AMD’s upcoming flagship GPUs should be 3x faster than RX 6900XT

When I see "long ways out" I have to admit I dismiss it.

Either it ends up being pretty normal for the time (because we normally don't know what is in the R&D pipe for most companies), or it is Bitboys Oy Glaze3D all over again.
 
I keep seeing 2.5x as the goal, which shouldn't be too hard, since the top cards with have 160 CUs, not 80. I also saw I think on RGT that they're above 50 percent with IPC+frequencies which may actually get them close to 3x overall (2x+1/2*2x=3x). A 25 percent improvement of IPC+frequencies would get them at 2.5x alone, and that's not at all unrealistic.

RDNA2 was what, between 30 and 40 percent more powerful than RDNA, CU per CU?

I'm a bit concerned about pricing, since Nvidia is supposed to be competitive even with AMD's projected gains.
 
One could hope for a bit of competition. I would think that the lack of silicon might still have a deep impact on these actually launching.
 
If this is when AMD move to chiplets in GPUs, I could actually see larger than normal performance ceiling rise on the high-end- but I wouldn't believe for a moment that performance in the midrange or perf/$ in general will go up by anything remotely close to 3x. If the top-end RX 7000 or whatever is somehow 2.5-3x as fast as 6900XT by virtue of chiplet scaling on top of IPC & clock gainz, it'll also be an even higher performance tier (7990?) and cost more accordingly. Feel free to quote the entirety of this post next year to see how I did lol.
 
Man, if anyone thinks that AMD is doing anything to get people to postpone buying the current generation and wait and see what's in store for ... 2022? they need to get their hypometers calibrated.
 
I dunno about these stories.
No one can even get the current gen to know if they are fast, so who gives a shit whats coming down the line :D
 
As others have said, why even bother to project what it will be? They might as well just shoot for the moon and say it will be 50x faster for all the good it will do or it means.
Not that I'll be buying another GPU in the next 2 years anyway it seems.
 
Last edited:
I think I'll just buy a bot and then just sell all the GPUs to everyone here on H at MSRP. Seriously though, this is getting ridiculous.
 
I don't see why it couldn't be that fast in certain situations. The chiplet thing could either be a pro or a con. Obviously, they get better yields, but if it introduces microstutter due to the infinity fabric or games don't scale right with the multi-GPU it could be a sidegrade.
 
Ow. When I read "should be three times than RX 6900XT" I hoped that it would be "should be three times more in stock than RX 6900XT"but in these weird times, the fact that it's "three times more performance" is a big disappointment.
 
NattyKathy said:
If this is when AMD move to chiplets in GPUs, I could actually see larger than normal performance ceiling rise on the high-end- but I wouldn't believe for a moment that performance in the midrange or perf/$ in general will go up by anything remotely close to 3x. If the top-end RX 7000 or whatever is somehow 2.5-3x as fast as 6900XT by virtue of chiplet scaling on top of IPC & clock gainz, it'll also be an even higher performance tier (7990?) and cost more accordingly. Feel free to quote the entirety of this post next year to see how I did lol.
Click to expand...
Also, the bandwidth requirements of the interconnect should be MASSIVE, if they figured that out, then I can see a 2 chiplet + IO die GPU being 2.5x.
I wonder how well it will scale compared to the 80CU non-chiplet version.
Also, I would expect the TDP of the 2 chiplet version to be pretty high.
 
Doubling the CUs make you wonder about the price tags, but if that trickle down the line-up, that would be excellent news.
NattyKathy said:
If this is when AMD move to chiplets in GPUs, I could actually see larger than normal performance ceiling rise on the high-end- but I wouldn't believe for a moment that performance in the midrange or perf/$ in general will go up by anything remotely close to 3x. If the top-end RX 7000 or whatever is somehow 2.5-3x as fast as 6900XT by virtue of chiplet scaling on top of IPC & clock gainz, it'll also be an even higher performance tier (7990?) and cost more accordingly. Feel free to quote the entirety of this post next year to see how I did lol.
Click to expand...
It is according to the article it would be going in the opposite direction I think (I could be missing what they meant or what chiplets mean):

The current flagship, Navi 21, offers 80 compute units and 5,120 shading units. Navi 31 will reportedly double those numbers, offering a whopping 160 CUs and over 10,000 shading units, delivered in two 80 CU chiplets on a single die.
 
LukeTbk said:
Doubling the CUs make you wonder about the price tags, but if that trickle down the line-up, that would be excellent news.

It is according to the article it would be going in the opposite direction I think (I could be missing what they meant or what chiplets mean):

The current flagship, Navi 21, offers 80 compute units and 5,120 shading units. Navi 31 will reportedly double those numbers, offering a whopping 160 CUs and over 10,000 shading units, delivered in two 80 CU chiplets on a single die.
Click to expand...
Chiplets are what allowed AMD to wreck Intel with CPUs.

And they more than likely will let AMD wreck nVidia finally.

What it means is amd can stick two GPU chiplets on a monolithic sized package and bridge the two gpus together into a single unified gpu architecture via infinity fabric and feed the two gpu chiplets a very huge level 3 cache.

Want 3 or 4 gpus unified into one? Done! Just stick more chiplets in and wire them to the fabric and allocate some more cache.

Meanwhile nvidia has to stuff thier competition into a big ass single smoking hot 5 GWatt chip.
 
Krenum said:
Say what?! :eek:

https://arstechnica.com/gadgets/202...ship-gpus-should-be-3x-faster-than-rx-6900xt/

"According to wccftech and well-known hardware leaker @KittyYYuko, AMD's next-generation Radeon graphics cards should be a massive upgrade—as much as three times faster than AMD's current flagship graphics cards"

Still a long ways out though...
Click to expand...

I hear they'll also be up to three times more Out of Stock ;)

It'll be interesting to see them when they're introduced. Since I've decided to skip this generation I may very well wait for these...
 
tangoseal said:
Chiplets are what allowed AMD to wreck Intel with CPUs.

And they more than likely will let AMD wreck nVidia finally.

What it means is amd can stick two GPU chiplets on a monolithic sized package and bridge the two gpus together into a single unified gpu architecture via infinity fabric and feed the two gpu chiplets a very huge level 3 cache.

Want 3 or 4 gpus unified into one? Done! Just stick more chiplets in and wire them to the fabric and allocate some more cache.

Meanwhile nvidia has to stuff thier competition into a big ass single smoking hot 5 GWatt chip.
Click to expand...
Ok so like what I thought that AMD rumored release is going in the complete opposite direction than chiplet right ?
 
Not a surprised when current AMD RDNA2 GPU's are using 256-bit memory bus and using plain GDDR6 memory. Add chiplet and 5nm and I can easily see 3x performance increase.
 
Last edited:
Yes. Seem's on par for course. Duh! Double the shit with a bit of tuning and ya!
Anyone who doesn't see that as an easy goal, Didn't own a GPU in the early 2000's.
Raw-Raw time !!!
:ROFLMAO::ROFLMAO::ROFLMAO:
 
I hope they can get AFR sorted so they can double output and decrease power by 15%. Each chiplet could render the next consecutive frame and have twice as long to do it there by reducing power and thermals.
 
OutOfPhase said:
When I see "long ways out" I have to admit I dismiss it.

Either it ends up being pretty normal for the time (because we normally don't know what is in the R&D pipe for most companies), or it is Bitboys Oy Glaze3D all over again.
Click to expand...

Ummm...Bitboys.....the good stuff
 
learners permit said:
AFR
Click to expand...

Eh, they have a different solution going. Basically the system only sees one GPU, which is the I/O die. The I/O die tells the chiplets what to do, and takes the info they handle and present it to the system. Essentially, it's just one big GPU even if it's hosted on multiple physical dies.

I don't know if they work like one big die or if they split up roles, like SFR back in the day. But however it'll work, it's a black box as far as the system is concerned.
 
You must log in or register to reply here.
Back
Top