https://www.tomshardware.com/news/intel-xe-hpc-ponte-vecchio-examined Skip to main content (*) ( ) [] Tom's Hardware [ ] Search [ ] RSS * * Reviews * Best Picks * Raspberry Pi * CPUs * GPUs * More + Laptops + SSDs + Motherboards + Cooling + Desktops + PC Builds + Monitors + RAM + PC Cases + Keyboards + Headsets + Mice + Power Supplies + VR Headsets + Windows Tips Forums Trending * Radeon RX 6700 XT * Raspberry Pi Pico * Ryzen 5000 Mobile * Rocket Lake * Apple M1 Tom's Hardware is supported by its audience. When you purchase through links on our site, we may earn an affiliate commission. Learn more 1. Home 2. News Intel's Ponte Vecchio Xe-HPC GPU Boasts 100 Billion Transistors By Anton Shilov 24 March 2021 Intel gives a glimpse on Ponte Vecchio's AI performance. * * * * * * Comments (9) Intel (Image credit: Intel) Intel uploaded a detailed description of its upcoming Ponte Vecchio Xe-HPC GPU for supercomputers just hours after outlining its IDM 2.0 strategy that involves usage of internal and external manufacturing capabilities. Ponte Vecchio, which employs components produced by Intel, Samsung, and TSMC using a variety of process technologies, demonstrates Intel's vision of the future in the best way possible. Intel's codenamed Ponte Vecchio is the company's first GPU based on the Xe-HPC microarchitecture that will be initially used for Argonne National Laboratory's Aurora supercomputer along with Intel's next-generation Xeon Scalable 'Sapphire Rapids' processor. The machine will be one of the industry's first supercomputers to feature over 1 ExaFLOPS FP64 performance. Intel (Image credit: Intel) Over time, the part will be available to other customers and Intel might even customize it as it is relatively easy to do given the fact that the Ponte Vecchio uses a disaggregated modular architecture, Intel's new approach to complex processors. Intel (Image credit: Intel) In fact, it would be impossible to build a monolithic Ponte Vecchio as it is a massive processor featuring 47 components, over 100 billion transistors, and offering PetaFLOPS-class AI performance (more on this later). The Ponte Vecchio includes the following tiles/chiplets: * 2 base tiles made using Intel's 10 nm SuperFin technology * 16 compute tiles produced by TSMC initially and then by Intel when its 7 nm technology is ready for high-volume manufacturing (HVM). * 8 Rambo cache tiles fabbed using Intel's 10 nm Enhanced SuperFin process * 11 EMIB links made by Intel * 2 Xe Link I/O tiles made by a foundry * 8 HBM memory stacks produced by a DRAM manufacturer At present, Intel only using its Ponte Vecchio Xe-HPC GPUs in the lab. While the modular design of the device allows the company to build it more or less cost efficiently, tailoring the design's thermals, voltages, and frequencies is tricky and will take some time. Intel (Image credit: Intel) One interesting thing to note about Intel's Ponte Vecchio description is that the chipmaker said that it offers 'PetaFLOPS-class AI performance.' There are numerous AI workloads that require different compute precision. Intel usually considers FP16 to be the optimal precision for AI, so when the company says that that its Ponte Vecchio is a 'PetaFLOP scale AI computer in the palm of the hand,' this might mean that that the GPU features about 1 PFLOPS FP16 performance, or 1,000 TFLOPS FP16 performance. To put the number into context, Nvidia's A100 compute GPU provides about 312 TFLOPS FP16 performance. Argonne National Laboratory's Aurora supercomputer is due in 2022. Topics Newsfeed GPUs See all comments (9) [ ] 9 Comments Comment from the forums * hotaru251 24 March 2021 22:22 i'm sorry....did they seriously use "magical tiles" in presentation? these arent products for little kids <_< Reply * Neilbob 24 March 2021 23:10 hotaru251 said: i'm sorry....did they seriously use "magical tiles" in presentation? these arent products for little kids <_< That made me wince slightly too. Are Intel channelling Apple now? Reply * Co BIY 24 March 2021 23:14 hotaru251 said: i'm sorry....did they seriously use "magical tiles" in presentation? Sounds better "chips our rivals make that we can't". Agree sounds like a placeholder word that slipped past into the final draft. Reply * Chung Leong 25 March 2021 06:29 hotaru251 said: i'm sorry....did they seriously use "magical tiles" in presentation? these arent products for little kids <_< Remember, the presentation is meant for members of the press. Reply * InvalidError 25 March 2021 12:44 That's some epic packaging-fu. When you have nearly infinite resources and want to stomp competition in the "if you have to ask the price, you can't afford it" market segment, you do what you gotta do. Reply * spongiemaster 25 March 2021 12:57 You sure this picture goes with this article? That looks like the new cooler bundled with the i9-11900k. Reply * TerryLaze 25 March 2021 14:20 spongiemaster said: You sure this picture goes with this article? That looks like the new cooler bundled with the i9-11900k. this might mean that that the GPU features about 1 PFLOPS FP16 performance, or 1,000 TFLOPS FP16 performance. To put the number into context, Nvidia's A100 compute GPU provides about 312 TFLOPS FP16 performance. If it really is 3 times this then it looks about right. Reply * comedichistorian 25 March 2021 14:40 Anyone else mistake "Ponte Vecchio" for "Pancho Villa"? No? Anyone? Reply * spongiemaster 25 March 2021 15:14 TerryLaze said: If it really is 3 times this then it looks about right. Way to miss the joke. It isn't 3 times that. That's eight A100's which is 2.5 PFLOPS. Reply * View All 9 Comments Show more comments Be In the Know Get instant access to breaking news, in-depth reviews and helpful tips. [ ] [ ] [ ] Contact me with news and offers from other Future brands [ ] Receive email from us on behalf of our trusted partners or sponsors [Sign me up] Thank you for signing up to Tom's Hardware. You will receive a verification email shortly. There was a problem. Please refresh the page and try again. No spam, we promise. You can unsubscribe at any time and we'll never share your details without your permission. (*) ( ) Most Popular Most Shared 1. [missing-im] 1 Sourcekit PiTray Mini Review: Brings Ports to Raspberry Pi CM4 2. [missing-im] 2 How to Delete Duplicate Files in Linux with Fdupes 3. [missing-im] 3 Best Gaming Keyboards for 2021: Mechanical, TKL & More 4. [missing-im] 4 Best B550 Motherboards 2021: AMD's More Affordable PCIe 4.0 Option 5. [missing-im] 5 Where to Buy an RTX 3060 (or 3060 Ti), 3070, 3080 or 3090 1. [missing-im] 1 Sourcekit PiTray Mini Review: Brings Ports to Raspberry Pi CM4 2. [missing-im] 2 How to Delete Duplicate Files in Linux with Fdupes 3. [missing-im] 3 Best Gaming Keyboards for 2021: Mechanical, TKL & More 4. [missing-im] 4 Best B550 Motherboards 2021: AMD's More Affordable PCIe 4.0 Option 5. [missing-im] 5 Where to Buy an RTX 3060 (or 3060 Ti), 3070, 3080 or 3090 Tom's Hardware is part of Future US Inc, an international media group and leading digital publisher. Visit our corporate site. * Terms and conditions * Privacy policy * Cookies policy * Accessibility Statement * Advertise * About us * Contact us (c) Future US, Inc. 11 West 42nd Street, 15th Floor, New York, NY 10036.