AMD Radeon RX 6000 Series Specs Leak: RX 6900 XT, RX 6800 XT, RX 6700 Series

[Vya Domus]

Not 100% true, as i know RDR2 can look better than it does with my 390X and that's up to a 420w just for 40-60 fps.

Sure but that's a really old GPU. Gotta look at things from the current generation.

 

[ebivan]

That's untrue. All AiBs have models at the same MSRP as the FE card

Sure, and those are just piling up on the shelves where you live? Come on, nobody (or almost nobody) got an 3080 for 700. Even the cheapest Zotac card is at 850 and still not available...

 

[Vya Domus]

Meanwhile I expect Nvidia are sat there with the Ti / Super variants ready for AMDs next move. No way they're letting AMD have the fastest cards come Christmas.

It will be funny to see a 3080 "ti" that's 5-8% faster.

Reality check, Nvidia probably didn't plan to use GA102 for the 3080 but rather a fully enabled GA104 like they always did which could have made room for a 3080ti. For some unknown reason (let's leave it at that for the time being) they decided GA104 wasn't enough so the whole stack got pushed up. There is no room for "ti" variants, or at least not for any that would matter. There is potential room for a 3070ti but that's because they disabled more SMs than usual, the 3070 was meant to be the 3080 and 3090 the Titan.

 

[Zach_01]

It will be funny to see a 3080 "ti" that's 5% faster.

Most probably they have nothing except the 20GB variants. When they move to TSMC’s 7nm I guess they will.

 

[efikkan]

That's usually true but unlike the FE which is a high-quality design with binned silicon and an expensive cooler, the MSRP uses the standard, cheap reference design and likely the cheapest-to-produce cooler they can get away with. So, compared to the FE it's terrible value for money.

Nonsense, there are excellent models like the Asus RTX 3080 TUF, unless you are going to overclock, which most buyers don't do.

Sure, and those are just piling up on the shelves where you live? Come on, nobody (or almost nobody) got an 3080 for 700. Even the cheapest Zotac card is at 850 and still not available...

Unprecedented demand doesn't mean almost nobody got/gets their products.

 

[ebivan]

Yeah sure, its not a supply issue Mr Huang

 

[efikkan]

Yeah sure, its not a supply issue Mr Huang

There is no reason to be snarky
Now you're down to semantics. You can certainly call it a supply issue when demand is significantly higher than the supply, as long as you don't mislead people into thinking there are production or shipment issues (which are different kinds supply issues), which several earlier posts have alluded to.

 

[MxPhenom 216]

nVidia messed up with Ampere, high prices (MSRP of 1000$ here for no logical reason), only 10gb of vram, and worse the overpriced cards don't even exist in the real life. I really hope AMD will have a real launch with reasonable prices.

I mean US prices were actually pretty surprising to me. Not nearly as much as I thought they would be for the performance.

Where they screwed the pooch I think was going with Samsung 8nm. Its just an extension of their 10nm which wasn't very good anyways.
Samsung yields are shit and GDDR6X is in short supply too.

With reports that Nvidia will be switching to TSMC 7nm for possible Super variants of the cards I think things will turn around for Ampere.

10GB of vram is fine. GDDR6X too and a lot more bandwidth.

Having said all that, my new build early next year with a 5800x, will probably also get a 6900XT.

 

[medi01]

...when demand is significantly higher than the supply...

Which can happen for 2 reasons:
1) Demand is too high
2) Supply is terribly bad

One needs hell of a green reality distortion field not to notice #2 (not even pre-orders are accepted)

 

[r9]

I mean US prices were actually pretty surprising to me. Not nearly as much as I thought they would be for the performance.

Where they screwed the pooch I think was going with Samsung 8nm. Its just an extension of their 10nm which wasn't very good anyways.
Samsung yields are shit and GDDR6X is in short supply too.

With reports that Nvidia will be switching to TSMC 7nm for possible Super variants of the cards I think things will turn around for Ampere.

10GB of vram is fine. GDDR6X too and a lot more bandwidth.

Having said all that, my new build early next year with a 5800x, will probably also get a 6900XT.

Nvidia released a card 30% faster than a $1200 card for $699 and they expected to sell only 5 cards. Lol

 

[MxPhenom 216]

Which can happen for 2 reasons:
1) Demand is too high
2) Supply is terribly bad

One needs hell of a green reality distortion field not to notice #2 (not even pre-orders are accepted)

Nvidia released a card 30% faster than a $1200 card for $699 and they expected to sell only 5 cards. Lol

Mhmm.

From what I understand, GDDR6X is in very short supply. Only Micron is making it right now. Also Samsung 8nm (10nm+) yields are pretty bad.

Or Nvidia is purposefully reducing the supply. If thats the case, itll just hurt Nvidia in the end since most people will just buy these new AMD cards if the performance is there.

 

[medi01]

From what I understand, GDDR6X is in very short supply.

Ah, sure thing, dude.
It is GDDR6x and not that oversized GA102 chip that is the reason.

Surely, $699 for a GPU 20-30% aster than $1200 card is nothing unusual either, totally expected and there are no reasons to see that something doesn't quite add up here.

 

[MxPhenom 216]

Ah, sure thing, dude.
It is GDDR6x and not that oversized GA102 chip that is the reason.

Surely, $699 for a GPU 20-30% aster than $1200 card is nothing unusual either, totally expected and there are no reasons to see that something doesn't quite add up here.

Had Nvidia been able to use TSMC I think the supply issues would be considerably less.

Demand for new cards this year seems way way higher than previously and I'm not entirely sure why.

 

[ebivan]

There is no reason to be snarky
Now you're down to semantics. You can certainly call it a supply issue when demand is significantly higher than the supply, as long as you don't mislead people into thinking there are production or shipment issues (which are different kinds supply issues), which several earlier posts have alluded to.

No reason to defend nvidia either. They screwed up. They made promises they could not deliver.
They didnt expect people to buy that extremely well priced extremely fast card? Who calculates expected demands over there? First year business school students? No, they made promises which they knew they wouldnt be able to keep.
Of course there are production issues, how could you deny that?
It's not importent if its Samsungs or Microns fault that there are almost no cards available. Nvidia wasnt ready to launch and they did anywasy. I really hope AMD sells a shitload of cards while Nvidia still cannot deliver for months!

 

[Umbral]

Possible theoretical AMD strategy:

6900 XT = 5% slower than 3090, $650 16GB RAM

6800 XT = 5% slower than 3080, 5% slower than 6900XT $450 16 GB RAM
6800 = 10~15% faster than 3070, 5% slower than 6800 XT $400 16 GB RAM

6700 = 10 % faster than 3060 TI, 10% slower than 3070 $325 12 GB RAM

AMD wins at price/performance at every tier. Not to mention 99% games optimized for consoles on AMD hardware.

Best Card for price/performance = 6800 vanilla

 

[efikkan]

No reason to defend nvidia either. They screwed up. They made promises they could not deliver.

What precisely did they screw up? Which exact promises did they fail to deliver on?

They didnt expect people to buy that extremely well priced extremely fast card?

They had tens of thousands available on launch day, comparable to the launch of Pascal and Turing, but that's not enough when the demand is probably >10x the supply. It's not much of a "screwup" considering there probably shipped more RTX 3080 cards on launch day than AMD's "flagship" Radeon VII did in its entire production run.

Who calculates expected demands over there? First year business school students? No, they made promises which they knew they wouldnt be able to keep.

You don't know much about how microchips are made do you?
It's not a matter of wanting to produce chips, the limit is wafer capacity in the foundries. You can be sure Nvidia ordered as much as they could. If you want to blame someone, blame Samsung who didn't make the factory larger when they planned it >5 years ago.

It's not importent if its Samsungs or Microns fault that there are almost no cards available. Nvidia wasnt ready to launch and they did anywasy.

Logic fail.
Waiting wouldn't have increased the total number of cards shipped to date.

I really hope AMD sells a shitload of cards while Nvidia still cannot deliver for months!

I certainly do hope AMD makes good cards, and for once make decent volumes. AMD cards have been in much shorter supply than Nvidia cards in the past five years, and very little of that can be blamed on miners.

 

[AsRock]

It will be funny to see a 3080 "ti" that's 5-8% faster.

Reality check, Nvidia probably didn't plan to use GA102 for the 3080 but rather a fully enabled GA104 like they always did which could have made room for a 3080ti. For some unknown reason (let's leave it at that for the time being) they decided GA104 wasn't enough so the whole stack got pushed up. There is no room for "ti" variants, or at least not for any that would matter. There is potential room for a 3070ti but that's because they disabled more SMs than usual, the 3070 was meant to be the 3080 and 3090 the Titan.

LOL, the 3090 is not a TITAN, nVidia point at it like if it was but it's not.

 

[Zach_01]

Ah, sure thing, dude.
It is GDDR6x and not that oversized GA102 chip that is the reason.

I bet its neither of those...

Surely, $699 for a GPU 20-30% aster than $1200 card is nothing unusual either, totally expected and there are no reasons to see that something doesn't quite add up here.

Now you are getting somewhere...

1. Ampere was not ready for September, but just wanted to create hype before RNDA2
2. nVidia will supply the market when they are ready for the 20GB models of 1000+$, and that was the goal from start. After the prices of RDNA2 of course.


Good supply but great demand...!!! I'm laughing...
nVidia did not fail...
Just didnt want to sell a 700$ MSRP flagship GPU...

 

[ebivan]

What precisely did they screw up? Which exact promises did they fail to deliver on?


They had tens of thousands available on launch day, comparable to the launch of Pascal and Turing, but that's not enough when the demand is probably >10x the supply. It's not much of a "screwup" considering there probably shipped more RTX 3080 cards on launch day than AMD's "flagship" Radeon VII did in its entire production run.


You don't know much about how microchips are made do you?
It's not a matter of wanting to produce chips, the limit is wafer capacity in the foundries. You can be sure Nvidia ordered as much as they could. If you want to blame someone, blame Samsung who didn't make the factory larger when they planned it >5 years ago.


Logic fail.
Waiting wouldn't have increased the total number of cards shipped to date.


I certainly do hope AMD makes good cards, and for once make decent volumes. AMD cards have been in much shorter supply than Nvidia cards in the past five years, and very little of that can be blamed on miners.

Yeah sure man, you're right and I'm wrong!
I just clicked on your name and read the last coupe of posts you wrote. Our little chat ends right here, as i don't really see you as someone who discusses. You are someone who is just always right and I can't compete with that. And I really share none of the opinions I read you have.

 

[R0H1T]

I certainly do hope AMD makes good cards, and for once make decent volumes. AMD cards have been in much shorter supply than Nvidia cards in the past five years, and very little of that can be blamed on miners.

Well technically speaking AMD's selling at least 2-5x as many GPUs as Nvidia, Intel probably outsells both. So it's not an Apples to Apples comparison, even if AMD were superior to Nvidia they'd have (nearly) double the issues keeping up with supplies of APU, CPU, consoles, embedded chips & finally dGPU - notice the last one probably has the least volumes.

 

[Yazzia]

Possible theoretical AMD strategy:

6900 XT = 5% slower than 3090, $650 16GB RAM

6800 XT = 5% slower than 3080, 5% slower than 6900XT $450 16 GB RAM
6800 = 10~15% faster than 3070, 5% slower than 6800 XT $400 16 GB RAM

6700 = 10 % faster than 3060 TI, 10% slower than 3070 $325 12 GB RAM

AMD wins at price/performance at every tier. Not to mention 99% games optimized for consoles on AMD hardware.

Best Card for price/performance = 6800 vanilla

Why would AMD sell a card basically the same as the 3090 for less than half the price? The delusion is real

 

[bug]

Look Infinity Cache Patent details:

"We propose shared L1 caches in GPUs. To the best of our knowledge, this is the first paper that performs a thorough characterization of shared L1 caches in GPUs and shows that they can significantly improve the collective L1 hit rates and reduce the bandwidth pressure to the lower levels of the memory hierarchy."

• "We develop GPU-specific optimizations to reduce inter-core communication overheads. These optimizations are vital for maximizing the benefits of the shared L1 cache organization."

• "We develop a GPU-specific lightweight dynamic scheme that classifies application phases and reconfigures the L1 cache organization (shared or private) based on the phase behavior."

• "We extensively evaluate our proposal across 28 GPGPU applications. Our dynamic scheme boosts performance by 22% (up to 52%) and energy efficiency by 49% for the applications that exhibit high data replication and cache sensitivity without degrading the performance of the other applications. This is achieved at a modest area overhead of 0.09 mm2 /core."

ADAPTIVE CACHE RECONFIGURATION VIA CLUSTERING - ADVANCED MICRO DEVICES, INC.

<div p-id="p-0001">A method of dynamic cache configuration includes determining, for a first clustering configuration, whether a current cache miss rate exceeds a miss rate threshold. The first cluste

www.freepatentsonline.com

https://adwaitjog.github.io/docs/pdf/sharedl1-pact20.pdf

Well, if it's L1 cache, it's going to be tiny. And please note they narrow down the specific scenario where that cache helps.
Don't get me wrong, caching works, it improves performance. But it's different from bandwidth. I mean, look at Intel's CPUs. Those have had pretty smart caches (three levels of them) that do a pretty good job. But they still only hide latency, they're not a substitute for bandwidth.
Of course, GPU usage patterns are "a little" different, but I hope you get the idea.

 

[Sp408]

Can someone explain the difference between the 8nm and 7nm? Does that actually make a difference in performance or anything important?

 

[bug]

Can someone explain the difference between the 8nm and 7nm? Does that actually make a difference in performance or anything important?

It's an important distinction, but it's too complicated to explain here. Read up on silicon manufacturing nodes if you have a week or so to spare

 

[R0H1T]

Can someone explain the difference between the 8nm and 7nm? Does that actually make a difference in performance or anything important?

Hard to say, unless Nvidia releases the exact same cards on 7nm we'll never really know how good TSMC's node is, it could be better or worse for Ampere. On most objective parameters TSMC is superior but again we know this isn't how it works in real life. Just look at Intel's 10nm debacle for instance.