Friday, February 23rd 2024
NVIDIA Expects Upcoming Blackwell GPU Generation to be Capacity-Constrained
NVIDIA is anticipating supply issues for its upcoming Blackwell GPUs, which are expected to significantly improve artificial intelligence compute performance. "We expect our next-generation products to be supply constrained as demand far exceeds supply," said Colette Kress, NVIDIA's chief financial officer, during a recent earnings call. This prediction of scarcity comes just days after an analyst noted much shorter lead times for NVIDIA's current flagship Hopper-based H100 GPUs tailored to AI and high-performance computing. The eagerly anticipated Blackwell architecture and B100 GPUs built on it promise major leaps in capability—likely spurring NVIDIA's existing customers to place pre-orders already. With skyrocketing demand in the red-hot AI compute market, NVIDIA appears poised to capitalize on the insatiable appetite for ever-greater processing power.
However, the scarcity of NVIDIA's products may present an excellent opportunity for significant rivals like AMD and Intel. If both companies can offer a product that could beat NVIDIA's current H100 and provide a suitable software stack, customers would be willing to jump to their offerings and not wait many months for the anticipated high lead times. Intel is preparing the next-generation Gaudi 3 and working on the Falcon Shores accelerator for AI and HPC. AMD is shipping its Instinct MI300 accelerator, a highly competitive product, while already working on the MI400 generation. It remains to be seen if AI companies will begin the adoption of non-NVIDIA hardware or if they will remain a loyal customer and agree to the higher lead times of the new Blackwell generation. However, capacity constrain should only be a problem at launch, where the availability should improve from quarter to quarter. As TSMC improves CoWoS packaging capacity and 3 nm production, NVIDIA's allocation of the 3 nm wafers will likely improve over time as the company moves its priority from H100 to B100.
Sources:
Q4 Earning Call Transcript, via Tom's Hardware
However, the scarcity of NVIDIA's products may present an excellent opportunity for significant rivals like AMD and Intel. If both companies can offer a product that could beat NVIDIA's current H100 and provide a suitable software stack, customers would be willing to jump to their offerings and not wait many months for the anticipated high lead times. Intel is preparing the next-generation Gaudi 3 and working on the Falcon Shores accelerator for AI and HPC. AMD is shipping its Instinct MI300 accelerator, a highly competitive product, while already working on the MI400 generation. It remains to be seen if AI companies will begin the adoption of non-NVIDIA hardware or if they will remain a loyal customer and agree to the higher lead times of the new Blackwell generation. However, capacity constrain should only be a problem at launch, where the availability should improve from quarter to quarter. As TSMC improves CoWoS packaging capacity and 3 nm production, NVIDIA's allocation of the 3 nm wafers will likely improve over time as the company moves its priority from H100 to B100.
64 Comments on NVIDIA Expects Upcoming Blackwell GPU Generation to be Capacity-Constrained
basicly the game is don't accually tell the public what we CAN do. tell them what we WILL Do then we can keep the price higher for longer
AMD, NVIDIA, Apple, Qualcomm... and now even Intel, who has a foundry business of their own and has always been a point of pride for the company, they're all using TSMC and more specifically, demand the N3 node they're using to build these GPUs. Supply constraints are a real concern, IMHO, and I wouldn't be surprised to see consumer-grade parts stick to a less advanced node such as N4 used on Ada simply to ensure that the supply situation remains stable.
They are selling every single graphics card they make, you can bet on that one. There's no "nGreedia" at work here.
The question is at what cost?
Well, in general yes, it is either sooner or later, or how much the retailer on-shelves duration stay lasts.
1. You can't know whether the supply is not artificially limited in order to keep the prices high.
2. German Mindfactory has sold so far approximately 8,420 pcs. RTX 4090, which since October 2022 means around 500 pcs. a month.
I guess (because there is no statistics data per country) Germany is one of the top largest GPU markets and a large part of the graphics production goes there.
Question - do you really think that Nvidia can make only 500 pcs. a month for Germany alone?
To put it in a context - 500 working AD102 dies could be made by 7 wafers.
I am no amd fanboy but its pretty clear that amds failure in the market has setup nvidia to basicly run the entire show and thats something that should be heavily scrutinised
something about a inch given and 40 billion miles taken ...
AMD is reliant on TSMC all the same, and the decision to place Radeon in a lower priority was a decision of their own - margins on EPYC and Threadripper products are insane, and Ryzen makes it up to the lower margins in sheer volume. Said margins aren't attainable on their graphics products, even in the workstation segment.
of course I know TSMC makes chips for _everything_
but I remind you that the new fabs in the states don't exist because TSMC wants them too they exist per us-gubberment mandate and gubberment-monies without either of those mandates.
tsmc would be far less inclined to build these new fabs so I would hardly call them 'desperate' are new fabs good for them in the long run? yep. in the next year or even 3 Nope.
and then there is the problem of who you are going to get to staff and run the US Fabs because there isn't enough labor force in that sector stateside ... yet
and AMD is going to miss the AI-supertrain to money town (again) and remain forever the underdog because the people in the gpu division are clueless
nobody disputes ryzen(anymore) but not relavent to nvidia/partners playing pin the tail on the price-fixing donkey
Intel is comming up fast but there idiotic choice to kill larrabee years ago is biting them hard
Nvidia basicly has a Run of the entire Gpu/AI market completely unchallanged in the DC for AI which is where all the money is going to be.
you put a giant bag-o-dollars in front a company they are going to make a grab at it and they WILL break every rule and step on every toe to get it
which is why a hawk eye needs to be keept on every move and statement they make as well as a eye on anybody else that stands to profit from the situation
cat being away mice gonna play. childern gonna break shit while the parents are out of town
On the other hand, TSMC reported back in 2022 that its monthly 5nm wafer capacity is 150,000 wafers. www.digitimes.com/news/a20220323PD215.html
And yes, Germany is not a small market. A small market is Africa, South America, Eastern Europe.
I guess in my country no one has bought an RTX 4090.
And no, RX 7800 XT is not doing that good. It has sold under 10,000 since September last year. That's ~1500 pcs. a month ~ 20-25 wafers.
Spoilered because of a bit grey area regarding no politically talk but it's in good faith and relevant to the conversation. No partisan talk, mods feel free to remove if its too much thanks. Assuming 100% yield and disregarding as many variables as possible... but no, really. Mindfactory would pale even in comparison with Amazon let alone the OEM market. Very easy to get a skewed perspective from data like this.
500 dies is 7 wafers. Amazon is a bad retailer. Don't buy there.
What is the best website to buy pc components in Germany?
germany/comments/7tlfsw