AMD Radeon RX 7000 RDNA3 To Launch Early December

[Makaveli]

Depending on price I think I will be going 6800XT to 7900XT

 

[Nopa]

Coming from 3090, DP 2.1, PCIe 5 x16, and being MCM will make this a true upgrade for me. Can't wait!

 

[evernessince]

Big difference between xtx and xt. Unexpected. No gddr6x, hmm... there should be visible difference in 2k and 4k against npeedia.

The 4090 has 1,008 GB of memory bandwidth vs the 7900XT's 960. If you factor in infinity cache it's more likely that AMD has the advantage in regards to memory subsystem when you consider the relief infinity cache provides for it.

AMD just call them 7900XT and 7800XT ffs, stop doing this dumb naming..

I'd theorize that AMD is calling it the 7900 XTX because they want to charge above 7900 XT prices.

There is no reason for 1% lows to suffer. It isn't an MCM in the traditional sense where there are two GPU cores and separate memories working on separate frames. This is multiple dies working together. Think zen 2 chiplets vs two-socket servers. In two socket servers the NUMA-ness of the system is a lot more exposed than with zen 2. Especially since AMD is only putting one compute die, so this is a ryzen 3700X(1 compute 1 io die), not a 3950X.

The performance of the GPU will be similar or identical to existing designs. There will be different tradeoffs, so games will need to be re-optimized for the new architecture, since the bottlenecks are in a different place. So expect performance in existing games to be worse, but in newer games that have access to RDNA3 GPUs to improve. Similar to other generations of GPUs, but more impactful.

There is almost no chance that AMD is releasing a GPU that requires optimization to see performance benefits over last gen cards. That'd almost be on the level of another bulldozer and DOA.

Prior MCM designs, including Zen 1, all saw performance benefits in all existing applications so there is no reason to believe the same wouldn't apply here. The only known disadvantage for MCM is when bouncing threads around CCDs. It remains to be seen whether that'll apply to GPUs given their highly parallel nature.

 

[mb194dc]

If it's 300w and a reasonable price AMD might have a winner.

Realistically the XTX will probably be $1500+...For me that's DOA.

Even if technologically impressive like the 4090, releasing ultra expensive cards to this environment is a commercial mistake imo.

 

[shovenose]

I have hope for a true dual slot card here, 7900xt, even if it’s a blower design I want it.

 

[ARF]

I'd theorize that AMD is calling it the 7900 XTX because they want to charge above 7900 XT prices.

Probably they want to move the whole new lineup a tier up. Bye-bye good RX 7800 XT will few disabled shaders, say hello to severely cut RX 7800 XT with Navi 32, which would normally have been only 7700 XT

The same as with the 12 GB "RTX 4080" drama that ended with a happy ending, fortunately.

 

[FeelinFroggy]

Did I read "It's rumored that the company is targeting a 100% performance uplift over the previous-generation"

Doubling the performance of their previous generation is beyond a rumor, it's a lie. We saw the same nonsense with Nvidia that their next gen cards would double performance.

We hear the same rumors about CPU's with potential 50% IPC improvement. Those numbers are never even close to reality so why even publish them?

 

[LupintheIII]

I only hope the latency won't be huge and due to the 2 MCMs. I guess AMD has this issue in the back of their heads and they have addressed the potential latency.
Are they really aiming to double the performance? wonder if they can pull this off.

RX 6900XT was double the shaders of 5700XT with half the memory bus on the same node, it is more than double the performance.
"7900XTX" or whatever is gonna be called is more than double the shaders of 6900XT with double the memory bus on a better node.
I really don't see how it is difficult to belive they can double performance.

I don't think latency will be an issue, but frame pacing will almost certainly be affected by MCM design. Even if they get close to 4090 average framerates, the 1% lows are almost sure to suffer because of MCM.

On the bright side, the total die size is smaller than 6900XT (with better yields because the die area is broken into smaller chiplets), so I am hoping for the same $999 MSRP on the 7900XT. AMD could realistically price it lower, but with nvidia's pricing so high up, I don't think AMD will have any incentive to do so.

My realistic guess is

7900XT = $999 to undercut the 4080, while outperforming it in rasterization.
7900XTX = $1299 to undercut the 4090, while being about 5% off in average rasterization performance (similar to 6900XT vs 3090).

This has the same vibe as "N21 = RTX 3070 because 256bit-bus"... have you stopped for a second thinking maybe AMD is aware 1% lows exist?

Did I read "It's rumored that the company is targeting a 100% performance uplift over the previous-generation"

Doubling the performance of their previous generation is beyond a rumor, it's a lie. We saw the same nonsense with Nvidia that their next gen cards would double performance.

We hear the same rumors about CPU's with potential 50% IPC improvement. Those numbers are never even close to reality so why even publish them?

RX 6900XT was double the shaders of 5700XT with half the memory bus on the same node, it is more than double the performance.
"7900XTX" or whatever is gonna be called is more than double the shaders of 6900XT with double the memory bus on a better node.
The fact Nvidia presentations are just a bunch of marketing claims with no substance doesn't mean everyone else is doing the same.

Based on the specs alone RX 7900 XT should eat 4080 for breakfast. It will all come down to pricing now. Price it $899 and they fly off the shelves like hot cakes, but I have a bad feeling that Lisa will accept Jensen's price hiking game and charge $1,199 USD maybe minus 50 bucks for this thing.

Those are not RTX 4080 competitors (it would get simply disintegrated), there is N32 for that.
Price will be accordingly

 

[evernessince]

Probably they want to move the whole new lineup a tier up. Bye-bye good RX 7800 XT will few disabled shaders, say hello to severely cut RX 7800 XT with Navi 32, which would normally have been only 7700 XT

The same as with the 12 GB "RTX 4080" drama that ended with a happy ending, fortunately.

That's entirely possible. It's up to customers to push back on that should it happen if we want to see well priced GPUs.

 

[Blue4130]

After the 4080 naming outrage and fiasco, where is the outrage over two very different 7900s being called the same thing?

 

[HTC]

After the 4080 naming outrage and fiasco, where is the outrage over two very different 7900s being called the same thing?

NOT the same thing: unlike the 4080s where you have TWO different chips (103 and 104 IIRC), here you have ONE chip (N31).

 

[Blue4130]

NOT the same thing: unlike the 4080s where you have TWO different chips (103 and 104 IIRC), here you have ONE chip (N31).

With vastly different specs and likely performance. Who cares what the silicon is.

 

[HTC]

With vastly different specs and likely performance. Who cares what the silicon is. View attachment 268051

Compare the difference between these and nVidia's two 4080 cards, and see for yourself.

 

[Blue4130]

Compare the difference between these and nVidia's two 4080 cards, and see for yourself.

You can't make a direct comparison like that, the architecture is too dissimilar.

 

[ymdhis]

Navi 33 seems to be the one you have to look out for... unfortunately it is rumoured to have 4-500$ price tag, because AMD is shifting the entire lineup higher.

It's silly that Polaris still has more performance per dollar than anything else in the budget region.

 

[Kapone33]

For me it is the reason at that earlier in the year I bought a 4K 144HZ panel and not too long ago a 5800X3D. The price will probably suck but at the same time I am confident that it will feel faster than my current 6800XT. I think I like the XT but if the XTX is priced right it will be mine. Merry Xmas indeed.

 

[HTC]

You can't make a direct comparison like that, the architecture is too dissimilar.

NOT what i mean: the difference between the rumored 7900 models is A LOT SMALLER than the difference between the 4080 models:

The 7900 XT has, relative to the 7900 XTX:

- 87.5% stream processors
- 83.33% memory
- same memory speed but 83.33% bus width
- 83.33% memory bandwidth

The 4080 12 GB has, relative to the 4080 16 GB:

- 78.95% CUDA cores
- 75% memory
- 91.3% memory speed but 75% bus width
- 68.48% memory bandwidth

Which do you think is worse?

 

[Blue4130]

NOT what i mean: the difference between the rumored 7900 models is A LOT SMALLER than the difference between the 4080 models:

The 7900 XT has, relative to the 7900 XTX:

- 87.5% stream processors
- 83.33% memory
- same memory speed but 83.33% bus width
- 83.33% memory bandwidth

The 4080 12 GB has, relative to the 4080 16 GB:

- 78.95% CUDA cores
- 75% memory
- 91.3% memory speed but 75% bus width
- 68.48% memory bandwidth

Which do you think is worse?

I can't say with certainty because it's not released, and it's not a linear scale between

NOT what i mean: the difference between the rumored 7900 models is A LOT SMALLER than the difference between the 4080 models:

The 7900 XT has, relative to the 7900 XTX:

- 87.5% stream processors
- 83.33% memory
- same memory speed but 83.33% bus width
- 83.33% memory bandwidth

The 4080 12 GB has, relative to the 4080 16 GB:

- 78.95% CUDA cores
- 75% memory
- 91.3% memory speed but 75% bus width
- 68.48% memory bandwidth

Which do you think is worse?

They are both bad. They both should have different names. On the cpu side, same core silicon is used but names are changed when two cores are disabled, why is it OK on gpu side to use the same name with a single letter to differentiate between two classes of product?

Names are entirely arbitrary, but I feel that it misleads consumers when they are too similar while being different products. Nvidia is famous for it and gets raked over the coals, amd seems to get a pass? Why is that?

 

[ratirt]

RX 6900XT was double the shaders of 5700XT with half the memory bus on the same node, it is more than double the performance.
"7900XTX" or whatever is gonna be called is more than double the shaders of 6900XT with double the memory bus on a better node.
I really don't see how it is difficult to belive they can double performance.

Sure but it was not just a bump in transistors you know and it was not MCM. I hope the MCM part will not create some sort of inefficiency or a bottleneck.
NV 4080 has twice shading units than a 3080 and it is not twice as fast. Maybe the same will be with 7000 series AMD. Time will tell.
Don't get me wrong, I would really appreciate twice the performance for the same price. Would that be the case? I doubt it.

 

[anachron]

I can't say with certainty because it's not released, and it's not a linear scale between

They are both bad. They both should have different names. On the cpu side, same core silicon is used but names are changed when two cores are disabled, why is it OK on gpu side to use the same name with a single letter to differentiate between two classes of product?

Names are entirely arbitrary, but I feel that it misleads consumers when they are too similar while being different products. Nvidia is famous for it and gets raked over the coals, amd seems to get a pass? Why is that?

I would argue that names are different though. 4080 16gb vs 4080 12gb imply that it's the same card with only a different amount of memory while 7900XT vs 7900XTX even though it's a single letter is obviously not the exact same card. Just like a 3080 is not the same as a 3080Ti.
Even if the XTX name is stupid, i think pretty much anyone will understand it's a higher model than just the XT.

 

[Blue4130]

I would argue that names are different though. 4080 16gb vs 4080 12gb imply that it's the same card with only a different amount of memory while 7900XT vs 7900XTX even though it's a single letter is obviously not the exact same card. Just like a 3080 is not the same as a 3080Ti.
Even if the XTX name is stupid, i think pretty much anyone will understand it's a higher model than just the XT.

The 4080 has their same number of different characters....

And on the amd side, both use 7900, how would someone think that they are different cards?

 

[anachron]

The 4080 has their same number of different characters....

And on the amd side, both use 7900, how would someone think that they are different cards?

A RTX 4080 is a RTX 4080, 12 and 16gb is an amount of memory, people will not take it as a model number by default.
XT vs XTX is obviously part of the model name since it doesn't refer to memory amount.

 

[Blue4130]

A RTX 4080 is a RTX 4080, 12 and 16gb is an amount of memory, people will not take it as a model number by default.
XT vs XTX is obviously part of the model name since it doesn't refer to memory amount.

7900 is 7900, the model name, is it not? Sorry, but I strongly disagree with you. You are free to your opinion, but you have not convinced me. No harm, no foul.

 

[ARF]

RX 6900XT was double the shaders of 5700XT with half the memory bus on the same node, it is more than double the performance.

Weird. Both are 256-bit, the difference is 14 (448 GB/s) vs. 18 Gb/s (576 GB/s) GDDR6, and 0 vs. 128 MB Infinity Cache.

 

[anachron]

7900 is 7900, the model name, is it not? Sorry, but I strongly disagree with you. You are free to your opinion, but you have not convinced me. No harm, no foul.

By this logic a RTX 2080 is the same as a RTX 2080 TI and a RTX 2080 Super since they are all "2080". You voluntarily choose to ignore a part of the model name.

So now you will tell me that "12Gb" and "16Gb" are also part of the model names in this case. Yes, but they directly imply that it's just a memory amount change. NVIDIA did already pull this kind of shenanigans with the RTX 3080 10GB vs 12GB, but at least the specs were relatively close outside of the memory change, unlike the 4080s. They could have called the 4080 16GB a 4080TI instead and it would probably have caused less uproar.

But as you said, we can agree to disagree in the end.