As per the latest figures, it looks like NVIDIA GPUs are powering the bulk of the supercomputers in the Top500 list with a total of 168 systems while AMD’s CPUs and GPUs power a total of 121 super computers. At the same time, Supercomputers housing AMD and NVIDIA GPU-based accelerators are largely running Intel CPUs which cover around 400 supercomputers and that’s a huge figure & while the number of systems running Intel CPUs are in a clear lead in quantity, AMD actually wins the crown for the fastest supercomputer around in the form of Frontier.
The number one system in the world is an AMD powered system and 4 of the top ten are AMD with NVDA holding 5 of the top ten spots…doc
Very good results from Nvidia tonight. Up $81 or 26.5% after hours. AMD going up in sympathy, up $10.70 or 9.9%
These are 2 of the most promising stocks tackling AI from the hardware side. On the software side I guess 2 of the most promising stocks would be Microsoft and Google. Both up after hours but much more moderately.
When it hit $58, I told my son he should buy back in (he sold at $100) on the downtrend. I called him last night and asked if he saw AMD afterhours price. He didn’t buy back in!
Your purchases in the 90-110 range look good right now.
For me, I worry that Intel will catch back up with AMD on the cpu manufacturing front or that ARM will start to outperform cpu’s. Also, AMD doesn’t so far seem to be making serious inroads on the gpu front vs Nvidia. Next, Nvidia has such a huge lead in AI. Lastly, I hope China leaves Taiwan alone. The increased tension on the technology front between USA and China concerns me. Only time will answer these questions…doc
If China disrupts TSMC, NVidia is badly screwed until things settle down (assuming we’re all alive). But compare Russia to China. But want to take over a neighboring country that at some time in the past they had owned. Russia’s approach is brutal and stupid. Putin’s committing war crimes, killing the citizens that would provide value if Ukraine became a Russian satrap, destroying the foundations of Ukraine’s value.
China knows that invading Taiwan would cost more, maybe even a lot more, in destruction of Taiwan’s value. More than Ukraine, which has substantial natural resources, Taiwan’s value lies in its intellectual properties. Destroy the factories and kill the people, and those factories will need to be rebuilt possibly without the minds that know how to do so. TSMC is setting up to flee Taiwan as its sole base of powers. All the money could easily be gone before China could claim it. To China, I think Taiwan’s valus is political. They stand as a symbol that things are not all rosy in China. Thus, I think China will threaten a lot, as they have for decades, but not actually invade. There is also the possibility that Taiwan has nukes. Some people are very good at keeping secrets.
But back to the main point. NV is most closely tied to Taiwan. AMD is still working with other foundries and investigating more. They’ll hurt, but less than NV. The big winner will be Intel. If China invades, I could see AMD take over the graphics and maybe data center lead and Intel own CPUs until AMD gets fully caught up at GF (many upgrades needed) or Samsung.
If AMD can continue to take market share from intel in the cpu realm and if AMD can continue to take market share from Nvidia, then AMD investors should be in pretty good shape even if AMD doesn’t surpass Nvidia in market cap - but one can hope…doc
A Microsoft spokesman clarified that AMD is ‘not involved in Athena’, but rather that Microsoft is investing to help AMD develop its own chips for AI use. This latest move is part of a broader push from the company to deepen its influence in the chipmaking industry; recent reports also claim Microsoft plans to produce its own silicon to power future systems running Windows 12.
Another quote from the TECHRADAR article above: In any case, I’m cautiously hopeful that AMD stepping into the AI processor game will prevent some of the best graphics cards being subject to shortages as hardware is increasingly guzzled up by AI developers.
AMD is making big strides, forging ahead with new hardware in a wide range of areas, from data centers to handheld gaming consoles (check out the dope-as-all-hell Asus ROG Ally. Now AI is the new frontier, and it’s good to see AMD won’t back down from a fight - but the real winner here will be Microsoft, which isn’t severing its relationship with Nvidia either; it’s playing both sides to come out on top.
The US figured the best way to keep China out of Taiwan is to become #1 in fab technology again. https://pr.tsmc.com/english/news/2977 I don’t think those two fabs will be enough to feed AMD and nVidia, but Intel is also building a lot of new US fabs. Even if Intel can’t* match TSMC at 4 nm or 3 nm, Intel should have enough capacity as a contract fab to feed all the companies who don’t want or need to be on the bleeding edge.
For the record, Intel’s new fabs are planned to be competitive with TSMC. I don’t expect Intel to get ahead of TSMC, but they should get a lot closer by next year.
AMD hired a guy away from IBM. Here is an old article about him from 2012. I think since Lisa worked at IBM before AMD that this will turn out to be a really good hire. Scroll down the page as it is closer to the bottom of this Harvard Business Review article:
This second article is authored by Guido and reflects his business IQ and his understanding of the priorities:
This last one is from the IBM Think blog: https://www.ibm.com/blogs/think/2018/06/ibm-ca/ For companies all over the world, the power of the mainframe is critical to the success of their enterprise transformations. As they transform toward cloud deployment models, many rely on the traditional strengths of the mainframe to maintain secure, reliable, and scalable computing. And for good reason. For example, nearly 80 percent of all enterprise data is managed on the mainframe.
Guido brings someone who has been working in the Enterprise sector, has connections, has the expertise and has a reputation - he’s a known commoditiy because of his time at IBM. Hopefully this turns out to be a huge move by AMD…doc
Today, AMD (NASDAQ: AMD) announced that SAP has chosen AMD EPYC™ processor-powered Google Cloud N2D virtual machines (VMs) to run its cloud ERP delivery operations for RISE with SAP; further increasing adoption of AMD EPYC for cloud-based workloads. As enterprises look toward digital modernization, many are adopting cloud-first architectures to complement their on-premises data centers. AMD, Google Cloud and SAP can help customers achieve their most stringent performance goals while delivering on energy efficiency, scalability and resource utilization needs.
Its good to see AMD continuing to get wins in the Enterprise workspace.
“AMD powers some of the most performant and energy efficient cloud instances available in the world today,” said Dan McNamara, senior vice president and general manager, Server Business Unit, AMD. “As part of our engagement with Google Cloud and SAP, SAP has selected AMD EPYC CPU-powered N2D instances to host its Business Suite enterprise software workloads. This decision by SAP delivers the performance and performance-per-dollar of EPYC processors to customers looking to modernize their data centers and streamline IT spending by accelerating time to value on their enterprise applications.”
AMD still taking advantage of its superior technology to gain more customers in enterprise applications. Now, when will we see this on the bottom line. I can’t wait for AMD to start getting traction in the AI arena as that will catapult AMD sales once again. Until then NVDA is still the 800 lb gorilla in AI…doc
AMD has a strategy where it now has three server CPU lines in the same socket. The mainstream Genoa, the high-core count, Bergamo, and the high-cache Genoa-X. As we have mentioned previously, the big win will be a future part combining reduced base-die cache / high core count chips with L3 cache stacked. We think it is at least a generation away, but Bergamo-Next-X is going to be a fascinating part.
I am wondering if these sku’s will cater to differing needs. For instance, will the high cache cpu be better with AI systems ie faster and more efficient…doc
Listening to a very interesting conversation youtube Moores Law is Dead about the new Sapphire Rapids turbo core. The guest says that they have run the new core and it seems to top out at 3ghz and the brochure says that particular chip needs a 1500 watt power supply. The problems they are seeing with the new Intel chip for workstation might be why there aren’t any benchmarks out yet is what they are thinking…doc
Sort of what I was talking about in my previous post. Basically if you have an application you are going to devote multiple processor chips/sockets to, the magic number is how many threads you can run without risking thrashing. So you run too many threads and back off until the thrashing goes away. Then you take out one or two more threads just so that something else that is running on the system occasionally won’t start thrashing.
Do this on all three choices and pick the one with the lowest price/performance ratio. (Or turn that around and pick the highest performance/price ratio. If you are going to be using a few thousand or so processors to do something like simulating the universe, see my previous post about tuning code to the specific hardware. May still be classified, but I have managed a factor of 1000 to change “Can’t be done with current hardware,” into “Sure here is the thirty or so lines of code that goes into the innermost loop.”
Took advantage of the time reversibility of physics. Same answer either way, but doing it backwards allowed a lot of the needed results to be copied from the previous iteration rather than recomputed each iteration. Oh, and the critical step was clearing an array to all zeros. (Very easy if you are building the hardware.
This seems to be a great site for research on a company. I will paste a link to the information guru focus collects on AMD. It made my head hurt going through all the data they collect on companies…doc
I saw reading the articles that these cpu’s were seen on a Dell and an HP workstation. New threadrippers coming out in the 4th quarter and 2024. I don’t think there is any competing product at this core count at this size…doc