Nvidia's SECRET Weapon | RTX 50 HUGE SM & Cache Changes

Nvidia's SECRET Weapon | RTX 50 HUGE SM & Cache Changes

Hot on the heels of Nvidia unveiling the Blackwell AI GPU, some fresh leaks for RTX 50 Blackwell gaming GPU have leaked. These latest leks on the RTX 50 specs and architecture unveil some HUGE changes to the SMs and cache of Nvidia's RTX Blackwell, giving us more insight into the next gen Nvdia GPU.

Join us as we examine the latest leaked info on RTX 50, which again includes updated hardware specifications, gaming and ray tracing and path tracing performance, and some very interesting insights into the RTX 50 architecture improvements. But what about RDNA 5? Will AMD be able to compete with this MONSTER graphics card?

LINKS

Whokeys 25% off code：RGT

AMAZON AFFILIATE LINKS

GreenManGaming Affiliate link -

SOCIAL MEDIA

[braindead]

A 50-70% increase in performance on the RTX 5090 over the RTX 4090 is exciting. I myself will not ever buy one but still cool to see that we have chips pushing the envelope. I am sure the power draw will be going up big time on these guys though and the price just as much!

[timotmon]

Here we go again. Hype the flagship GPU because it'll be awesome, everything else on down will be dogshit.

[HoodHussler]

The best part? 5090 is going to be at an all time low $3000! Absolutely affordable by everyone

[raysparks]

I hear that DLSS 4 will pull frames from the future using a system called Synchronistic Homogenic Inferance Technology

[svenweyers]

For the high end, I believe the uplift in performance... and the 5090 will probably be 2000$... the low end probably will live of the hype of the high end and is there is a significant uplift in that section, I will actually be surprised... I expect however that we'll get more of the same garbage like the "4060" that no uplift at all because it is a 4050...
I hope Nvidia does something that starts to restore the confidence... but I think Jensen is more interested in finding his next jacket than in the gaming market segment...

[delatroy]

It's cool to think that a narrow AI is used to optimise GPU layouts so the AI can get smarter to compute faster to optimise GPU layouts further

[burandon]

Paul knows everything because he's a vampire. He can read minds and grant immortality.

[ImUrMISERY]

Yea AMD will always have to be Price to performance lol Nidia is on a whole different level and im paying

[ivofixzone]

After "PS5 is a BEAST" and "PS6 is a MONSTER" Paul has been pretty modest on AMD and NVidia SECRET upcoming cards 🤣

[disclaimer]

they need to price RTX 50 better and more aggressive than Ampere, no more turring pricing Ada lovelace pricing was wierd and horrible, thats all I want to see.

[toddsimone]

We're gonna need some very fast CPU's to keep up. I hope Intel and AMD are on their A-game.

[FunFunFun]

4090 is suddenly $499 and many crying owners

[Jackjack.]

I cannot wait for the RTX 6090:)
But, seriously, This is the first time I have skipped one generation (4090).
Currently have the 3090 and can't wait for my 5090 new build in about 12 months most likely. Probably will not be able to get a hold of it for at least a few months after it launches unfortunately 🤔🧐

[Thomashand]

I'll happily buy the PS5 Pro, building or upgrading my pc makes no sense

[johndoh]

I don't think TSMC can get a 30% transistor density improvement using their "4" node, no matter WHAT it's named, because Nvidia is ALREADY using a custom N4 node. So if Nvidia is staying on a "4" node they aren't getting that big improvement in density.

For that kind of improvement you'd have to be on N3. I can't even begin to fathom that TSMC would want to put the work into an N4 node to try to squeeze out that kind of density.

So, if Nvidia stays on an N4 node which is certainly possible I could see TSMC having worked on it to clock it faster, and I could see getting a slight density improvement of 5 - 10%.

I could also see Nvidia using it to have a lower cost than jumping to N3, especially if they're going to add more cache on the die since cache doesn't scale as well when moving to higher density process nodes.

[kaystephan]

192GB of HBM3e VRAM at a speed of what? 8TB/s? Sounds perfect for GTA VI when it eventually comes to PC in 2026 or so 😂

[ImUrMISERY]

3200mhz boost then over clock on top of that will be insane

[johndzwon]

RT On/Off: Barely discernible visual difference, with a much bigger jump in GPU cost/price. Good for NVIDIA, not so good for everyone else.

[vincentsmith]

512 bit, that could indicate an increase in VRAM?! More than 24 GB for the 5090...?It seems to me that 32 GB od VRAM is possible, when factoring in the ray-tracing & path-tracing performance jump.

[Alexx_]

I'll tell you how things unfold. As usual. NVIDIA cards will still have a significant advantage in raytracing, probably even greater than it is now. Don't believe in this hype nonsense. AMD will catch-up with FSR and FR using machine learning. But NVIDIA shows a new tech, and AMD will need another year to come up with their own version, usually slightly inferior. R&D is the key to stay on top. NVIDIA is spending a lot of money on R&D, much more than AMD and Intel.