Nvidia shows signs in [2023]

  • Thread starter Deleted member 2197
  • Start date
Status
Not open for further replies.
So where does GF 4200 goes according to your opinion?


I would list each SKU according to it's MSRP which falls into some market segment. It doesn't matter if they have 6s or 9s somewhere in their completely made up names which mean nothing.
You're not doing a good job of showing me how you'd separate those skus...

When referring to Low end, Mid-Range, High end, and Enthusiast tiers... you are comparing the power of a range products within a certain generation to each other... not categorizing them by prices.

By that logic AMD could overprice a shitty GPU and it would be classified as "high end" simply because it's within some arbitrary price range....


It's literally a tiered list of how the products compare to each other... and they will typically have MSRPs which reflect their tier within the stack. Those names... signify where they generally sit in the stack.
 
Last edited:
When referring to Low end, Mid-Range, High end, and Enthusiast tiers... you are comparing the power of a range products within a certain generation to each other... not categorizing them by prices.
No. Because there is zero sense in pricing segments in such approach since they will change constantly with introduction of new SKUs even within one lineup. This is just wrong and completely unusable. Pricing segments determine the demand you have at each price point which in turn allow you change the margin in products between them and predict the amount of units you need to produce for each segment.

What you're describing is just false idea of what these segments are. As I've said already "mid-range" is not in the middle of the price range, it's a fairly stable price range into which a product either fits with it's MSRP or not. No amount of $5000 products above it would change this range.

By that logic AMD could overprice a shitty GPU and it would be classified as "high end" simply because it's within some arbitrary price range....
They can and they did exactly that with Vega 20 for example. Anyone can put any product into any price tier but if it's bad for that tier then it won't be able to compete and the manufacturer will have to lower the price - and move it into a different segment with that.

It's literally a tiered list of how the products compare to each other...
No, it's not. It's a number of segments on the price scale.
 
No. Because there is zero sense in pricing segments in such approach since they will change constantly with introduction of new SKUs even within one lineup. This is just wrong and completely unusable. Pricing segments determine the demand you have at each price point which in turn allow you change the margin in products between them and predict the amount of units you need to produce for each segment.

What you're describing is just false idea of what these segments are. As I've said already "mid-range" is not in the middle of the price range, it's a fairly stable price range into which a product either fits with it's MSRP or not. No amount of $5000 products above it would change this range.


They can and they did exactly that with Vega 20 for example. Anyone can put any product into any price tier but if it's bad for that tier then it won't be able to compete and the manufacturer will have to lower the price - and move it into a different segment with that.


No, it's not. It's a number of segments on the price scale.
That's what they are RANGES.... They don't change constantly when new SKUs get added... those new SKUs simply fit into those ranges... and those equate to a certain price range..

4050ti is still a low-end GPU...
4060ti is still a mid-range GPU..
4070ti is still a mid-range GPU.. it's at the higher end of that range
4080ti is still a high-end GPU... at the higher end of that range
ect..

If the 4050ti started at $1000 and the 4080ti was $3000 would all of those GPUs be high end or enthusiast??? cmon now... No...the 4050ti would still be Nvidia's low end GPU,...ect ect..

I honestly can't believe you're having a hard time understanding this simple basic concept. It's not the price.. it's where the GPU falls within the entire generational stack. The pricing scales from there.
 
That's what they are RANGES.... They don't change constantly when new SKUs get added... those new SKUs simply fit into those ranges... and those equate to a certain price range..

4050ti is still a low-end GPU...
4060ti is still a mid-range GPU..
4070ti is still a mid-range GPU.. it's at the higher end of that range
4080ti is still a high-end GPU... at the higher end of that range
ect..

If the 4050ti started at $1000 and the 4080ti was $3000 would all of those GPUs be high end or enthusiast??? cmon now... No...the 4050ti would still be Nvidia's low end GPU,...ect ect..

I honestly can't believe you're having a hard time understanding this simple basic concept. It's not the price.. it's where the GPU falls within the entire generational stack. The pricing scales from there.
Okay. So what is the point of what you're describing and how is this used by the companies making the products?

Also in this approach are the ranges different or the same between a $$250-2000 30 series and a $$300-1500 40 series lineups? If the "mid-range" is in the middle of said ranges then how exactly has it gotten more expensive between 30 and 40 series?
 
Okay. So what is the point of what you're describing and how is this used by the companies making the products?

Also in this approach are the ranges different or the same between a $$250-2000 30 series and a $$300-1500 40 series lineups? If the "mid-range" is in the middle of said ranges then how exactly has it gotten more expensive between 30 and 40 series?
The point is to easily be able to categorize a GPU into a certain segment based on it's relative power. If someone buys a 4070... I could easily recommend them settings based on where that product falls within the stack of that generation. It also gives them an idea of what price range those products will fall into relative to the others. These companies know how they will segment their GPUs.. and price them accordingly. Any updated SKUs are meant to give people more options and pad out the market.

970 - $320 MSRP
1070 - $380 MSRP
2070 - $499 MSRP
3070 - $499 MSRP
4070 - $599 MSRP
5070 - $???? MSRP
 
The point is to easily be able to categorize a GPU into a certain segment based on it's relative power.
What does that give the manufacturer? Pricing segments is a tool used by them, they are completely useless for consumers. How does a manufacturer use that tool if it's what you're proposing - something which is constantly changing between lineups and different manufacturers (since they tend to cover different price ranges by their lineups; a 480 is a high end Radeon card by your logic for example)?
 
What else could it be seen as? It's a single monolithic GPU, same as every Hopper, now with HBM3e instead of HBM3 memories. The other chip is CPU called Grace and it's the same (including memory) as the earlier superchips.
Oh, I thought that's GH200 is a two GPU superchip.
 
So where does GF 4200 goes according to your opinion?


I would list each SKU according to it's MSRP which falls into some market segment. It doesn't matter if they have 6s or 9s somewhere in their completely made up names which mean nothing.
i wonder why they waste time with naming then... just sell them packaged with just [ nvidia: "pricetag" ]
 
i wonder why they waste time with naming then... just sell them packaged with just [ nvidia: "pricetag" ]
They do pretty much that in proviz these days. But with GeForce they obviously want you to feel like you need to buy the same "class" because this way you're the sucker who's fallen for the marketing and buy cards based on their pointless names instead of their actual important qualities.
 
The only improvement possible from a half a year delay would be to their margins.
Sure but they probably had a early 2022 GB10x designs with different configurations. I suspect GB20x is a new and different set of configurations with a few tweaks they may wanted to do.
 
Sure but they probably had a early 2022 GB10x designs with different configurations. I suspect GB20x is a new and different set of configurations with a few tweaks they may wanted to do.
Configurations are decided months before launch, you don't delay a launch a year and a half prior to that if you want to change some configurations.

This delay (if there even was one; no one gave any promises about Blackwell launch time) is likely linked to what competition will have in the same timeframe and/or production predictions (allocations, costs, etc) than them doing anything with Blackwell which they weren't planning to do previously.
 
I do wonder if the process, memory bit busses and L2 Cache sizes will look like this:

GB202 - TSMC N3E/N3P, 512-bit bus, 64MB L2 cache.
GB203 - TSMC N3E/N3P, 384-bit bus, 48MB L2 cache.
GB205 - TSMC N4P/N4X, 256-bit bus, 32MB L2 cache.
GB206 - TSMC N4P/N4X, 192-bit bus, 24MB L2 cache.
GB207 - TSMC N4P/N4X, 128-bit bus, 16MB L2 cache.

With GDDR7 and larger busses Nvidia probably saves on some die sizes, like I could see GB205 on par with AD103 or even more powerful on TSMC N4P with half the cache with a die size around 300mm2 maybe? Could price that at $700 for a 5070 Ti that's probably faster than a 4080 by 10-20% depending on other improvements?
 
To elaborate more ...
According to the source, Saudi Arabia has already purchased a batch of at least 3,000 NVIDIA H100 accelerators, the market value of which will reach $40,000 each. Developers from the United Arab Emirates are trying to keep up with their neighbors in the macro region and not only buy such accelerators in thousands of pieces, but also develop their own language model Falcon. The state-sponsored Technology Innovation Institute in Abu Dhabi has the key resources to succeed in this space: cash, energy and the ability to attract qualified personnel from around the world.
...
According to unofficial estimates, TSMC will not be able to ship more than 550,000 NVIDIA H100 accelerators by the end of this year, giving priority to processing orders from US companies. Other customers won’t get much in this situation, so they’re actively competing for the right to get the accelerators they want. Developers in Saudi Arabia want to build their supercomputers not only on the basis of the H100, but also on the A100 accelerators that are already available to them. In addition, they will use hybrid accelerators from NVIDIA’s recently announced Grace Hopper family. In specialized projects in Saudi Arabia, Chinese professionals are hoping to use their talents who have been unable to find jobs there due to US-imposed sanctions.
 
Status
Not open for further replies.
Back
Top