AMD: Navi Speculation, Rumours and Discussion [2019-2020]

I'm still skeptical about this magical cache, like if it was so easy to just add more cache why haven't it be done before? even in a smaller scale...But at the same time what I find more intriguing is why Sony is being so extremely secret about It's GPU...We have no pics of the chip, or even de PCI or anything. Maybe it's AMD prohibiting it's revealing until they officially lunch the architecture? Or maybe this specs are AMD doing the same as Nvidia "leaking" false specs to hide the real one.

For me I just want to be able to play at 1440p at 120FPS for less than 400 bucks.

 

That is an interesting thought. The 128CU CDNA chip with 4096HBM would be the ~500mm2 ASIC in that case.

I don't think you can draw any conclusions of GDDR6 pricing based on a +enthusiast segment GPU with a newly developed signaling spec and uses 24 GDDR6X ICs.

IMO- that has been possible for over a year now with the RTX2060S and RX5700/XT though it still depends on the game and using sweetspot IQ/performance settings. As long as you have realistic allowances of your minimums, it is doable, and Freesync definitely helps.

 

Yes but none are less than 400 bucks new and you still need to lower the quality and forget about some effects. I want a more flawless experience with some performance to spare to maintain it for a reasonable time.

 

5700XT's can be had new for under $400.

There's an MSI for 375 after rebate at newegg

 

Ok. How about all the 384bit implementations of GDDR6 non-X releasing at $2500 or more?

 

Six DevIDs (also SKUs?) for Sienna Cichlid in recent commit:

0x73A0
0x73A2
0x73A3
0x73AB
0x73AE
0x73BF

Phoronix.com - Linux 5.9 Gets More Fixes For AMD RDNA2 GPUs, Promotes Navi 12

 

My guess is, since Navi is one feature set all the way through, the lowest end cards (128bit bus?) will have Rx590-ish performance for $250-200, but that might only hit 1080p at best. If you want more than that though, well... hopefully the 6500... 6400? and 3060 will be $299.

 

Or maybe it's hard to have an efficient cache, and you need the right architecture to exploit it fully. So, it was not on the table before RDNA2.

 

You also need suitable workloads.

 

The forking of the path between gaming (RDNA) and compute (CDNA) surely would have helped in this regard.

 

Yep AMD is certainly more free to do something radical that is entirely gaming focused.

 

That already exists and it's called the RX 5500XT. There's no reason to launch such a similar card unless they can make it significantly cheaper.

 

I was thinking in 6600, 6700 for 350-400 maybe(for me that would be about 600 dollars because of taxes..... ) as it seems AMD will compete with price so I expect more cheaper prices than Nvidia,

Yeah but something like that out of nowhere? Like sure AMD can innovate, as saw that with Zen but just creating a big cache to "forget" about ram interface? I find hard to believe that Nvidia with its huge R+D bugged hasn't try it or even AMD with RDNA with a more simpler implementation like from zen 1 to zen 2 where Zen 2 is basically zen 1 but split into 3 parts.

But I'm all in to be surprise in this.

 

Hmm... Zen2 has double the FP units & associated bandwidth compared to Zen, it has improved branch predictors, µOPs cache, TLB, load/store bandwidth, L3 cache size, L2 cache latency, reorder buffer, No need for NUMA, and a lot of other improvements.

 

Yes Zen 2 is an improvement. My point was that AMD first created the modular Arch created the CCX and then move to the next step dividing them, they didn't start with the zen 2 approached. I would be surprise they didn't include a simpler version of this magical cache in the first RDNA1 and use it to boost performance.

 

Moving to chiplets allowed AMD to position themselves well vs Intel's 10nm architectures. AMD did moar-corez (16c desktop and 64c server) to compete with Intel. It's not realistic to do that utilizing NUMA.

 

intel iris pro had an edram cache of 128mb specifically for the reason of saving memory bandwidth. I could imagine amd doing something similar.

 

That is true .

Also AMD has talked about bringing over cpu designers , perhaps they are just better at optimizing large amounts of cache ?

 

Nvidia aren't really in the business of making the fastest GPUs, they're in the business of making the most profitable ones. And they excel at it.
If nVidia really wanted to bring the fastest gaming GPU they could possibly make to the PC market, they wouldn't use Samsung's 8nm node. They'd go with TSMC's N7+.
They also wouldn't use GDDR6X, they'd use HBM2.
They can't sell GA100-like gaming GPUs for $10 000 in the gaming market in enough quantities to justify the investment, so there's no such product.


Putting a lot of cache into a GPU or SoC makes it considerably larger. If your economics dictate a maximum die area, using lots of cache will eat away at the available area for execution resources and significantly lower the perf-per-mm^2 compared to a competitor that just uses faster external memory (e.g. XBOne with eDRAM vs. PS4 with wider GPU).

My point is that lots of cache isn't a magic bullet. If true, it's just a means to compensate for not using a wider VRAM bus, at the cost of having a larger chip to fabricate.