Ep 195: We are Live! Talking Lattice, Marvell, Google, AMD, Broadcom, IBM and Meta
Leading global tech analysts Patrick Moorhead (Moor Insights & Strategy) and Daniel Newman (Futurum Research) are front and center on The Six Five analyzing the tech industry’s biggest news each and every week and also conducting interviews with tech industry “insiders” on a regular basis.
The Six Five represents six (6) handpicked topics that will be covered for five (5) minutes each.
Welcome to this week’s edition of “The 6-5.” I’m Patrick Moorhead with Moor Insights & Strategy, co-host, joined by Daniel Newman with Futurum Research. On this week’s show we will be talking:
3:33 Special Guests!
12:47 Lattice Development Conference
18:07 Marvell IA Day
26:12 Google Gemini AI 1.0 and New TPU
36:12 AMD Advancing AI Event
45:38 Broadcom Q4 FY23 Earnings
50:06 AI Alliance Led by IBM and Meta
Special Guests!
Lattice Development Conference
It was great to catch up with members of the @latticesemi leadership team and discuss its first, annual, Developer Conference, company performance, and what to expect in the future.
To pull off a dev conference, you have to have a vibrant ecosystem and big partners. I saw all… https://t.co/pQtZe6OAAD
— Patrick Moorhead (@PatrickMoorhead) December 6, 2023
Great week ahead in the Silicon Valley!
First, joining @LatticeSemi for its Developer Conference where The @SixFivePodcast will be joining to have conversations with top executives and leaders along with its partners. Looking forward to catching up with CEO Jim Anderson and… pic.twitter.com/vC3QFaY1L3
— Daniel Newman (@danielnewmanUV) December 4, 2023
Marvell IA Day
Good day overall by @MarvellTech. All public information, but a good refresh, in particular, how @MarvellTech plays in AI. Highlights on moving that AI data and processing it. Good thing that the company made the decisions it did 3-4 years ago.
Company did a good job showing… https://t.co/RUKk6nUUOm pic.twitter.com/Nv7PmUdl2h
— Patrick Moorhead (@PatrickMoorhead) December 6, 2023
Some thoughts on Marvell as its industry analyst day kicks off.
Overall, an excellent opening presentation from COO Chris Koopmans.
Generally speaking, @MarvellTech contributions to #AI and broader data infrastructure are not well understood.
The company is seeing rapid growth… pic.twitter.com/WCPWHqm7yu— Daniel Newman (@danielnewmanUV) December 5, 2023
Excellent day @MarvellTech
Always gracious hosts and deep with technology and innovation. Thanks to CEO Matt Murphy and COO Chris Koopmans and your team for hosting us in Santa Clara.
Networking is one of the unsung heroes of #AI and the #GenAI boom.
Marvell is extremely… pic.twitter.com/BmJmYkaQNn— Daniel Newman (@danielnewmanUV) December 6, 2023
I agree with @MarvellTech's self assessment that it's ready for the #GenAI gold rush given its strength in data infrastructure bringing compute, storage & most importantly 😎#connectivity together #Marvell2023IADay pic.twitter.com/Gbe4h2Z8cZ
— Will Townsend (@WillTownTech) December 5, 2023
The acceleration of #network switching is mindblowing 🤯 & @MarvellTech has been responsible for pushing it to new heights over the last two decades #Marvell2023IADay pic.twitter.com/FbTFcVsBDr
— Will Townsend (@WillTownTech) December 5, 2023
Google Gemini AI 1.0 and New TPU
It's good to see Google diversify its processor supply chain. AWS, Meta, and Microsoft recently released some of its newer DC AI chips.
There are not a ton of details on v5p, but at first glance, it's not setting many spec records. It is outperforming prior TPUs, which is… https://t.co/ez9NeaDOLa
— Patrick Moorhead (@PatrickMoorhead) December 6, 2023
As for Gemini 1.0:
-should improve Bard if the demos translate to reality
-nice to see the good-better-best and the good fits on-device
– smartphone industry sigh of relief for on-device; were many concerned companies out there asking for solutions
-wondering if @googlecloud… https://t.co/HbwBC8TB3K pic.twitter.com/lwhPasY1lx— Patrick Moorhead (@PatrickMoorhead) December 6, 2023
A powerful 6-minute demo to better understand the new multimodal AI capabilities of #GeminiAI.
The rapid advancement of #GenAI to do things as well or better than humans is profound.
Can you imagine where we will be one year from now? $GOOG https://t.co/HjX8fwUxmo
— Daniel Newman (@danielnewmanUV) December 6, 2023
The breaking #AI news continues to come fast and furious. Today Google is in focus.
While we are awaiting AI updates from @AMD today, @Google has launched its newest model #GeminiAI.
This will enable a new and improved Bard experience by utilizing the mid-range Gemini model,… https://t.co/c7b1cUWGT1
— Daniel Newman (@danielnewmanUV) December 6, 2023
AMD Advancing AI Event
https://cnbcmediahub.com/item/f05c1cf3-dcaa-4da2-a6b0-5ee00b9336d3
CEO @LisaSu kicks off it's AI event:
-AI history, verticals
-key to everything is GAI (yep)
-revised TAM to $400B (DC accelerators)
– strategy E2E, 'broad, open, co-in ovation' (last 2 AMD historical strengths)
-'GPU at the center' (versus ASICs)$AMD ##AMDAI pic.twitter.com/jkfXmAYD4R— Patrick Moorhead (@PatrickMoorhead) December 6, 2023
The AMD MI300X specs & performance versus @NVIDIA H100 per AMD.
Wow @AMD must have done some work to make this a training solution. And inference perf is big, big, big.
Partners seeing this?$AMD $NVIDIA $AMDAI pic.twitter.com/4TU8fv0Bfi
— Patrick Moorhead (@PatrickMoorhead) December 6, 2023
Microsoft CTO @kevin_scott on @AMD:
– lots of custom silicon history (@Xbox)
– then Epyc
– past 2 years, working together on AI
– supports $MSFT strategic AI objectives
– enthusiastic on MI300x, in preview TODAY $AMD #AMDAI pic.twitter.com/o9O5nAaxF8— Patrick Moorhead (@PatrickMoorhead) December 6, 2023
Now, going from card competitive performance to "platform" AMD MI300X Platform vs NVIDIA H100 HGX. What do Partners say? $AMD $NVDA pic.twitter.com/lqXHuLjCIh
— Patrick Moorhead (@PatrickMoorhead) December 6, 2023
.@OracleCloud @Karan_Batta on stage with @LisaSu.
– MI300X on bare metal coming
– working with Data bricks and Mosaic
– no dates mentioned $AMD $ORCL pic.twitter.com/SYL5GxL48c— Patrick Moorhead (@PatrickMoorhead) December 6, 2023
AMD has fielded some competitive AI hardware previously, but software has held it back in AI. ROCm did well in HPC with massive installs, and has been developed for over a decade, but HPC isn't all AI.
ROCm 6 *appears* to change this with a focus on LLMs..
Wow MI300 supported… pic.twitter.com/CzOJiOW7Pu
— Patrick Moorhead (@PatrickMoorhead) December 6, 2023
Now @Meta …
-Epyc since 2019
-Instinct since 2020
-ROCm benchmarking, getting better
-big Pytorch work
– adding MI300X into data enters 🔥
– fastest design to deployment in its history (wow)
– big perf gains on LLAMA models on ROCm$AMD pic.twitter.com/D1Zbzhodo9— Patrick Moorhead (@PatrickMoorhead) December 6, 2023
Now @DellTech with President, GIS (DC and edge), Arthur Lewis:
-PowerEdge adding MI300X
-8 cards, some wild bandwidth
-Using Infinity Fabric too (interesting)
– 'focus on out of the box LM simplicity'
-'Open for business, taking orders now' (crowd cheers) $DELL pic.twitter.com/4LBBwiQ2Wt— Patrick Moorhead (@PatrickMoorhead) December 6, 2023
Adding competition in GPU Ethernet for AI, low latency networking. Was wondering how that hole got plugged, or at least the strategy to do that. Via UltraEthernet. RDMA at scale. Mixing vendors.
AMD's Forrest Norrod chatting with:
– @AristaNetworks BoD Andy Bechtolsheim… pic.twitter.com/66HEP6ydhn— Patrick Moorhead (@PatrickMoorhead) December 6, 2023
Ok didn't expect this from AMD on on-device, PC software. It's 1.0 FWIW.
Hawk Point 16 NPU TOPS. 39 system total. Steikes me as similar phase 1 AI strategy as Intel.
Combined TOPS harder to program, less efficient than giant NPU for FM inference, but it's a start. pic.twitter.com/XrPExET6NG
— Patrick Moorhead (@PatrickMoorhead) December 6, 2023
And for Phase 2 of AI PC, @LisaSu quickly shows a roadmap. @microsoft's Pavan Davuliri doing a good job discussing the Windows strategy for AI. Cloud, hybrid, on-device, all connected. All about Copilot. Many phases. $MSFT $AMD https://t.co/UTffsFHzbt pic.twitter.com/8ge8W1MlpC
— Patrick Moorhead (@PatrickMoorhead) December 6, 2023
Now @Lenovodc @KirkSkaugen:
– #1 HPC in Top500
– E2E corporate strategy; public, private, and personal AI from DC to DC edge to PC
– hybrid approach, bringing AI to the data (edge) SE 455
– key is making it simple and CHOICE
– inference will become the dominant workload (yup)
-… pic.twitter.com/Iz2rrIKJbB— Patrick Moorhead (@PatrickMoorhead) December 6, 2023
The TAM for data center AI accelerators has gone from an expected $150 Billion in 2027 to $400 Billion in 2027.
This is why I continue to say it isn’t winner take all. And AMD and others will have a significant role to play. $AMD #GenAI #AI pic.twitter.com/F8iMyFNbkX
— Daniel Newman (@danielnewmanUV) December 6, 2023
The @NVIDIA compare slide for LLM and #GenAI
Of course $NVDA has its newest H200 coming but this is based upon what is out in the market today.
Impressive comparisons and benchmarks. $AMD pic.twitter.com/lsezvLRiog
— Daniel Newman (@danielnewmanUV) December 6, 2023
Run more models on a single server.
More compete metrics and an important note about improved Capex.
At some point, $$$ will matter and more offers in the market should play a role in bringing up supply and down cost for GPU/ AI accelerators.#GenAI $AMD $NVDA pic.twitter.com/zLO0oriyzz
— Daniel Newman (@danielnewmanUV) December 6, 2023
Oracle also in the boat to support MI300x and support it as part of its #GenAI service.
Mentioned the ROCm ecosystem as important. Expecting to hear more about that. pic.twitter.com/uKUtGQli5A— Daniel Newman (@danielnewmanUV) December 6, 2023
Microsoft detailing that it will have the new @AMD MI300x VM Instances available today.
TLDR: partner ecosystem matters a lot.
Look for the key partners that align with $AMD to get a sense of the demand and interest in broadened the supplier ecosystem. $MSFT #AI #GenAI pic.twitter.com/Iflo6zMSkd
— Daniel Newman (@danielnewmanUV) December 6, 2023
Now we move to software.
The availability of competitive AI chips like MI300X is important but software has been the sticky point for a long time.
Being able to win developers and deliver performant #AI using open-source tools versus CUDA will play a big part in adoption. $AMD pic.twitter.com/lqwJvdyGaL— Daniel Newman (@danielnewmanUV) December 6, 2023
Broadcom Q4 FY23 Earnings
https://cnbcmediahub.com/item/f05c1cf3-dcaa-4da2-a6b0-5ee00b9336d3
https://investors.broadcom.com/news-releases/news-release-details/broadcom-inc-announces-fourth-quarter-and-fiscal-year-2023
AI Alliance Led by IBM and Meta
The AI Alliance spearheaded by @IBM and @Meta isn’t a surprise, but I’m surprised it took this long for one to be created.
At AMD, we used to call this the “virtual gorilla”.
The project areas are spot on and I urge you to read between the lines.
It’s an impressive first… pic.twitter.com/5so90Uw6iS
— Patrick Moorhead (@PatrickMoorhead) December 5, 2023
Congratulations to @IBM and @Meta in launching the "AI Alliance" this morning.
The development of a broad ecosystem of contributors and implementers of a technology as transformational as #AI requires collaboration and consideration that has to not only put commercial interests,… pic.twitter.com/QNSvgoo0uc
— Daniel Newman (@danielnewmanUV) December 5, 2023
Disclaimer: This show is for information and entertainment purposes only. While we will discuss publicly traded companies on this show. The contents of this show should not be taken as investment advice.
cisco academie
