AMD New Hires: Dan McNamara from Intel, Joshua Friedrich and Brad McCredie from IBM
by Dr. Ian Cutress on January 21, 2020 12:00 PM EST- Posted in
- CPUs
- AMD
- Enterprise
- GPUs
There is always a fairly fluid movement of engineers in the companies we cover, but recently AMD has made a number of substantial hires into several of its biggest departments.
The newest hire as reported by AMD is of Dan McNamara, former Senior Vice President of Intel’s Network and Custom Logic Group (formerly the Programmable Solutions Group) for several years and one of Intel’s hires from the Altera acquisition, having spent 11 years at Altera. Dan is set to be AMD’s SVP and GM of the Server Business Unit. This means that Dan’s role will expand through to accelerate AMD’s EPYC portfolio in order to engage better with AMD’s customers about server solutions built through AMD hardware. This is a slight jump away from his previous focus of SoCs, ASICs, and FPGAs, which may make some readers think that AMD might be going in that direction: Forrest Norrod is still heading up AMD’s Enterprise, Embedded, and Semi-Custom Business Group. Dan’s hiring was the focus of a recent AMD blog post about promotions and new hires.
While not specifically promoted by AMD in that post, the company has also made two key hires, both of which have spent the last 20+ years at IBM. First on that list is Dr. Bradley (Brad) McCredie, which AMD actually hired back in June. Brad started at IBM back in 1991 focusing on packaging and mainframes, eventually having spent over 28 years at IBM which includes stints in POWER system development and also holding the position of President of the OpenPOWER Foundation. He is now set in a role in AMD as a Corporate Vice President of GPU Platforms, but specifically will cover the execution of AMD’s data center strategy covering CPU and GPU, reporting directly to Forrest Norrod.
The other IBM hire is Joshua (Josh) Friedrich, a 20-year IBM veteran with roles in POWER5 clock gating, the POWER6 frequency lead, the POWER7 Chip Power Lead, the POWER8 Chip Circuit Lead, POWER9 concept/high-level design and uncore development, and his final role was developing future POWER designs at IBM. Within AMD, Josh’s role is listed as Corporate Vice President, and a spokesperson states that Josh’s role is in CPU/GPU integration technologies, reporting to CTO Mark Papermaster. That isn’t a lot to go on, as it could cover APUs or something more unique, and on probing AMD for more information, they’ve confirmed that it’s more on the platform/solution side to create differentiated products.
There is one departure to note: Scott Aylor, the Corporate Vice President and GM of AMD’s Data Center Solutions Group, is currently on leave and is set to leave the company at a future date. Dan McNamara is taking over his role, and CRN is reporting that Aylor’s departure is not related.
Title image, from left to right: Brad McCredie, Dan McNamara, Josh Friedrich
Update 1/22: Our moles have done some extra digging, and AMD hired two other long-time IBM employees in 2019.
Greg Wetli, who AMD hired back in February 2019 to manage server processor validation, spent 31 years at IBM in POWER processor validation as well as different aspects of chip design and tooling as far back as POWER4.
Norman James, hired back in March 2019 as an AMD Fellow on system architecture, spent 23 years at IBM starting as a senior engineer on POWER6 before working through to lead engineer on IBM's Lead Engineer of Cognative Systems, focusing on deep learning and machine learning.
Facebook
Twitter
RSS
43 Comments
View All Comments
Ian Cutress - Tuesday, January 21, 2020 - link
I think he was referring to the IBM POWER hires. Not everything is Intel vs AMD.twotwotwo - Tuesday, January 21, 2020 - link
Man, I agree you can't really guess much, but Friedrich's experience with POWER power (ha) and clock gating, and the "CPU/GPU integration" wording, sounds half-relevant to trying to reduce idle/low-load power use in mobile parts, which is maybe the top problem to solve to compete well there. You can't rely on sales from shortages and gaming laptops forever. :)And, of course, any work started by new hires today won't be in released products for years, so we shouldn't collectively hold our breath.
alufan - Wednesday, January 22, 2020 - link
hmmm early benchmark leaks from the new 4k laptop chips are showing they not only beat intel but have better battery life as well, however these are leaks so we must wait for the proper releases to be sure...but i for one cannot see AMD allowing a Ryzen release to fall behind an intel one maybe in the bulldozer etc days but not nowKevin G - Tuesday, January 21, 2020 - link
Interesting moves.One thing that rumored in the past was that AMD would be using the same socket as POWER systems. This was play to help reduce costs from both AMD and IBM’s perspectives. AMD cancelled G3MX which was part of that plan.
The idea that has been ping ponging around for a few years is who would buy Xilinx. For awhile it seemed like IBM and then nVidia. Makes me wonder if AMD would pursue such a move now.
rocketbuddha - Tuesday, January 21, 2020 - link
Nope. POWER is direct competition to AMD now that EPYC is basically better than Intel Xeon in every shape and form and IBM always tries to compare POWER to XEONs rather than OPTERONs. IBM has offloaded its foundries to Global Foundries and then inked a deal with Samsung for 7nm FinFET for future POWER 10 + series.HStewart - Tuesday, January 21, 2020 - link
I would not say that - each CPU has it advantage, AMD has more cores but Intel has advance instruction sets like AVX 512. POWER has it own line of softwareKorguz - Tuesday, January 21, 2020 - link
" but Intel has advance instruction sets like AVX 512" yea that only a handful of software actually uses, and the performance and power hit intels cpus take, almost makes it not worth it to use, point is ????AshlayW - Wednesday, January 22, 2020 - link
To be fair, 64 Zen2 cores with AVX2 can actually provide similar FPU throughput to 28 SKL cores with AVX512; at lower power use.Kevin G - Wednesday, January 22, 2020 - link
Context is important and hence why I mentioned the G3MX socket that AMD abandoned in favor of G34. This around the time JEDEC had abandoned work on FB2-DIMMs to replace the much maligned FB-DIMM standard. Intel and IBM had already started work on their next generation memory buffers and memory controllers to be placed on their respective CPU dies. AMD was looking to adopt the JEDEC standard before it dissolved. What eventually happened is that the memory buffers moved to the motherboards and if AMD needed to leverage the work they already put into that buffered serial memory controller, they'd need a source for a buffer chip or develop one themselves. The solution was to approach IBM and move to a common platform so that the buffer chip could be leveraged on AMD and IBM's systems would benefit from reduced costs due to higher volume of necessary parts. That deal fell through and AMD went with socket G34 by going a dual die route with their high end client chips. This was all when AMD was short on resources and starting to collapse.Conceivably AMD and IBM could have shared another socket with the launch of Epyc and the Power9 having some vast overlap in IO features (512 bit wide memory interfaces, high speed serial links for coherency based upon PCIe PHY etc.). Even if they were not electrically compatible at the socket level, being identical mechanically would save IBM and AMD money in similar platform component costs as those components would be manufactured in higher volume. Both of these companies are still relatively low volume in comparison to Intel's Xeon line up right now.
A shared platform has happened in the past. AMD licensed their FSB for the Athlons from DEC. AMD's chipsets at the time could be used for DEC's Alpha line up at the time. Similarly AMD licensed out Hypertransport and oddly the first Hypertransport product was from Transmeta. Mechanical compatibilities appear randomly for low volume and embedded stuff as a cost savings measure. For example, I have a PowerPC chip that fits into socket 7 via a PCB interposer.
FunBunny2 - Tuesday, January 21, 2020 - link
This is funny. For so many years, no one over 30 could be trusted in IT (esp. coding). The geezers have re-taken control!!! :)