We’re reacting to some of the big announcements that came out of Hot Chips 33 this week. We discuss these announcements and many more stories on this week’s Rundown.
Intel’s big push to expand their foundry capabilities in the US has just scored a big win. The company announced that the US Department of Defense will be using those facilities to product Rapid Assured Microelectronics Prototypes, or RAMPs for use in defense applications. The RAMP program is designed to encourage domestic chip production for a variety of needs while ensuring the chips are produced in-country in the event of issues with foreign supply chains. The entire book value of the agreement hasn’t been discussed and Intel must hit several milestones in the next 3 years in order to be the primary supplier.
If you wanted a good reason to patch critical exploits as soon as possible, look no further than this story from back in 2020. A bug affecting Citrix Application Delivery Controllers was announced in mid-December 2019. The patch for this was issued just over a month later but active exploiting was taking place as early as January 8th. One of the organizations that was targeted? The US Census Bureau. The organization was under attack on January 13th even though they took almost two weeks to find out that fact. The intruders were able to create rogue admin accounts for a foothold but were unable to deploy enough backdoors to get back in with their new privileges. The affected servers have been patched by now but this just highlights the vulnerabilities of government agencies we’ve heard about over the past few weeks.
The US Department of Energy just announced a new 44-petaflop exascale supercomputer called Polaris. But this news comes as a surprise, since Argonne Labs was already working on an exascale supercomputer known as Aurora. The addition is explained by a switch in suppliers: AMD and NVIDIA will be powering the new Polaris supercomputer, which will come online well before the long-delayed Intel Xeon-powered Aurora. Once again, Intel’s failure to meet deadlines presents an opportunity for their competitors.
HPE had a bit of a reshuffling of their executive ranks last week. There’s a new Greenlake Cloud Services Solutions group headed by current Aruba COO Vishal Lall, which is focused on the SaaS offerings from Ezmeral, Lighthouse, and others. The bigger news is the impending departure of the CTO, Kumar Sreekanti. He was formerly the CEO of BlueData, which was acquired in 2018. Whomever replaces Sreekanti will be taking over the Greenlake Platform Development side of the house as well as HPE continues to bet their future on Greenlake.
At their recent NGINX Sprint 2.0 event, NGINX (which is now owned by F5) reiterated their commitment to open source. They’re not just committing to keeping their world-beating webserver open and free, they’re also supporting open source projects across the spectrum. How does this differ from the typical approach of for-profit open source product companies?
We’ve been talking about the proposed NVIDIA purchase of ARM for quite a while now. It appears that the regulatory agencies are starting to weigh in and the news isn’t good for fans of the deal. The Competition and Markets Authority of the United Kingdom has expressed concerns about the deal due to potential harm to NVIDIA competitors. The thinking is that NVIDIA could restrict access to ARM chips or related products to help reduce competition. The remedy for this situation that was proposed by NVIDIA was rejected by the CMA and now they are doing a more in-depth analysis.
Researchers at mnemonic Labs found a new bug in TLS security last week that seems to affect more than just one provider. The exploit takes advantage of the Server Name Identification (SNI) filtering system. Using a special tool, attackers can exfiltrate data through the SSL hello packets. It’s a weird bug because even if the server is on a block list from a reputation or policy block the hello packets are still exchanged, allowing attackers to put data where it shouldn’t be located. While the linked article headline specifically mentions Cisco products, the researchers found the vulnerability in Palo Alto, Fortinet, and F5 gear as well. There is no timeline for a fix right now but the proof-of-concept is already in the wild so the clock is ticking.
There’s been a lot of news coming out of Hot Chips 33 this week. Since you are the resident chip expert here at the Rundown I thought it would be good to react to some of the big announcements. We’re indebted to our friends over at Serve the Home, who provided amazing coverage of this event.
Intel is focusing on RISC for their data processing card, which is confusingly called an IPU instead of a DPU. The so-called Mount Evans IPU packs an Arm Neoverse N1 core with P4 programming from Barefoot. This is in contrast to their exotic FPGA-based cards, one of which packs Xeon-D!
Esperanto showed a RISC accelerator card that packs an amazing 6558 RISC-V cores into a single Glacier Point Open Compute slot. Each of these features custom vector/tensor instructions, each with its own DRAM. These are amusingly called Minions and live in Neighborhoods as part of a Minion Shire in a mesh network on the chip.
Cerebras is famous for using an entire silicon wafer as a single massive chip, and now these have a fancy rack-sized chassis. The Wafer-Scale Engine 2 packs 2.6 trillion transistors, 850,000 AI-optimized processing cores, and 40 gigabytes of on-chip memory. And their new MemoryX module boasts up to 2.4 PB of RAM and flash!
Processors are appearing everywhere, and RAM leader Samsung is now packing them into memory modules! Their Aquabolt-XL packs a stripped-down RISC-32 processor into a high-bandwidth memory module for compute offload.
The Gestalt IT Rundown is a live weekly look at the IT news of the week. It broadcasts live on Facebook every Wednesday at 12:30pm ET. To watch along, “Like” our Facebook page. Be sure to subscribe to Gestalt IT on YouTube for even more weekly video content.