal Graphical Simulation (2016, background): “Some authors state that it is possible to achieve a speed-up of almost 10 times [40] in high-level computer vision algorithms, when more conservative ones, and in a wider perspective, claim for 2.5 times on average [24].”
Efficient Parallel Graph Exploration on Multi-Core CPU and GPU (2011, neutral): “GPU debates [1], [30].”
Accelerating CUDA graph algorithms at maximum warp (2011, result): “We also refer interested readers to other studies on the impact of architectural traits on various applications [19, 21], other than graph algorithms.”
Where is the data? Why you cannot debate CPU vs. GPU performance without the answer (2011, background): “[3] provide a compelling argument that fair comparisons are necessary; CPU and GPU performance evaluations are only valid when both have been fully optimized for their respective platforms, taking into consideration specific architectural features for each device and optimizing the kernels to take a”
Horton Tables: Fast Hash Tables for In-Memory Data-Intensive Computing (2016, background): “Recent work in high-performance databases that leverages BCHTs has shown that SIMD implementations of BCHTs, as well as larger data processing primitives, are often more than 3× faster than the highest performing implementations that use scalar instructions alone [6, 44, 60].”
Second-generation PLINK: rising to the challenge of larger and richer datasets (2014, background): “We have made the development of GPU-specific code a low priority since their installed base is much smaller than that of multicore processors, and the speedup factor over well-written multithreaded code running on similar-cost, less specialized hardware is usually less than 10x [10,11].”
Medical image processing on the GPU - Past, present and future (2013, result): “language has led to many reports of large computational gains compared to optimized CPU implementations, though some of these speedups have been questioned (Lee et al., 2010).”
Verifiable Computation with Massively Parallel Interactive Proofs (2012, background): “CPU implementation. However, GPUs are increasingly widespread, cost-effective, and power-efﬁcient, and they offer the potential for speedups in excess of those possible with commodity multi-core CPUs [17, 14]. We obtain server-side speedups ranging from 40-120 for the general-purpose protocol due to”