wccftech NVIDIA GeForce RTX 5090 GPU Rumor Alleges 512-bit Monster, 16 GDDR7 Memory Modules, Brand New PCB Design

[bonami2]

  Quote

The rumor comes from Chiphell Forum member, Panzerlied, who has been active in the scene in talking about the different aspects of NVIDIA's GeForce RTX 50 "Blackwell" GPUs. The member has previously mentioned the Blackwell GPU SKUs under the GB2** family and also the performance for the upcoming flagship compared to the RTX 4090.

 

The NVIDIA GeForce RTX 5090 GPU has been said to feature a 512-bit memory interface on several occasions and has been verified by other reliable leakers and it looks like we have the first layout pointed out within this rumor. There are going to be a total of 16 DRAM modules featured on the RTX 5090 PCB. All of these modules will be based on the GDDR7 standard and we know that initial dies will stick to 28 Gbps speeds. The GPU will have four memory modules on the top, five each on the sides, and two at the bottom. This new configuration will require a brand-new PCB design that will be capable of hosting the increased memory layout.

Expand  

NVIDIA GeForce RTX 5090 GPU Rumor Alleges 512-bit Monster, 16 GDDR7 Memory Modules, Brand New PCB Design

WCCFTECH.COM

---

New rumors for the NVIDIA GeForce RTX 5090 "Blackwell" GPUs have surfaced, revealing a brand new PCB & a 512-bit GDDR7 memory interface.

 

[Snakecharmed]

No skimping out on memory (although the -90 cards wouldn't do that anyway) this time. 32 GB 5090 incoming for...$1999? Because I simply expect generational price hikes from Nvidia at this point.

[ENTERPRISE]

  On 25/05/2024 at 17:43, Snakecharmed said:

No skimping out on memory (although the -90 cards wouldn't do that anyway) this time. 32 GB 5090 incoming for...$1999? Because I simply expect generational price hikes from Nvidia at this point.

Expand  

It will be at LEAST that price for the base cards/founders. I Reckon the AIB cards will be astronomical.

[Sir Beregond]

I still really doubt the 512-bit rumor for a consumer product in the GeForce stack. 

 

But who knows, Nvidia could decide $2500 is the way to go and 512-bit gives them more vram capacity until higher density GDDR7 modules come out. For now their density is the same as GDDR6X (2GB per module).

[bonami2]

I read an article recently showing some 4000 tier card having pretty good bandwith bottleneck. Maybe Nvidia hit a wall or something and had no other option than increase it back to 512bit

[Sir Beregond]

  On 26/05/2024 at 01:49, bonami2 said:

I read an article recently showing some 4000 tier card having pretty good bandwith bottleneck. Maybe Nvidia hit a wall or something and had no other option than increase it back to 512bit

Expand  

Doubt it. The move from 40-series to 50-series also sees a move to GDDR7 which will have better bandwidth/speed, and again will also see a larger L2 cache.

 

I mean I could be totally wrong about the 5090 and they might go all out on it, but the rest of the stack? Probably going to mirror the 40-series a ton in terms of 256-bit and 192-bit products in the 80 and 70 tiers.

[Barefooter]

From Tom's Hardware

RTX 5090 may be surprisingly svelte — twin-slot, twin-fan model on the way, says leaker

 

  Quote

Is Nvidia giving its most powerful GPU a diet?

 

Nvidia’s 50-series GPUs are expected to arrive by October or November this year, just in time for the holiday season. As usual, we expect the company to release the flagship RTX 5090 and RTX 5080 GPUs first, showing the way with its Founders Edition (FE) models. However, rumors of a hulking 5090 FE design have been rebuffed by a well-regarded leaker, who says previously touted super-fat graphics card designs are inaccurate.

 

It was initially thought that the 5090 FE would be based on the unreleased 4090 Ti, taking up four slots inside a PC case. It would thus probably have a triple-PCB layout, with three fans interspersed throughout the GPU shroud, according to various sources. Now, popular leaker @kopite7kimi has denied the rumors on X, saying that the RTX 5090 FE will come with a twin-slot, twin-fan design.

If this new information is true, it makes the RTX 5090 a much slimmer design than the 4090 FE, which was delivered with a triple-slot cooling solution. Other AIB partners offer larger even 4090 designs that could take up almost four slots. A smaller and slimmer flagship GPU would be a welcome change after several generations of larger and larger GPUs. The problem of heavier graphics cards has come to the point that Gigabyte introduced a motherboard that can support 128-pound GPUs, but still didn’t solve the issue with the GPU itself cracking under its own massive weight.

 

The leaker doesn’t share information about whether the 5090 will have lower power consumption than the 4090, but they said that the new GPU’s cooling design will be more efficient. We wouldn’t be surprised if the 5090 turns out to require less power for its performance level. Nvidia has been at the forefront of AI chip development, and it powers the top three most efficient supercomputers as of May 2024. We would like to think that Nvidia’s developments in this sector would trickle down to its consumer and gaming business, allowing end-users to enjoy more power without requiring the largest possible PC cases.

Expand  

 

[EHW Ai]

Kinda makes me wonder if they'll do away with that cluster**** that is the 12VHPWR connector.

[Fluxmaven]

Some beefed up monster GPU sounds neat but I wouldn't be a buyer. If they focus on efficiency with decent bump in performance and keep the prices the same, I might be replacing my 3090. 

 

If the 5080 was a bit faster than a 4090 with ≤ 300w power draw at $1000, I think they would have a winner. 

 

That really shouldn't be much to ask for, but Ngreedia just can't help themselves. 

[The Pook]

  On 29/05/2024 at 19:33, Fluxmaven said:

Some beefed up monster GPU sounds neat but I wouldn't be a buyer. If they focus on efficiency with decent bump in performance and keep the prices the same, I might be replacing my 3090. 

 

If the 5080 was a bit faster than a 4090 with ≤ 300w power draw at $1000, I think they would have a winner. 

 

That really shouldn't be much to ask for, but Ngreedia just can't help themselves. 

Expand  

 

I don't even need a big performance bump if the power savings are significant and I get more VRAM. 

 

Give me ~+5-10% 3080 Ti performance @ 200W with DLSS 3 and 24GB of VRAM and I'll buy it yesterday. 

[EHW Ai]

  On 29/05/2024 at 22:51, The Pook said:

 

I don't even need a big performance bump if the power savings are significant and I get more VRAM.

Expand  

 

...so you really didn't want a 4090, eh?

 

  On 29/05/2024 at 22:51, The Pook said:

Give me ~+5-10% 3080 Ti performance @ 200W with DLSS 3 and 24GB of VRAM and I'll buy it yesterday. 

Expand  

 

boy, do i have a 3090 Strix OC for you

 

(yesterday is here)

Edited May 30, 2024 by iamjanco

[The Pook]

  On 30/05/2024 at 00:52, iamjanco said:

 

...so you really didn't want a 4090, eh?

Expand  

 

if it's free, yes 

 

(thx in advance) 

 

  On 30/05/2024 at 00:52, iamjanco said:

boy, do i have a 3090 Strix OC for you

Expand  

 

is it a magic 3090 with DLSS 3 frame gen? 

Edited May 30, 2024 by The Pook

[EHW Ai]

  On 30/05/2024 at 01:03, The Pook said:

is it a magic 3090 with DLSS 3 frame gen? 

Expand  

 

only if you've got a working magic wand.

[SoloCamo]

I just want a card near 4090 performance with 20gb+ of vram and a 300w or less tdp and I'm sold.   What I feel like I'm going to end up with is whatever midrange AMD GPU they put out for their next gen since they aren't going to have a proper halo card.  I'm hoping it's closer to a 7900XTX performance wise with a much lower power envelope.  If neither AMD nor Nvidia deliver I guess I'll be riding with my 6900XT yet again.  I took a 290X from 2013 to 2019 or so, I can hold off another gen if needed.   It also helps that there aren't a ton of games that I actually care about that require more horsepower...

Edited May 31, 2024 by SoloCamo