AMD "Matisse Refresh" Processor SKUs Include 3900XT, 3800XT, and 3600XT

[ARF]

Yes and that's why no one can give you the answer to

I know the answer, I was just arguing to the guy who expects AMD not to compete when Dr Lisa Su explicitly said that AMD will be back in the enthusiast market.
And other AMD personnel said that you must expect 4K level disruptive performance from Navi 2X.

 

[Vayra86]

A simple problem:

Navi 10 is 251 sq. mm with performance 100% and performance per watt 100%.

How much faster will Navi 21 be if it is 505 sq. mm with (>) 50% higher performance per watt ?

Not a problem. If you think that somehow translates to double perf you got it wrong. And it does need double perf to kill the Nvidia stack.

We have been here a half dozen times before, overhyped AMD GPU prior and pretty weak product post leunch. Will it be different? I do think it will not be shit, as RDNA was already not shit. But performance crown? Nah, not happening...

 

[ARF]

Not a problem.

A problem.

"A mathematical problem is a problem that is amenable to being represented, analyzed, and possibly solved, with the methods of mathematics."

mathematical problems - Google Search

But performance crown? Nah, not happening...

Nvidia has only 2080 Ti with no where in sight anything better than it..
Performance crown is quite easy to achieve today, Nvidia has exactly zero N7 GPUs as of today.

 

[efikkan]

The performance crown might be obtainable if Nvidia's next gen is somehow delayed. But when it arrives, it could offer anywhere from ~30-80% more performance. It's very hard to predict considering it's a new major architecture and transition to a new mature node.

But honestly, I have more expectations from Zen 3 than RDNA2…

 

[GeorgeMan]

A problem.

"A mathematical problem is a problem that is amenable to being represented, analyzed, and possibly solved, with the methods of mathematics."

mathematical problems - Google Search

Nvidia has only 2080 Ti with no where in sight anything better than it..
Performance crown is quite easy to achieve today, Nvidia has exactly zero N7 GPUs as of today.

Isn't ampere to launch on September?

 

[Elysium]

I'm...surprised. Not really sure why these chips are coming out. Gaming-focused? Meh? I mean, did they absolutely need products to fight Comet Lake? The way AMD supporters themselves constantly, constantly talk about AMD's chips coming within a few percentage points of Intel for gaming purposes, you'd think this stack would be DoA because no-one should be buying them if they come within a few points and Zen 3 is less than six months away...right? Bit of a waste of silicon really.

 

[ARF]

The performance crown might be obtainable if Nvidia's next gen is somehow delayed. But when it arrives, it could offer anywhere from ~30-80% more performance. It's very hard to predict considering it's a new major architecture and transition to a new mature node.

But honestly, I have more expectations from Zen 3 than RDNA2…

Zen 3 will be the swan song of the AM4 platform. I honestly don't expect anything from it. They will not surprise us with Zen 3.
Zen 4 will be much more important milestone in the development of the Zen-based micro-architectures

RDNA 2 is also very important because AMD needs to become competitive in the graphics department again.
We need that innovation with something new in the graphics, too.

Isn't ampere to launch on September?

Maybe, maybe in March 2021, who knows... We will see but honestly why would launches from the corps coincide?

I'm...surprised. Not really sure why these chips are coming out. Gaming-focused? Meh? I mean, did they absolutely need products to fight Comet Lake? The way AMD supporters themselves constantly, constantly talk about AMD's chips coming within a few percentage points of Intel for gaming purposes, you'd think this stack would be DoA because no-one should be buying them if they come within a few points and Zen 3 is less than six months away...right? Bit of a waste of silicon really.

Yeah, pretty much
Refresh of Zen 2 is not needed at this time. Many people said the Comet Lake was DoA. If it was DoA, why would AMD need to act now?

 

[efikkan]

Zen 3 will be the swan song of the AM4 platform. I honestly don't expect anything from it. They will not surprise us with Zen 3.
Zen 4 will be much more important milestone in the development of the Zen-based micro-architectures

Honestly I don't know, there are so much contradicting information on Zen 3, but if it's anywhere close to the 20% IPC gains some are claiming, then it's a major overhaul which should excite everyone.

AMD doesn't plan to use DDR5 until 2022, so it makes me wonder if they will delay Zen 4 until 2022, or if it will support both memory types. Not that DDR5 matters to me, I'm just curious.

I have one machine overdue for a replacement, and another which is getting close. The next 1-1.5 years will be exciting, perhaps I'll build one machine from each team?

RDNA 2 is also very important because AMD needs to become competitive in the graphics department again.

We need that innovation with something new in the graphics, too.

I don't disagree with the need for competition and innovation in the graphics market. But my expectations of RDNA 2 is not that high, AMD have disappointed again and again.

 

[TheoneandonlyMrK]

The performance crown might be obtainable if Nvidia's next gen is somehow delayed. But when it arrives, it could offer anywhere from ~30-80% more performance. It's very hard to predict considering it's a new major architecture and transition to a new mature node.

But honestly, I have more expectations from Zen 3 than RDNA2…

If you're just guessing based on Huang it's 20/400% better, we will see eh.
And it's less new than it seams both no all companies iterate even Ryzens had some up carry over from the BD line, no one starts from scratch.

Literally any thread Can become the GPU argument thread on a whim.

 

[ValenOne]

The problem with that logic is that those things dont scale linearly

Near-linear scale can be reached e.g. 7870 (dual geometry/dual rasterization units, 20 CU, 32 ROPS, 256-bit bus) scaling into R9-390 (quad geometry/quad rasterization units, 40 CU, 64 ROPS, 512-bit bus).

7870 = dual shader engines GCN
R9-290/290X = quad shader engines GCN
...
RX 5700/5700 XT = dual shader engines RDNA

 

[btarunr]

What are your thoughts on the reasons for AMD to bump 3xxx series, while so close to Zen3 release?

Zen 3 cycle begins with a September announcement, and mid-to-late Q4 availability, with high availability only by December-January, if not later. That's too much ground ceded to Intel marketing on the "gaming" plank.

As reviews have shown, 10900K only enjoys a mid-thru-high single digit gaming performance lead, and a 10-15% nT non-gaming performance deficit over the 3900X. So AMD's goal would be to bring Intel's gaming performance lead down as much as possible (at least restore the ~2.5-4% lead it had with CFL), and push its nT performance lead close to 20%.

They weren't embattled before and they sure as hell ain't now.

18-22% price-cuts suggest otherwise.

 

[Caring1]

18-22% price-cuts suggest otherwise.

Or it's an indicator that pricing is way too high in the market, and they can increase sales while maintaining profitability by reducing their prices.

 

[InVasMani]

Honestly I don't know, there are so much contradicting information on Zen 3, but if it's anywhere close to the 20% IPC gains some are claiming, then it's a major overhaul which should excite everyone.

AMD doesn't plan to use DDR5 until 2022, so it makes me wonder if they will delay Zen 4 until 2022, or if it will support both memory types. Not that DDR5 matters to me, I'm just curious.

I have one machine overdue for a replacement, and another which is getting close. The next 1-1.5 years will be exciting, perhaps I'll build one machine from each team?


I don't disagree with the need for competition and innovation in the graphics market. But my expectations of RDNA 2 is not that high, AMD have disappointed again and again.

I'd actually prefer not to have to jump to DDR5 immediately if it can be helped. It can defiantly get expensive quickly when you need to replace not just the CPU and MB, but the memory as well.

 

[Vayra86]

Zen 3 will be the swan song of the AM4 platform. I honestly don't expect anything from it. They will not surprise us with Zen 3.
Zen 4 will be much more important milestone in the development of the Zen-based micro-architectures

RDNA 2 is also very important because AMD needs to become competitive in the graphics department again.
We need that innovation with something new in the graphics, too.




Maybe, maybe in March 2021, who knows... We will see but honestly why would launches from the corps coincide?




Yeah, pretty much
Refresh of Zen 2 is not needed at this time. Many people said the Comet Lake was DoA. If it was DoA, why would AMD need to act now?

Its all crystal ball category statements, isn't it. From both of us I'm just a bit more of a pessimist, it seems to end up closer to the truth every time... Not just with AMD - Nvidia too. You won't find me overhyping the next release in ANY camp - even Pascal was better than expected, but the price increase went along with the additional performance that gen offered. So bottom line... the next gen will be yet another next gen like every other, carefully balanced out against the current economic and market reality. There will be some GPUs that are great value, and others that are overpriced, and the stacks will again be extremely close to one another, for the most part.

It sounds boring, and yes, it really is The interesting bits for next gen will be how each camp scales with RT content and without it, while using their architectural update. How large the dies will be, how power management affects 'quality of life'... and how the perf/dollar will land. Whoever has the fastest... the vast majority is not in the market for it anyway. That 2080ti is really not a product in the stack IMO, it is priced way out of it. The details matter a lot more, and so far AMD has yet to show us details that truly bring benefits on the points mentioned here. RDNA was a small step in that direction, but still massively behind the curve.

When it comes to CPUs though, AMD has a much more reliable track record as of late, and it is actually them that are leading the curve, despite a minor, situational performance deficit (in ST). Intel has more catching up to do here; in die size/yield/node, in power management (quality of life - enter the delid or solder the lid - still too hot, scales like shit when turboing)... The metrics are really the same, and none of them include me believing any sort of hype or marketing.

Don't think anyone disagrees about competitiveness, but saying it often won't make it so.

 

[efikkan]

I'd actually prefer not to have to jump to DDR5 immediately if it can be helped. It can defiantly get expensive quickly when you need to replace not just the CPU and MB, but the memory as well.

Usually, I keep CPU, motherboard and memory together when upgrading computers, reusing it in another build as a secondary computer, file server, etc. But if you have a lot of memory, it's understandable that it will cost a lot to replace it.
More importantly, I see far too many builders waste money on overpriced overclockable memory. Using something like 4000 MHz memory is not going to yield a substantial gain, and you are probably not able to run it reliably at that speed for 5+ years. People should rather buy affordable memory running at the highest JEDEC speed of their CPU and spend the money save on a CPU or GPU that's one tier higher.

 

[ARF]

Its all crystal ball category statements, isn't it. From both of us I'm just a bit more of a pessimist, it seems to end up closer to the truth every time... Not just with AMD - Nvidia too. You won't find me overhyping the next release in ANY camp - even Pascal was better than expected, but the price increase went along with the additional performance that gen offered. So bottom line... the next gen will be yet another next gen like every other, carefully balanced out against the current economic and market reality. There will be some GPUs that are great value, and others that are overpriced, and the stacks will again be extremely close to one another, for the most part.

It sounds boring, and yes, it really is The interesting bits for next gen will be how each camp scales with RT content and without it, while using their architectural update. How large the dies will be, how power management affects 'quality of life'... and how the perf/dollar will land. Whoever has the fastest... the vast majority is not in the market for it anyway. That 2080ti is really not a product in the stack IMO, it is priced way out of it. The details matter a lot more, and so far AMD has yet to show us details that truly bring benefits on the points mentioned here. RDNA was a small step in that direction, but still massively behind the curve.

When it comes to CPUs though, AMD has a much more reliable track record as of late, and it is actually them that are leading the curve, despite a minor, situational performance deficit (in ST). Intel has more catching up to do here; in die size/yield/node, in power management (quality of life - enter the delid or solder the lid - still too hot, scales like shit when turboing)... The metrics are really the same, and none of them include me believing any sort of hype or marketing.

Don't think anyone disagrees about competitiveness, but saying it often won't make it so.

The part which I don't like is that Zen 3 is end of road for AM4 support. No possible future upgrade on the line with anything.
It's a very serious reason not to buy anything CPU this year or H1 next year, and if you can wait for the new AM5 platform, then better wait.

 

[ValenOne]

The part which I don't like is that Zen 3 is end of road for AM4 support. No possible future upgrade on the line with anything.
It's a very serious reason not to buy anything CPU this year or H1 next year, and if you can wait for the new AM5 platform, then better wait.

DDR5 is different from DDR4. Higher memory bandwidth would be required when AMD supports AVX-512 or quad pipeline 256-bit AVX 2 FMA3 with 16 CPU cores. Refer to X299 platform's four DDR4 memory channels as an example.

 

[efikkan]

The part which I don't like is that Zen 3 is end of road for AM4 support. No possible future upgrade on the line with anything.
It's a very serious reason not to buy anything CPU this year or H1 next year, and if you can wait for the new AM5 platform, then better wait.

Well, the only thing that can't be upgraded would be the CPU.
Frankly, I seem to disagree with many in terms of CPU support. The reality with AM4 support ended up being having decent support for ~1 more generation, then perhaps partial support with some major compromises beyond that. For this support to matter to me, it needs to last at least 3-4 years of decent support and properly maintained BIOSes. There is just no point in upgrading a CPU after 1 year, that's a waste of money. The sad truth is that motherboard BIOSes are barely maintained beyond 1 year, so until they come up with motherboards which commits to proper long term support, I'm going to buy a new motherboard whenever I need a new CPU.

So for me, the choice between Zen 3 or Zen 4 would not be about sockets, but performance and CPU features (like AVX-512?).

 

[ARF]

I haven't seen any hint that AMD intends to support AVX-512, when even Intel who created it pretty much dropped it as an option.

 

[efikkan]

I haven't seen any hint that AMD intends to support AVX-512, when even Intel who created it pretty much dropped it as an option.

Neither have I, but they better bring it sooner rather than later. Once some productive tools starts to use it, it will be a big deal. Even Ice Lake-U features AVX-512, so all of Intel's segments should feature it going forward (except Atom etc.).

 

[ARF]

Neither have I, but they better bring it sooner rather than later. Once some productive tools starts to use it, it will be a big deal. Even Ice Lake-U features AVX-512, so all of Intel's segments should feature it going forward (except Atom etc.).

As of 2020, there are no AMD CPUs that support AVX-512, and AMD has not yet released plans to support AVX-512.

Advanced Vector Extensions - Wikipedia

en.wikipedia.org