Nvidia’s In Trouble - AMD RDNA 5 Leaked Specs

Thank You to PCTechForum For Sponsoring This Video!

Nvidia’s In Trouble - AMD RDNA 5 Leaked Specs
AMD's RDNA 5 based RX 9000 GPUs are set to launch by the end of 2025 if leaks are to be believed meaning it will be releasing a full year before it was originally expected. With that being the case more leaks are popping up regarding its potential design and performance targets. Today we break down the latest leaks to find out just how fast the AMD RDNA 5 RX 9900 or 9950 XTX will be.

Sources

FULL DISCLOSURE
Amazon has given me affiliate links which send me a percentage of each purchase when you use them. Additionally all Vip-cdkdeals links are sponsored.
END DISCLOSURE

Specs

Thanks to Vip-cdkdeals for sponsoring this video. 30% off code: GPC20

Music
Outro: Everyone by Monplaisir

[GraphicallyChallenged]

Thank You to PCTechForum For Sponsoring This Video!
How fast do you think RDNA 5 will be?

[rurutuM]

talking about RDNA5 when 4 isn't even out yet. what is this a 2027 release?

[mdd]

How many times have we heard this in the last 6-7 years? '1080Ti in trouble', '3090 Killer', it never ends....; and is all BS

[Giovanni-Giorgio]

Youtubers: "Nvidia is in BIG trouble now!"
Meanwhile: Nvidia worth is more than all competitors combined..

[PyroPeter]

Impressive how often you Think NVIDIA is in trouble ☝🏼😅😂

[onfmysl]

NVidia is never concerned. It never hurts their sales no matter how good AMDs GCs are

[LucianPrimeOneMillion]

Yeah... So have you and alot of other content lying creators said since 2019.

[korigun]

Can't believe how often Nvidia gets in trouble

[rangersmith]

Have you heard the latest on the RX 10990XTX-X? It's a gazillion times faster than a R9-270.

[schwarz.leandro]

Let’s put that mustache on the line? lol

[FMBriggs]

These AMD videos are a meme at this point, like flat earthers declaring they finally found proof.

[vulcand]

Nvidia will only be concerned if AMD releases a card with 48 to 64GB. Now that is what I would expect to pay $2000 for.

[Riyozsu]

When has Radeon's flagship ever beaten Nvidia's flagship?

[yubrandomain]

"Nvidia should be concerned", cannot tell you how many times i've heard this but always hope Nvidia will be pushed. I just got my 7800X3D to replace my 14900k last night. I love to see what AMD got ready to slay Nvidia, give me a reason to not use my waterblocked 4080.

[edragyz]

2:43
Why even mention this?

THIS JUST IN: GPU GENERATION BUILT TO COMPETE WITH THE RTX 6000 SERIES IS FASTER THAN THE RTX 5000 SERIES OMG GUYS WOW

[chy.]

The only way AMD can compete and eat away market share is if they put performance close to a NVIDIA card, but price it several hundred less.

[Daniel-hjjt]

The problem with monolithic die designs....the larger you go, the fewer die's you can fit on a single wafer....AND the worse your yields will be, which is cost that's pushed onto the consumer. The 7900xtx's GCD has an area of 203 mm sq, a 30 % redection in die size would put that at 140mm sq for 48 WGPs....the die size of the 4090 is 609 mm sq...140 into 609 is 4.35, which would mean that just by moving to a smaller node, they could fit 208 WGPs into an area the size of a 4090.

The MCD's in the 7900xtx were manufactured on a 6nm node. Moving to 3nm, offers a logic density increase of 1.7x, which would shrink the MCD area down to 197 mm sq. 806 mm sq of silicon isn't that much when it's split across chiplets....but ok maybe it is. So subtract that MCD area from that 609 limit, and you're left with 412 mm sq. A 2.9x increase over the 7900xtx, which would still put in a 140 WGPs.

But wait! What about Zen 4/5 vs Zen 4/5c? Same architecture, identical performance clock for clock, 25% reduction in size....one would think they'd employ the same design philosophy into their next fence swing of a GPU, right? Which would put it back up to 175 WGPs in that same 609 mm sq area.

But all of that is theoretical, right?

Lets take a look at what we actually know.

So how about this, go take a look at the die of Strix point that shows off how each component fits into the die. The total die area of strix point is 232 mm sq (a 54mm increase over hawkpoint due to the inclusion of 4 additional CPU cores, and 2 additional WGPs) Strix point has 8 WGPs....and takes up 11% of the die area, or 25.52 mm sq....meaning you could fir 72 WGPs....into that 232 mm sq package...609 / 232 = 2.625 x 72 WGP - 189....and strix point is manufactured on the 4nm node! 3nm offers a density increase of 10-15% (and AMD has committed to sticking with the newest node for each new product release going forward) which would mean they could get between 208 and 218 WGPs into that 609 mm sq die area, just based off the manufacturing tech being used today.

And keep in mind, strix halo is set to move from 12 cores to 16, and from 8 to 20 WGPs, double the cache and a larger NPU....because where strix point is a monolithic design, Halo is moving to chiplets.

[elibolton]

I'm pretty saure RGT said that they weren't going to be making anything to compete at the high end and that they weren't going to beat the 7900xtx. He's been saying that over and over again.

[JoeEbitDa]

As a disgruntled RTX 4090 owner with a fire extinguisher next to his PC. I am happy to wait for something that might beat the RTX 5090, but I also want some AI capability, how long will it take for that to work with cards that are still just rumours.

[lazylemymaitland]

Ahhh… that sucks for me if the rx 9950 xtx is better than the rtx 5090, because I would love to have a AMD gpu that’s powerful, but waiting a year from now or more is to long, so I’m gonna get the 5090, I have a 4060, and I could hold out but… no one knows me for my patience