As previously noted, AI is where its at for Nvidia and AMD.
Quote
Nvidia sold half a million H100 AI GPUs in Q3 thanks to Meta, Facebook — lead times stretch up to 52 weeks: Omdia
Having earned $14.5 billion on datacenter hardware in the third quarter of fiscal 2024, ... Nvidia sold nearly half a million A100 and H100 GPUs, and demand for these products is so high that the lead time of H100-based servers is from 36 to 52 weeks.
it should be noted that virtually all companies that purchase Nvidia's H100 GPUs in large quantities are also developing [their own] custom silicon for AI, HPC, and video workloads. As a result, their purchases of Nvidia hardware will likely decline over time as [users] transit to their own chips.
As previously stated, I think AI will make "advanced/high-end" human "thinking/problem-solving/etc" obsolete in less than 100 years -- probably less than 50.
Possibly partly due to Nvidia 4090s being diverted from the Gaming Market by the thousands -- to become AI GPUs:
Quote
Nvidia RTX 4090 GPU Pricing Is Officially Ludicrous
The time to buy was about six months ago, apparently.
Looking at Amazon, for example, the only card available is the Zotac AMP Extreme card, which goes for $2,267—almost $700 over MSRP.
Newegg has four cards in stock, the least expensive being the same one from Amazon—Zotac AMP Extreme—for $2,249.
I checked. When I checked, the cheapest Newegg RTX 4090 was $2249 (as in the article). In a quick check, the cheapest Zotac AMP Extreme at Amazon was $2174.
I expect the prices can change hour to hour -- as Amazon's has.
More AI news appearing. Nvidia is big on GPUs for AI. AMD is big on high end CPUs for AI -- and working up the AI GPU ladder.
Thus, as previously noted, it is rumored/leaked that there will be fewer and higher priced Nvidia and AMD Gaming GPUs next year.
Quote
NVIDIA Sold “Half a Million” AI GPUs In Q3 2023, Courtesy of Demand From Major Tech Firms
sold nearly a whopping "half a million" AI GPUs in the quarter, mainly consisting of the Hopper H100 and the Ampere A100. The sale volumes are a sight to see in the industry, and it has led NVIDIA to a position where it has exerted dominance in the markets. NVIDIA's plans to sell millions of GPUs in 2024 seem achievable,
it looks like Meta and Microsoft were on top of the list, securing 150,000 units of NVIDIA's current arsenal of AI GPUs.
NVIDIA has also seen a decent share of orders from Chinese markets, amounting to around 30% of the company's orders in Q3 2023
NVIDIA CEO Jensen Huang Believes That The US Needs “Decades” For Chip Supply Chain Independence
[Huang quote]: "We are somewhere between a decade or two decades away from supply chain independence, as I mentioned earlier, our systems comes from 35,000 parts, and eight of them comes from TSMC. Supply chain independence is going to be challenging but we are going to try it and we should endevaour it but total independence of supply chain isn't practical for a decade or two."
Through incentives like the CHIPS Act, the US poured somewhere around $52 billion into the markets in the form of subsidies, tax credits, and covering up costs of R&D. It was seen as one of the biggest packages offered to tech firms ever, which ultimately prompted chip companies like TSMC and Intel to capitalize on it through setting up their foundries in the US. .. neither of the above-mentioned developments has reached a conclusive stage yet.
AMD Opens The Company’s Largest R&D Center In India To Accelerate Development of Next-Gen CPUs, GPUs & SOCs
[Per AMD] "We are pleased to inaugurate our largest global design center in Bengaluru today. This investment strengthens our relationship with India and showcases our confidence in the exceptional engineering talent that the country has to offer."
The Technostar center is built upon 500,000 square feet of land, which not only makes it AMD's largest R&D center, but both parties plan on 3,000 engineers in the coming years, acting as a catalyst for regional technological developments. The campus will be responsible for advancements within CPUs, GPUs, adaptive SoCs, and FPGAs for both personal computers and data centers. Moreover, the center will consist of a hefty portion of "Indian" professionals and workforce, hence showing a contribution to the "Made in India" policy
Quote
According to the UN estimates, India has overtaken China in having the largest population in the world with population of 1,425,775,850 at the end of April 2023.
The USA Population is nominally 336 million. Tiny by comparison to India and China -- whose total populations are approaching 3 Billion people (not there yet).
AMD's Current & Future Ryzen CPUs To Get Precise & Detailed Specifications Listings As Hybrid Becomes The Norm
AMD has previously promised to make it easier for users to distinguish between Ryzen AI and non-Ryzen AI PCs through new branding.
It should be pointed out that AMD's hybrid strategy is so far unlike Intel's hybrid strategy. .. the fundamental [AMD] ISA (Instruction Set Architecture) is pretty much the same [across CPUs] with the only differences in clock speeds, cache, and power targets.
Intel's hybrid core approach is way more unique and uses two different core architectures that cannot be compared to each other
AMD made it clear that it won't follow Intel's P-Core/E-Core design for its hybrid future.
Because AMD uses similar instruction-sets/cores throughout a CPU and Intel uses different instruction-sets/cores in its CPUs, its easier/cheaper to design efficient-software for AMD CPUs.
TSMC's Roadmap Reveals Progressing Of Cutting-Edge Semiconductor Processes: Development On 1nm, Eyeing Over 1 Trillion Transistors By 2030
AMD has been leveraging TSMC's chiplet designs for its latest consumer, data center, and now the most recent MI300 accelerator chips. Intel also released its Meteor Lake chips which are the blue team's first chiplet-design for consumer platforms, hinting at the fact that chiplets are the future, with TSMC one step ahead. Intel itself uses chips fabricated on TSMC's process technologies to power Meteor Lake. The company projects 3D Hetero Integration to reach a whopping "one trillion transistors" by 2030.
AMD & Intel Motherboard Partners Eye Q3 2024 As Next-Major CPU Platform Refresh: 700-Series For AM5 & 800-Series For LGA 1851
Both AMD and Intel are said to introduce their latest chipsets which motherboard makers will utilize in next-gen products but AMD will be flexible based on the choices Intel makes since they aren't in a hurry
On the AMD camp, Team Red and its motherboard partners will introduce the new 700-series PCH family based on the same AM5 socket. AMD has committed to a 2025+ plan for its AM5 socket and the introduction of a new PCH doesn't mean that existing motherboards will get any less support than what they are getting now. The new PCH will be designed to offer new and improved features but as we have seen with AM4, the older chipsets will still retain compatibility with newer CPUs and may also get support for the same features through BIOS updates by motherboard manufacturers.
So, AMD will NOT be making current Motherboards obsolete. Continued support is planned.
I don't post much about AMD Threadripper CPUs because they are not for gaming. However, they are a big part of AMD industrial-application successes.
So, an older article just for comparison to gaming CPUs and completeness:
Quote
AMD Ryzen Threadripper PRO 7000WX CPUs Come In 6 Flavors, All Zen 4 Powered For Multi-Threading Insane Performance
The AMD Ryzen Threadripper PRO 7000WX CPU family will be based on the Zen 4 core architecture and feature six SKUs. All four SKUs are designed to offer record-shattering performance & bleeding edge efficiency.
Starting at the top, we have the AMD Ryzen Threadripper PRO 7995WX CPU which will offer a total of 96 cores, 192 threads, and up to 384 MB of L3 cache. This chip will offer a 2.50 GHz base clock & a boost clock of up to 5.15 GHz. This gargantuan of a chip will have 50% more cores/threads versus the Zen 3-based 5995WX while running at 14% higher clock speeds at a 25% higher TDP (350 vs 280 Watts).
Next up, we have the AMD Threadripper PRO 7985WX, 7975WX, 7965WX, featuring 64/128, 32/64 & 24/48 configurations. While the 7985WX runs at a base clock of 3.20 GHz, the other two chips will be operating at 4.0 - 4.20 GHz at the base frequency while all chips boost up to 5.35 GHz. The cache counts are maintained at 256 MB, 128 MB, and 96 MB, respectively.
Relevant to Chip Production in the USA by AMD, Intel, Nvidia and others:
Quote
Semiconductor industry proposes new 'Chipmaker's Visa' for H1B program — program would address extreme talent shortages in chipmaking industry
U.S. semiconductor industry needs more workers.
The U.S. semiconductor industry is facing a big talent problem: it will be short of 67,000 employees by 2030, according to estimates by the Semiconductor Industry Association (SIA), a lobbying group for the sector. Efforts to develop local talent in the U.S. are ongoing, but there is still a big gap, which is why the industry depends on engineers, computer scientists, and technicians from abroad.
However, the U.S. H-1B Visa system is making it tough to bring in and keep these workers, so the U.S. chip industry is calling on the U.S. government to rethink it, .. One of those options is a new type of visa specifically for the semiconductor industry.
Although the USA population is increasing, the part of the population with the innate ability to do Science and Engineering tasks is not reproducing at "replacement rate". That (and other factors) mean that each year there are fewer fully qualified USA born Scientists and Engineers available to USA industry -- but, every year, more are needed . Thus, every year, there are more and more USA jobs available for qualified Foreign Born Scientists and Engineers.
Steady Optimizations & Development Have Put The RADV Vulkan Way Ahead of AMD's "AMDVLK" Drivers, Up To 200% Ray-Tracing Gain With Radeon GPUs
Phoronix reports that the new update to the MESA 24.0 was in the works for almost a month by the developer Friedrich Vock, and while the change isn't that significant in terms of the work put into it, it is disclosed that developers have optimized the conversion of 1D ray launches to 2D so that now more threads in a GPU are utilized, resulting in enhanced performance.
The Frontier Supercomputer Sets New Records In The Space of LLM Training, Courtesy of AMD's EPYC CPUs & Instinct GPUs
The Frontier supercomputer is the world's leading supercomputer and the only Exascale machine that is currently operating. This machine is powered by AMD's EPYC & Instinct hardware which not only offers the top HPC performance but is also the 2nd most efficient supercomputer on the planet. A submission report on Arxiv by individuals has revealed that the Frontier supercomputer has reached the ability to train one trillion parameters through "hyperparameter tuning", setting a new industry benchmark.
AMD Unveils Ryzen 8000G Series Processors: Zen 4 APUs For Desktop with Ryzen AI
Starting with the top tier SKU, the Ryzen 7 8700G, it is based on AMD's Pheonix mobile silicon and, as such, benefits from eight fully-fledged Zen 4 cores driving sixteen threads (8C/16T). At the top end, it has a 5.1 GHz turbo frequency with a 4.2 GHz base frequency. Encompassing the true nature of an APU for AMD's AM5 desktop platform, it includes the Radeon 780M integrated graphics, which is RDNA3-based and supports AMD's latest features, such as Hyper-RX and Fluid Motion Frames.
NVIDIA Secures a Huge AI GPU Order From India, Worth Over Half A Billion Dollars
Yotto's CEO and co-founder Sunil Gupta revealed that the new order includes 16,000 units of NVIDIA's H100 and GH200 AI GPUs, which are expected to be delivered by March 2025. Moreover, the firm has previously placed an order for 16,000 NVIDIA H100 GPUs, also due for delivery by July. Hence, a rough estimate suggests that total order valuation has indeed exceeded a billion-dollar mark, placing Yotto in the list of NVIDIA's "esteemed" clients. Moreover, NVIDIA's CEO Jensen Huang has himself expressed the immense potential held in Indian markets and has predicted it to be one of the largest in the world moving ahead.
Quote
In 2023, India overtook China as the country with the largest population in the world, with more than 1.43 billion people. China now has the second largest population in the world, with just above 1.4 billion inhabitants.
Quote
As of 1 January 2023, the population of the EU is slightly over 448 million people.
Quote
U.S. Population Estimated at 335,893,238 on Jan. 1, 2024
Something to "think about" when considering buying one's next PC.
Quote
CES 2024 Showed Us How AI PCs Will Dominate Next-Gen Markets, Revolutionizing The Way We See PCs
With the influx of AI in almost every major tech segment in the world, it seems like artificial intelligence is the way to go moving forward. However, in recent times, the term "AI PC" has taken a new high in terms of its usage by tech giants, and based on what we have seen up till now, companies are adjusting their product portfolios in a way to incorporate the power of AI, especially in the case of personal computers.
AMD removes Taiwan branding from CPUs, says change wasn't made to appease China
"AMD removed the country of diffusion from all new CPU and APU products in 2023 to align with the product marking process for our other products," an AMD spokesperson told Tom's Hardware, indicating that the company will no longer mark (silkscreen) any of its consumer or enterprise processors with the country where the silicon die inside the chip was originally fabricated (diffused). In the case of all of AMD's leading-edge products, those silicon dies are created in Taiwan at TSMC.
Notably, AMD still marks the processors as "Made in Malaysia," which signifies where the processor die was integrated into the final chip package.
TSMC's average wafer price jumped 22% in one year — nearly all semiconductor industry growth now comes from more expensive products, not higher production volumes
the average selling price (ASP) of a TSMC 300-mm wafer increased to $6,611 in the fourth quarter, an increase of 22% in a single year, as noticed by analyst Dan Nystedt, who attributes this increase to the ramp of TSMC's N3 (3nm-class) process technology. Bernstein Research's Stacy Rasgon also points out that most semiconductor industry growth now comes from increased pricing, and not an increase in the number of processors shipped.
A partly misleading article. Part of the price increase is because the Wafers contain more "chips" and "transistors" (i.e. more product costs more).
The RX 7600 XT is a niche product. Buy one only after careful consideration. Most folks would do better elsewhere.
Quote
AMD Radeon RX 7600 XT review: double the memory and higher clocks, still Navi 33
RX 6700 XT still provides stiff competition at the same price.
As we saw with the RTX 4060 Ti 16GB, the RX 7600 XT ends up targeting a very specific niche. Yes, it's faster than the RX 7600, but it also costs more. For gaming purposes, there are a lot of titles where the extra VRAM isn't even that useful, and then it's just down to clock speeds. It's not necessarily a worse proposition than the RX 7600, but then we've opined that that [RX 7600] was one of the most disappointing AMD GPUs of all time.
