News AMD exec sarcastically teases RX 90700.05XTXT Max GPU with 320GB of VRAM and '1.2 Gigawatts' power supply requirement

[-Fran-]

We've been able to have 32GB video cards for a good while, but both AMD and nVidia have excused themselves in "but muh costs! too expensive for ya!".

The real question here is: why are we still putting up with it as PC people? Are we settling like Console people? There used to be a time when GPUs had exchangeable memory! Imagine that: being able to just put more memory onto your GPU with a minimal loss of performance.

🤷

Regards.

 

[Imaletyoufinish]

There have been more than a few leaks of actual pricing.
Its not good.
9070: $649
9070xt: $749

This very article mentioned those leaks and advised you to wait for the actual announcement. There were earlier leaks saying $479 as well for the XT but it doesn't matter because AMD can decide on pricing next week. Btw, the $649, $749 prices were conversions of Canadian dollars. Even if those are accurate for Canada that doesn't mean that will be official US MSRP since up north tends to get higher prices anyway.

 

[alceryes]

We've been able to have 32GB video cards for a good while, but both AMD and nVidia have excused themselves in "but muh costs! too expensive for ya!".

The real question here is: why are we still putting up with it as PC people? Are we settling like Console people? There used to be a time when GPUs had exchangeable memory! Imagine that: being able to just put more memory onto your GPU with a minimal loss of performance.

🤷

Regards.

Haha! I had one of these. I think it was the ATI all-in-wonder 16MB (yes megabyte not gigabyte) expandable to 32MB, or something like that.

The issue with upgradable/expandable VRAM on GPUs is threefold -
1) Adding any type of expansion slots increases manufacturer cost substantially vs. onboard soldered chips.
2) Memory connected via an expansion slot will be slower than soldered VRAM.
3) Properly cooling memory connected via an expansion slot is challenging.

These weren't issues back in the day, but they are very much an issue with the speeds and thermals achieved by the VRAM on today's GPUs. I'm not saying it can't be done, just that it would make even mid/low tier cards way more expensive than they are now (if you can believe it). 😉

 

[wastanley734]

And this is why Nvidia is kicking AMDs a… in the GPu game … too busy making jokes while they should be making a better product. Even instinct 300 AI accelerators while competitive against Hopper 100s they are still behind Blackwell.

Or maybe because of the illegal activities. The customers not supporting a competitive market, giving up their own spending rights. Then say crap like you said and say we don't live in a capitalist economy. We don't. Because we don't live in a free and open market. Because NVIDIA just "closed source" the supply to control the demand to validate the price inflations.

 

[wastanley734]

Haha! I had one of these. I think it was the ATI all-in-wonder 16MB (yes megabyte not gigabyte) expandable to 32MB, or something like that.

The issue with upgradable/expandable VRAM on GPUs is threefold -
1) Adding any type of expansion slots increases manufacturer cost substantially vs. onboard soldered chips.
2) Memory connected via an expansion slot will be slower than soldered VRAM.
3) Properly cooling memory connected via an expansion slot is challenging.

These weren't issues back in the day, but they are very much an issue with the speeds and thermals achieved by the VRAM on today's GPUs. I'm not saying it can't be done, just that it would make even mid/low tier cards way more expensive than they are now (if you can believe it). 😉

Soldered ddr4 cpu memory is faster than memory sticks? I think if it was then motherboard manufacturers would have tried that marketing scheme already. It's more an allocation problem now. Memory needs allocated for the rt, ai, and cores. So have memory stick slots for dgpu market is at an all time high. But people demanded sff cards. When we need a card the size of motherboard for a gpu. Quad channel ram for the cores, quad channel for the ai, quad channel for the rt cores. Without a cooler attached. Hardware development is holding us back. Zero innovations.

 

[alceryes]

Soldered ddr4 cpu memory is faster than memory sticks?

In short, yes, or at least it would be if manufacturers put time and money into it. Manufacturers are already having to consider signal integrity with super-fast, low latency DDR5. A cool short read regarding DDR5 signal integrity -
https://www.signalintegrityjournal..../post/1998-ddr5-signal-integrity-fundamentals

Now multiply those signal integrity issues due to the DDR6, DDR6X, and DDR7 used on current GPUs. In electronics, a soldered trace will always have less resistance/better signal integrity than two metals that are just touching.

I think if it was then motherboard manufacturers would have tried that marketing scheme already.

Slotted memory is too much the norm, but soldered memory has been a thing for over a decade now in laptops. It's used mainly as a space saver though.
If manufacturers put their engineers and a bunch of time and money into it, they could show the advantages of soldered RAM, but they just make it 'as fast as' slotted RAM and call it a day.

 

[acadia11]

Or maybe because of the illegal activities. The customers not supporting a competitive market, giving up their own spending rights. Then say crap like you said and say we don't live in a capitalist economy. We don't. Because we don't live in a free and open market. Because NVIDIA just "closed source" the supply to control the demand to validate the price inflations.

How does a non-competitive market benefit the consumer. Your view doesn’t hold water for a few reasons

1) capitalism by nature does not promote competition it promotes making money for individuals or a conglomerate of individuals through the free ability of individuals to own the means of production. Monopoly is often the end state and the very danger of capitalism play the game of monopoly and you’ll understand why.

2) nvidia yes has a closed eco-system but so did apple? And finally AMD was the first to 1ghz processing in the consumer arena and the jump started its fortunes in its battle with the Intel behemoth that was in 90s, it simply had a better product and give it a chance. AMd closed or open eco-system does not offer better GPUs than Nvidia, yes, software companies take full advantage of Nvidias a proprietary technology and its cost prohibitive to optimize for a GPU which has much lower market share … but AMD hasn’t given them a reason to do so for a minute. As stated ImHo this execs attitude is exactly why it’s a losing battle instead of giving the customer what he wants and there by gaining what they want $$$’s … he trolls his consumers. Smooth move. And a bit arrrogant.