NVIDIA GeForce RTX 5090 Features 16+6+7 Phase Power Delivery on 14-Layer PCB

[Chrispy_]

Business buying these up for AI means that the likely $3000 MSRP of these is going to be irrelevant, as they'll be scalped hard at launch, probably 50% more than MSRP and so hard to get your hands on one.

Yes, if you have infinite money you can throw it away on one for gaming, but the game developers are still targeting far more modest hardware and 5090 owners aren't something they're even paying any attention to.

 

[L0stS0ul]

Greed. In the end they will be left with hundreds of thousands of unsellable chips that no one will buy, even for Deep Learning

 

[TechBuyingHavoc]

Greed. In the end they will be left with hundreds of thousands of unsellable chips that no one will buy, even for Deep Learning

I don't think it works like that. Either people WILL buy them and the prices remain elevated, or they don't and the prices will fall until people do start buying them.

 

[TokyoQuaSaR]

utilize a 14-layer PCB, possibly due to the increased complexity of GDDR7 memory modules and power delivery. Usually, GPUs max out at 12 layers for high-end overclocking designs.

This is pretty much entirely due to the pin numbers of the GPU. 512b memory bus requires a huge amount of IO. This and the huge demand for power supply current. In the end, the more rows and columns of IOs, the more layers needed, that's pretty much it.

 

[Wirko]

This is pretty much entirely due to the pin numbers of the GPU. 512b memory bus requires a huge amount of IO. The more rows and columns of IOs, the more layers needed, that's pretty much it.

PCIe 5.0 makes things even more complicated. Given the memory chip layout, some lanes probably run under the memory chips and/or other lanes. This is quite unfavourable because these wires require a specific geometry (distance to ground, to other lanes etc).

 

[oddrobert]

High end as we knew it is dead. It's either "HI-FI" or "MID-FI" if we compare GPUs to headphones or speakers. Either you pay A LOT to get true high end (5090), or just a lot and get mid end, advertized as high end (5080). There is nothing in between and that's by design. Nvidia wants to be a luxury brand. I would laugh to anyone writing that GPU could be a luxury 10 years back, but here we are

Luxury brand which need to be avoided at all cost. Their marketing bubble is incredibly strong. As well-informed costumer run away.
Giving allure of buying something more expensive but better, innovative and prestigious. LIE.

Indulging in Exurban Opulence of a Modern GPUs? (>300$)

5
"HI-FI"
4090/5090 class
Premium halo product of second order UPSELL.

4
"MID-FI"
4070 Ti S class
Premium halo product first order OVERSPEND.

3
"HIGH END"
7800xt /7900 GRE/9070xt class
Way too EXPENSIVE for a toy it is, THE MORE YOU SPEND THE MORE YOU SAVE

2
"MID TIER"
6750xt/7700xt
SUFFICIENT
AMD native 1440p HDR

1
"VALUE"
ENOUGH
¬200euro used

0
"FREEDOM TIER"
PERFECT
no GPU, no gaming
healthier, wealthier, and wiser

 

[Wirko]

0
"FREEDOM TIER"
PERFECT
no GPU, no gaming
healthier, wealthier, and wiser

Ah, and you finally have enough free time for golfing, tennis, sailing and horse riding!

 

[TheinsanegamerN]

so this thing is most likely gonna cost $3000 min?

i still remember when $700 was for the top-dog, now it won't even get you mid-range xx70ti class card

I too remember a time when our money supply was LESS THEN HALF what it is today, when $40k a year was a decent wage, and we were coming out of a massive economic collapse that suppressed the cost of almost everything. I remember when a full wafer from TSMC cost $2000, not $30,000.

I don't think it works like that. Either people WILL buy them and the prices remain elevated, or they don't and the prices will fall until people do start buying them.

Inflation is a thing, but people dont like hearing the truth. Everyone seems to think a $500 RTX 4090 is just around the corner.....

 

[AusWolf]

I too remember a time when our money supply was LESS THEN HALF what it is today, when $40k a year was a decent wage, and we were coming out of a massive economic collapse that suppressed the cost of almost everything. I remember when a full wafer from TSMC cost $2000, not $30,000.

The problem is that $40k a year is still a decent wage. Only prices have gone up, wages haven't (at least here in the UK).

 

[mb194dc]

Business buying these up for AI means that the likely $3000 MSRP of these is going to be irrelevant, as they'll be scalped hard at launch, probably 50% more than MSRP and so hard to get your hands on one.

Yes, if you have infinite money you can throw it away on one for gaming, but the game developers are still targeting far more modest hardware and 5090 owners aren't something they're even paying any attention to.

Possible, still nothing on the front end that looks remotely like paying for the several hundred billion spent on GPUs to go in LLMs so far.

5090 is seemingly like a Titan card and will have 0.5% of the market at most.

 

[Chrispy_]

Possible, still nothing on the front end that looks remotely like paying for the several hundred billion spent on GPUs to go in LLMs so far.

5090 is seemingly like a Titan card and will have 0.5% of the market at most.

It's only the massive megacorps filling datacenters with NVL72 racks at $3M a rack before operating costs.

The other 7 million tech companies around the world who don't need datacenter-scale tech are just buying 4090s. We have a few pods of 7 GPUs in a 4U chassis, and they do happen to be be Quadros but only because we need a Quadro-specific software feature. For many ML and LLMs applications a few 4090s in 8-12U of rack space replace multiple 42U racks of traditional compute, so that's why 4090s are so expensive. The business supply wholly outstrips Nvidia's production capability, even with sanctions against the largest customer (China), so they're stupid expensive for consumers.

 

[Wirko]

I remember when a full wafer from TSMC cost $2000, not $30,000.

That was when the manufacturing cost per transistor was dropping sharply every year... now no more.