• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Discussion Ada/'Lovelace'? Next gen Nvidia gaming architecture speculation

Page 89 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
It's not so simple. You're effectively doubling the memory cost and memory power consumption. Not to mention any of the additional costs of doing the clamshell on top of that.

Of course cost is an issue. If it's an option I don't see the issue paying more for more memory.

In the 3000 series, they could have used 2GB chips and doubled RAM on the 3060 Ti, 3070 (256 bit bus).

Though this generation the bus is very narrow so more problematic with only 2GB chips forcing you to double up chips on the other side of the board. I don't get why they can't produce 4GB packages for this.
 
Of course cost is an issue. If it's an option I don't see the issue paying more for more memory.

If it is 20 bucks a chip (I have no idea what GDDR6X costs), that would be an extra 120 bucks in cost just for the memory on the 4070/Ti for instance. I imagine a board would have to be another $150-$200 more than the already high prices.
 
If it is 20 bucks a chip (I have no idea what GDDR6X costs), that would be an extra 120 bucks in cost just for the memory on the 4070/Ti for instance. I imagine a board would have to be another $150-$200 more than the already high prices.

Then you can buy the one with less memory. I said: "Option".
 
If it is 20 bucks a chip (I have no idea what GDDR6X costs), that would be an extra 120 bucks in cost just for the memory on the 4070/Ti for instance. I imagine a board would have to be another $150-$200 more than the already high prices.
I think AD106 and AD107 should use standard GDDR6.
For comparison.
Radeon RX 5500 XT 8GB 14gbps had $30 higher MRSP than the 4GB version.
Radeon RX 6500 XT 8GB 14gbps 18gbps had $20 higher MRSP than the 4GB version.
I don't see why would this option cost >$100, unless Nvidia or OEM were being greedy.
 
Last edited:
If it is 20 bucks a chip (I have no idea what GDDR6X costs), that would be an extra 120 bucks in cost just for the memory on the 4070/Ti for instance. I imagine a board would have to be another $150-$200 more than the already high prices.
I am quite entertained by these posts. let's see.

I don't know memory prices, BUT, it will add $(150-200) on a $800 card.
 
I think AD106 and AD107 should use standard GDDR6.
For comparison.
Radeon RX 5500 XT 8GB 14gbps had $30 higher MRSP than the 4GB version.
Radeon RX 6500 XT 8GB 14gbps 18gbps had $20 higher MRSP than the 4GB version.
I don't see why would this option cost >$100, unless Nvidia or OEM were being greedy.

Additional Issue here is the narrow bus though, they are using the max density 2GB chips. So you would need an all new more expensive board design as well as the additional chips. But I agree, it shouldn't cost more than $100, to go from 8GB to 16GB. Especially given current memory pricing.
 
I think AD106 and AD107 should use standard GDDR6.
For comparison.
Radeon RX 5500 XT 8GB 14gbps had $30 higher MRSP than the 4GB version.
Radeon RX 6500 XT 8GB 14gbps 18gbps had $20 higher MRSP than the 4GB version.
I don't see why would this option cost >$100, unless Nvidia or OEM were being greedy.

5500 XT was 128-bit, so it was using the same amount of chips. The real price gap on the 6500 XT was more than $20, and production of the 8 GB was more or less ended immediately due to being too late to sell to miners.
 
5500 XT was 128-bit, so it was using the same amount of chips. The real price gap on the 6500 XT was more than $20, and production of the 8 GB was more or less ended immediately due to being too late to sell to miners.
True, but I don't think 1Gbit and 2Gbit memory chips cost the same, but cost difference should be lower than 2x.
What was the price gap between those RX 6500XT? Was It $75-100?
 
True, but I don't think 1Gbit and 2Gbit memory chips cost the same, but cost difference should be lower than 2x.
What was the price gap between those RX 6500XT? Was It $75-100?

I'll have to look into that. But IIRC Sapphire wanted to sell the 6500 XT 8 GB for $300ish I think.
 
I'll have to look into that. But IIRC Sapphire wanted to sell the 6500 XT 8 GB for $300ish I think.
The Sapphire Pulse 6500XT 8GB is the only version to make it to UK retail and it was £30 more than the 4GB MSRP of £200. It was actually in stock until a few weeks ago, still at the launch MSRP.
 
I am quite entertained by these posts. let's see.

I don't know memory prices, BUT, it will add $(150-200) on a $800 card.

Well I can tell you with pretty high confidence it's somewhere between 15-25 dollars a chip. So adding an additional 6 chips would be somewhere between 90-150 in additional cost just for the memory.

The Sapphire Pulse 6500XT 8GB is the only version to make it to UK retail and it was £30 more than the 4GB MSRP of £200. It was actually in stock until a few weeks ago, still at the launch MSRP.

The 4 GB was probably nowhere near 200 pounds by then. In the US the 8 GB didn't last very long. Maybe it just took the UK retailer that long to sell off whatever they bought.
 
Well I can tell you with pretty high confidence it's somewhere between 15-25 dollars a chip. So adding an additional 6 chips would be somewhere between 90-150 in additional cost just for the memory.



The 4 GB was probably nowhere near 200 pounds by then. In the US the 8 GB didn't last very long. Maybe it just took the UK retailer that long to sell off whatever they bought.
I read memory prices are in the doldrums presently and expected to further deteriorate this year.

Is GDDR6 demand outpacing supply?
What's the node used?
What's the die area?
What are the relative values to DDR5 & GDDR5?
 
Well I can tell you with pretty high confidence it's somewhere between 15-25 dollars a chip. So adding an additional 6 chips would be somewhere between 90-150 in additional cost just for the memory.

You can search online if you know the part number. This is the fast GDDR6X that NVidia uses on the RTX 4080:

https://www.avnet.com/shop/apac/products/micron/mt61k512m32kpa-24-u-3074457345647662003/

Even at 100K parts, they want $30/chip. These are 2GB chips. More expensive than I would have thought. That's $120 for 8GB. I would think NVidia can get a better deal.

I found older, slower GDDR6 for about $18/chip.
 
Is there a big difference in performance in your opinion?

TPU seems to love the Trio X. Seems like the best 4090 to get at the lowest price, as long as you don't plan to increase the manual power limit beyond 480W.
 
Well I can tell you with pretty high confidence it's somewhere between 15-25 dollars a chip. So adding an additional 6 chips would be somewhere between 90-150 in additional cost just for the memory.



The 4 GB was probably nowhere near 200 pounds by then. In the US the 8 GB didn't last very long. Maybe it just took the UK retailer that long to sell off whatever they bought.

6500XT's are from £140 to £200 right now at the same shop that sold the 8GB model. 24Gbit chips (as have been mentioned) are sorely needed in GDDR form, lets hope GDDR7 comes in a wider variety of sizes than 1GB & 2GB per chip.
 

GPU-Z of a final 4070 seemingly leaked. Fill rate and compute power roughly in line with the 3080.
 
24Gbit chips (as have been mentioned) are sorely needed in GDDR form, lets hope GDDR7 comes in a wider variety of sizes than 1GB & 2GB per chip.
I went a-looking and noticed that Samsung is finalizing GDDR6W. This has two layers of DRAM inside the chip, doubling capacity per chip. It goes to 22 Gbps, while GDDR6X apparently tops out at 21 Gbps.

However, since it has double the pins, I assume that it doesn't actually help to provide double the VRAM for a certain bus size.

 

TPU seems to love the Trio X. Seems like the best 4090 to get at the lowest price, as long as you don't plan to increase the manual power limit beyond 480W.
I would be reducing power. You can as far as I recall reduce the power a great deal before it begins effecting performance levels.
 
Back
Top