Nvidia shows signs in [2023]

[Remij]

The PC GPU market will be what we dictate it is. The demand is there, it's just the pricing that isn't.

They will be forced to drop prices if people aren't buying. Problem is - we really need a large drop in sales, cuz the extreme margin increase that Nvidia is getting from these GPU's means they can absorb some drop in sales.

The 4080 is the equivalent of the 3070 in Ampere equivalence. It's a cut down upper midrange die for $1200. It's already here. You know this, we all know this. No amount of parroting your dishonest claims about this will change this. You've already admitted that you'd call a 750Ti a 'high end' GPU if Nvidia priced it at $1000, so your opinion is frankly just asinine.

No... it wont. The less dependent they become on gaming, the less ability we have to dictate anything.

Nvidia will divert resources into the market which is making them the most money. You say the demand is there.... and you'll see what Nvidia will charge consumers to fulfill that demand... since there's greater demand for their chips in AI data centers where their customers have no problem paying much more.

Suddenly prices will have gone way up... there's less product on the shelves... AMD wont be competing on the enthusiast/high end... so where are you going to go?

 

[Seanspeed]

No... it wont. The less dependent they become on gaming, the less ability we have to dictate anything.

Nvidia will divert resources into the market which is making them the most money. You say the demand is there.... and you'll see what Nvidia will charge consumers to fulfill that demand... since there's greater demand for their chips in AI data centers where their customers have no problem paying much more.

Suddenly prices will have gone way up... there's less product on the shelves... AMD wont be competing on the enthusiast/high end... so where are you going to go?

Again, Nvidia cant produce more H100's by cutting production of Lovelace GPU's. This just doesn't help them when wafer production is not the bottleneck.

And this idea that 'Nvidia will divert resources into the market which is making them the most money' is a claim we've been hearing for like the past 7-10 years. Divert some percentage of resources yes, but not all. Given that companies like to grow, the general smarter idea is to make money off of multiple large sources of income if possible. Diversification is also a critical hedge for major companies. It's generally not ideal to have all your money coming only from one market, especially in the tech business where demands can swing hard.

 

[TopSpoiler]

Nvidia's response to MI300:

NVIDIA Unveils Updated GH200 'Grace Hopper' Superchip with HBM3e Memory, Shipping in Q2'2024

www.anandtech.com

 

[DegustatoR]

I mean it is in the middle of the Ada Lovelace product stack. Considering what NVIDIA is currently offering, it is midrange in both price and performance.

"The middle of the stack" isn't "mid-range". The latter is a clearly defined pricing category, not something "in the middle" of something.
4070Ti is a $900 card with equivalent performance to a 3090Ti which wasn't mid-range and thus 4070Ti isn't "mid-range" in any way or form.
People wanting things to be cheap doesn't make things into what they want.

And I do feel weird even discussing this, again.

 

[Malo]

Ampere 3070 for $500 is apparently no longer a mid-range because the 2080Ti for $1200 it was equivalent to was an enthusiast halo card. Which means $500 3070 was the new enthusiast range, making $200-300 mid-range? Turing 2070 at $500 being faster than 980Ti but no one ever considered 980Ti a mid-range, and no one ever considered 2070 an enthusiast halo card. So if 2070 is now an enthusiast halo GPU historically, what does that make the 2080Ti?

And thus that makes 4000 series a completely new set of ranges, from godlike to extreme premium godlike. I love all these moving goalposts to suit the narrative, because the only thing that matters is what Nvidia says the price is.

 

[DegustatoR]

Ampere 3070 for $500 is apparently no longer a mid-range

It was although $500 is the upper border of "mid-range" so any card on that point can be considered lowest high end as well.

The point is that you could call a $900 card "mid-range" if you would have had a different card with similar performance at <$500 alongside it or at some earlier point in time. This is literally the only way such an argument can be made.

 

[techuse]

By every metric other than price the 4070ti is a midrange GPU. Performance relative to its contemporaries, die size, power consumption etc. All are indicative of a mid range GPU.

 

[DegustatoR]

By every metric other than price the 4070ti is a midrange GPU.

So by every metric other than the one which determines if something is "mid-range" or not then. Gotcha.

 

[homerdog]

By every metric other than price the 4070ti is a midrange GPU. Performance relative to its contemporaries, die size, power consumption etc. All are indicative of a mid range GPU.

Even on price it is midrange. It is in the middle of the range.

 

[techuse]

So by every metric other than the one which determines if something is "mid-range" or not then. Gotcha.

I disagree. Pricing a product high doesn't make it high end.

 

[DavidGraham]

China's tech giants (TikTok, Baidu, Tenecent, Alibaba) have placed an order of 5 billion dollars worth of NVIDIA GPUs, to be delivered this year.

https://twitter.com/i/web/status/1689395774579253248

And on other news, chip packaging giant Amkor will double CoWoS-like 2.5D capacity next year for 5 big clients, with over 80% for Nvidia.

https://twitter.com/i/web/status/1689448507747221504

 

[trinibwoy]

What do those capacity numbers refer to? Wafers per day/week/month?

 

[GhostofWar]

China's tech giants (TikTok, Baidu, Tenecent, Alibaba) have placed an order of 5 billion dollars worth of NVIDIA GPUs, to be delivered this year.

Isn't there some block from the us gov on nvidia selling its data centre gpus to china? I didn't really look into this when it was doing the rounds so not sure how it plays into this, but from those tweets I guess it's a none issue.

 

[DavidGraham]

What do those capacity numbers refer to? Wafers per day/week/month?

Per month.

https://twitter.com/i/web/status/1689513350114484224

Isn't there some block from the us gov on nvidia selling its data centre gpus to china? I didn't really look into this when it was doing the rounds so not sure how it plays into this, but from those tweets I guess it's a none issue.

To get around the block, NVIDIA is selling some castrated versions of A100 and H100 to the Chinese market, they are called A800 and H800, they have less memory and NVLink bandwidth than the original versions.

 

[DegustatoR]

I disagree. Pricing a product high doesn't make it high end.

Pricing is what determines at which market segment a product is targeted.
If it has enough qualities to compete in that segment then it is a product from that segment.
If you *think* that it doesn't then it's your own problem. It is called "unrealistic expectations".

 

[Seanspeed]

Pricing is what determines at which market segment a product is targeted.
If it has enough qualities to compete in that segment then it is a product from that segment.
If you *think* that it doesn't then it's your own problem. It is called "unrealistic expectations".

Nvidia want people who are used to spending $300 on a decent midrange GPU to now spend $500+ for that same kind of product. And so on throughout the range. They want to normalize a huge increase in average buying price for a GPU across the board. They want us to start thinking that $300 is lower end, $500-600 is midrange, etc.

They are trying to see what they can get away with. It is up to us to be stubborn and not accept this so that such a normalization in the shifting of 'market segments' does not happen. And it's why many of us get quite frustrated seeing certain people trying to defend such a drastic shift...

 

[DegustatoR]

Nvidia want people who are used to spending $300 on a decent midrange GPU to now spend $500+ for that same kind of product.

So to buy now something with a similar performance to an older decent midrange $300 GPU (like say RX590 with its $280 MSRP) I have to pay $500+? Which GPU do you have in mind I wonder?

They want to normalize a huge increase in average buying price for a GPU across the board.

They want to sell products which they make. If the only way to improve perf/price left is to extend the upper limit of high end then what would you prefer them to do:
a) Stop at $700-1000 point which was the upper limit of high end previously. Means nothing above 4070Ti with ~3090Ti performance.
b) Extend the product range above $1000 with 4080 and 4090 SKUs for those who can and are willing to pay as much for such products.
There are no other options because the prices are dictated by production chain. The only solid other option is to not make new GPUs at all - something which AMD has been doing with N32 for almost a year now, selling N21 in its place.

They want us to start thinking that $300 is lower end, $500-600 is midrange, etc.

No, they don't. It is you who is thinking this way, not them. They make midrange products just like they did - 4060, 4060Ti, likely a 4050 down the road, old 3050 is still in production too although this may have drifted into low end already.

It is up to us to be stubborn and not accept this

Okay, so you don't accept this. What is happening because of that? The two (three if we count the one where Nvidia does nothing new at all for gaming) options you can choose are listed above.

 

[trinibwoy]

Nvidia want people who are used to spending $300 on a decent midrange GPU to now spend $500+ for that same kind of product. And so on throughout the range. They want to normalize a huge increase in average buying price for a GPU across the board. They want us to start thinking that $300 is lower end, $500-600 is midrange, etc.

They are trying to see what they can get away with. It is up to us to be stubborn and not accept this so that such a normalization in the shifting of 'market segments' does not happen. And it's why many of us get quite frustrated seeing certain people trying to defend such a drastic shift...

What's missing from this debate is a common understanding of "same kind of product". There are a few different ways you can define it.

Some examples

• Lets you play newish games at 1080p / 60 / medium settings
• Only takes up 2 PCIe slots
• Requires only one power connector
• Costs < $300
• Product name ends in "60"
• Chip name ends in "06"
• Is in the middle of the SKU lineup

Everybody seems to be using a different definition of "midrange" which is why the conversation is just going in circles.

For the record my definition is the first one. A midrange card plays games at midrange settings. Feel free to argue about the definition of midrange settings

 

[DegustatoR]

For the record my definition is the first one

You can play games in such resolution on medium settings on a 1060. Or you mean new games? Then it's getting harder but still you can get there with a 4060 for the most part, especially since medium textures will solve the VRAM issues completely. That's a mid-range card from pricing point of view.

 

[techuse]

If Honda arbitrarily doubles the price of a Civic in 2024, is it now a higher class of car?