Ascend 910C

J. Soria, D. Tan, R. Chiari, Yi-long Wu, L. Paz-Ares, J. Wolf, Sarayut Lucien Geater, S. Orlov, D. Cortinovis, Chong-Jen Yu, Maximillian Hochmair, A. Cortot, C. Tsai, D. Moro-Sibilot, R. Campelo, T. McCulloch, P. Sen, M. Dugan, Serafino Pantano, F. Branle, C. Massacesi, G. Castro

R. Kapoor, P. Ho, N. Campbell, I. Chang, A. Deykin, F. Forrestal, N. Lucas, Bei Yu, D. Arnold, M. Freedman, M. Goldman, H. Hartung, E. Havrdová, D. Jeffery, Aaron E Miller, F. Sellebjerg, D. Cadavid, D. Mikol, D. Steiner, E. Bartholome, M. D'hooghe, Massimo Pandolfo, B. Wijmeersch, V. Bhan, G. Blevins, D. Brunet, V. Devonshire, P. Duquette, M. Freedman, F. Grand'Maison, F. Jacques, Y. Lapierre, Liesly Lee, S. Morrow, M. Yeung, M. Dufek, P. Kaňovský, I. Štětkářová, M. Talábová, J. Frederiksen, M. Kant, T. Petersen, M. Ravnborg, L. Airas, I. Elovaara, J. Eralinna, T. Sarasoja, A. A. Khedr, D. Brassat, B. Brochet, W. Camu, M. Debouverie, D. Laplaud, C. L. Frenay, J. Pelletier, P. Vermersch, Sandra Vukusi, K. Baum, A. Berthele, Juergen Faiss, P. Flachenecker, R. Hohlfeld, M. Krumbholz, Christoph Lassek, M. Maeurer, S. Meuth, T. Ziemssen, O. Hardiman, C. McGuigan, A. Achiron, D. Karussis, R. Bergamaschi, V. Morra, G. Comi, S. Cottone, Luigi M. E. Grimaldi, G. Mancardi, L. Massacesi, U. Nocentini, M. Salvetti, E. Scarpini, P. Sola, G. Tedeschi, M. Trojano, M. Zaffaroni, S. Frequin, R. Hupperts, J. Killestein, H. Schrijver, Ronald Van Dijl, E. V. Munster, M. Czarnecki, W. Drozdowski, W. Fryze, H. Hertmanowska, Jan Ilkowski, A. Kaminska, G. Kłodowska-Duda, M. Maciejowski, Ewa Motta, R. Podemski, A. Potemkowski, T. Róg, K. Selmaj, Z. Stelmasiak, A. Stępień, A. Tutaj, J. Zaborski, A. Boyko, Z. Chefranova, E. Evdoshenko, F. Khabirov, S. Sivertseva, E. Yakupov, J. C. Á. Cermeño, A. Escartín, O. Fernández, A. García‐Merino, Miguel Ángel Hernández Pérez, G. I. Ayuso, J. M. Lallana, X. M. Gairin, C. Oreja-Guevara, Albert Saiz Hinarejos, M. Gunnarsson, J. Lycke, C. Martin, F. Piehl, H. Roshanisefat, Peter Sundstrom, M. Duddy, B. Gran, T. Harrower, J. Hobart, M. Lee, P. Mattison, R. Nicholas, O. Pearson, W. Rashid, D. Rog, B. Sharrack, E. Silber, Ben Turner, Anna C. Williams, J. Woolmore, C. Young, D. Bandari, Joseph Berger, A. Camac, S. Cohan, J. Conway, K. Edwards, M. Fabian, J. Florin, S. Freedman, D. Garwacki, M. Goldman, D. Harrison, Craig Herrman, Deren Huang, A. Javed, S. Kamin, G. Katsamakis, B. Khatri, A. Langer-Gould, S. Lynch, D. Mattson, T. Miller, Augusto A Miravalle, H. Moses, S. Muley, James Napier, A. Nielsen, A. Pachner, G. Pardo, M. Picone, Derrick Robertson, W. Royal, Christopher Sheppard, B. Thrower, C. Twyman, E. Waubant, J. Wendt, V. Yadav, R. Zabad, Greg Zarelli

H. Liao, Jiajin Tu, Jing Xia, Hu Liu, Xiping Zhou, Honghui Yuan, Yuxing Hu

Deep neural networks (DNNs) have been successfully applied to a great variety of applications, ranging from small IoT devices to large scale services in a data center. In order to improve the efficiency of processing these DNN models, dedicated hardware accelerators are required for all these scenarios. Theoretically, there exists an optimized acceleration architecture for each application. However, considering the cost of chip design and corresponding tool-chain development, researchers need to trade off between efficiency and generality. In this work, we demonstrate that it is practical to use a unified architecture, called Ascend, to support those applications, ranging from IoT devices to data-center services. We provide a lot of design details to explain that the success of Ascend relies on contributions from different levels. First, heterogeneous computing units are employed to support various DNN models. And the datapath is adapted according to the requirement of computing and data access. Second, when scaling the Ascend architecture from a single core to a cluster containing thousands of cores, it involves design efforts, such as memory hierarchy and system level integration. Third, a multi-tier compiler, which provides flexible choices for developers, is the last critical piece. Experimental results show that using accelerators based on the Ascend architecture can achieve comparable or even better performance in different applications. In addition, various chips based on the Ascend architecture have been successfully commercialized. More than 100 million chips have been used in real products.

P. Ghia, A. Pluta, M. Wach, D. Lysák, M. Šimkovič, I. Kriachok, Á. Illés, J. de la Serna, S. Dolan, P. Campbell, G. Musuraca, A. Jacob, E. Avery, Jae Hoon Lee, G. Usenko, M. Wang, T. Yu, W. Jurczak

Acalabrutinib is a Bruton tyrosine kinase inhibitor approved for patients with chronic lymphocytic leukemia (CLL). ASCEND is the pivotal phase 3 study of acalabrutinib versus investigator’s choice of idelalisib plus rituximab (IdR) or bendamustine plus rituximab (BR) in patients with relapsed/refractory (R/R) CLL. In the primary ASCEND analysis (median 16.1-month follow-up), acalabrutinib showed superior efficacy with an acceptable tolerability profile versus IdR/BR; here, we report final ~4 year follow-up results. Patients with R/R CLL received oral acalabrutinib 100 mg twice daily until progression or unacceptable toxicity, or investigator’s choice of IdR or BR. A total of 310 patients (acalabrutinib, n = 155; IdR, n = 119; BR, n = 36) were enrolled. At median follow-up of 46.5 months (acalabrutinib) and 45.3 months (IdR/BR), acalabrutinib significantly prolonged investigator-assessed progression-free survival (PFS) versus IdR/BR (median, not reached [NR] vs 16.8 months; P < 0.001); 42-month PFS rates were 62% (acalabrutinib) versus 19% (IdR/BR). Median overall survival (OS) was NR (both arms); 42-month OS rates were 78% (acalabrutinib) versus 65% (IdR/BR). Adverse events led to drug discontinuation in 23%, 67%, and 17% of patients in the acalabrutinib, IdR, and BR arms, respectively. Events of clinical interest (acalabrutinib vs IdR/BR) included all-grade atrial fibrillation/flutter (8% vs 3%), all-grade hypertension (8% vs 5%), all-grade major hemorrhage (3% vs 3%), grade ≥3 infections (29% vs 29%), and second primary malignancies excluding nonmelanoma skin cancer (7% vs 2%). At ~4 years follow-up, acalabrutinib maintained favorable efficacy versus standard-of-care regimens and a consistent tolerability profile in patients with R/R CLL.

L. Ybanez, A. Colle, E. Nilsson, F. Berg, G. Galla, M. Tassisto, J. Rivenc, F. Kapaun, G. Steiner

If a cold source such as liquid H2 or cryogenic fuel is available, cryogenic and high-temperature superconductivity technologies are promising to significantly increase performance of electric propulsion systems. A first study for aeronautic applications shows that the power density of electrical components could be multiplied 2 to 3 times to reach more than 30kW/kg for electric motors and power electronics, the weight of cables can be considerably reduced, and the efficiency increased by more than 50% with a voltage between below 500 V for multi-megawatt applications. With ASCEND, AIRBUS intends to demonstrate the potential and feasibility of a cryogenic and superconducting powertrain to breakthrough aircraft electric propulsion performance.