AMD Radeon RDNA2 Navi (RX 6500, 6600, 6700, 6800, 6900 XT)

There is a piece of this puzzle that we are forgetting...the drivers.
In 2019 the 5700xt struggles to keep pace with the 2070s in certain games.
https://i.ibb.co/Bn0myVB/outter-worlds-slow.jpg
https://i.ibb.co/99CbJcN/jedifallenorder-slow.jpg
https://i.ibb.co/vJTKwrq/gears-5-slow.jpg


I didn't see improvements to the 5700xt until early this year, 2020. 5-6 months after release and after their December mega driver release. And, it was on newer titles that the 5700xt was matching or beating a 2070s. No fanfare. No ticker-tate parade.

https://i.ibb.co/Bqt9hjc/sws1440p.jpg
https://i.ibb.co/JzYNRkB/HD0.jpg
https://i.ibb.co/KbjdhrF/Gears-5.jpg
https://i.ibb.co/Zz056G1/F1-2020.jpg
https://i.ibb.co/XF7ZJdX/death-stranding.jpg

The only game that seems to struggle for 5700xt is Watch Dogs: Legions. However, per the dev's own release notes, today, they optimized for the RX 30 series (a bit sus. IMO). And still people are reporting problems with that game. So, I treat it as a one off until it's finally optimized for RX series.

Lets not forget:
5700xt: 40 CUs, 64 ROPs, 160TMUs, 2560 Shading Units, 10.3B Transistors @ 225watts with a 251 mm² die size
2070s : 40 SMs, 64 ROPs, 160TMUs, 2560 Shading Units, 13.6B Transistors @ 215watts with a 545 mm² die size

The point is that it's very possible that RTG will optimize the performance of the RX 6000 series as times goes on. I can only hope it doesn't take as long as the 5700xt.
 
I don't think PCIe Resizable BAR is exclusive to RDNA architecture either; need to look into Linux driver code for a list of supporting hardware though.
It's not exclusive to RDNA, it goes back all the way to the first GCN. See the amdgpu_device_resize_fb_bar() function in AMD's Linux kernel mode driver.


Yes, it's not about the bandwidth, it's about cache coherence.

If the GPU can use system memory just like its own local video memory (and vice versa, the CPU can use local video memory as if it was system memory), you have to either synchronise the GPU and CPU caches using some cache coherence protocol over PCIe bus - preferably something more complex than bus snooping - or completely disable caching for this physical memory pool, with a detrimental effect on perfromance.
For GPU access to system memory, all modern x86 CPUs have automatic cache coherence via snooping built into their PCIE controllers. You can see in the Vulkan GpuInfo database that any system memory heaps (those without the DEVICE_LOCAL bit) all have the HOST_COHERENT bit set, meaning any GPU writes to system memory are automatically coherent with the CPU, no matter whether caching is enabled or not.

Going the other way, CPU access to GPU memory on AMD GPUs is always considered coherent, but not automatically in hardware. Instead it's because the kernel mode driver explicitly flushes/invalidates the GPU's "host data path" caches every time a command buffer is submitted from user space. Thus to user-space drivers like Vulkan/DX12 it appears that anything they wrote to GPU memory gets seen by the GPU automatically when the command buffers execute.

I guess AMD in only making a big deal about it now because resizable BARs require BIOS / motherboard / OS support that's fairly recent, and PCIE 4.0 finally has high enough bandwidth that direct CPU access to GPU memory is starting to make sense for more use cases.
 
Lets not forget:
5700xt: 40 CUs, 64 ROPs, 160TMUs, 2560 Shading Units, 10.3B Transistors @ 225watts with a 251 mm² die size
2070s : 40 SMs, 64 ROPs, 160TMUs, 2560 Shading Units, 13.6B Transistors @ 215watts with a 545 mm² die size
This is dishonest comparison. 2070S is heavily cut down chip, 5700 XT isn't. Those transistors are for the full chip.
 
This is dishonest comparison. 2070S is heavily cut down chip, 5700 XT isn't. Those transistors are for the full chip.
Then you can take it up with:
https://www.techpowerup.com/gpu-specs/geforce-rtx-2070-super.c3440
https://www.anandtech.com/show/14586/geforce-rtx-2070-super-rtx-2060-super-review
https://www.guru3d.com/articles-pages/geforce-rtx-2060-and-2070-super-review,4.html
etc, etc where I obtained the information. But to be honest, I seriously doubt anyone is trying to be "dishonest".
:D
 
There is a piece of this puzzle that we are forgetting...the drivers.
In 2019 the 5700xt struggles to keep pace with the 2070s in certain games.
https://i.ibb.co/Bn0myVB/outter-worlds-slow.jpg
https://i.ibb.co/99CbJcN/jedifallenorder-slow.jpg
https://i.ibb.co/vJTKwrq/gears-5-slow.jpg


I didn't see improvements to the 5700xt until early this year, 2020. 5-6 months after release and after their December mega driver release. And, it was on newer titles that the 5700xt was matching or beating a 2070s. No fanfare. No ticker-tate parade.

https://i.ibb.co/Bqt9hjc/sws1440p.jpg
https://i.ibb.co/JzYNRkB/HD0.jpg
https://i.ibb.co/KbjdhrF/Gears-5.jpg
https://i.ibb.co/Zz056G1/F1-2020.jpg
https://i.ibb.co/XF7ZJdX/death-stranding.jpg

The only game that seems to struggle for 5700xt is Watch Dogs: Legions. However, per the dev's own release notes, today, they optimized for the RX 30 series (a bit sus. IMO). And still people are reporting problems with that game. So, I treat it as a one off until it's finally optimized for RX series.

Lets not forget:
5700xt: 40 CUs, 64 ROPs, 160TMUs, 2560 Shading Units, 10.3B Transistors @ 225watts with a 251 mm² die size
2070s : 40 SMs, 64 ROPs, 160TMUs, 2560 Shading Units, 13.6B Transistors @ 215watts with a 545 mm² die size

The point is that it's very possible that RTG will optimize the performance of the RX 6000 series as times goes on. I can only hope it doesn't take as long as the 5700xt.

That was an impressive exercise in cherry picking. My recollection is that the 5700xt was pretty competitive with the 2070 super when the latter first launched.
 
I am. There's no way I believe that in some games the 6800 will give me higher FPS than a 3090 until I see it in reviews. But if it's true...then damn AMD really deliver here.

Is there an announcement on when the NDA will finish?

Let's say average clock of the 3090 is 1900 MHz. It has 112 rops so that's 213 GPixels/s
Let's say the average game clock of a RX 6800 is 1815. It has 96 rops so that's 174 GPixels/s

Pretty interesting. Means it's not a fillrate advantage or a a compute advantage. Shouldn't be texture sampling either. Definitely not bandwidth. 3090 should even have a geometry advantage. I wonder if the difference is a straight up power limit in some way. Wonder if a 3090 undervolted with roughly the same clocks would actually perform better. I am very curious. Does SAM actually a dress a significant bottleneck that gone under the radar? Or are some of these games very cache friendly in some way, where the infinity cache leads to big gains.
 
Let's say average clock of the 3090 is 1900 MHz. It has 112 rops so that's 213 GPixels/s
Let's say the average game clock of a RX 6800 is 1815. It has 96 rops so that's 174 GPixels/s

Pretty interesting. Means it's not a fillrate advantage or a a compute advantage. Shouldn't be texture sampling either. Definitely not bandwidth. 3090 should even have a geometry advantage. I wonder if the difference is a straight up power limit in some way. Wonder if a 3090 undervolted with roughly the same clocks would actually perform better. I am very curious. Does SAM actually a dress a significant bottleneck that gone under the radar? Or are some of these games very cache friendly in some way, where the infinity cache leads to big gains.

We will see in a month. maybe be drivers, or settings or another 10 diff reasons.


upload_2020-10-30_19-4-8.png



No reviews? If true this doesn't look good at all...
 
Lol no one could miss the sarcasm in that post. I'm really curious to see 3rd party benchmarks.

Well now I'm really doubting AMD's numbers. This is only done when you want to hide ur performance rather than showing it..

Looks like now we know why we didn't have an NDA date.
 
I'd actually say AMD is very confident in the launch if they decided to publish such extensive bench results now without starting pre-orders and the other stuff... They'd get shredded if something is way too off when someone tries to replicate their numbers two weeks later.
 
Reviews should be good. I wonder if default clocks were used on the latest benchmarks or if there were any overclocks.
 
I am. There's no way I believe that in some games the 6800 will give me higher FPS than a 3090 until I see it in reviews. But if it's true...then damn AMD really deliver here.

Is there an announcement on when the NDA will finish?
Why not? Some games a Vega 64 can match or beat a 1080ti.
 
Main thing I'm curious for from 3rd party reviews is frame times , minimums.

RT I don't think we're going to get anything too meaningful until we see games developed in future , targeting both , I.e not under RTX SDK or whatever . Are any of the existing implementations even compatible?
 
Main thing I'm curious for from 3rd party reviews is frame times , minimums.

RT I don't think we're going to get anything too meaningful until we see games developed in future , targeting both , I.e not under RTX SDK or whatever . Are any of the existing implementations even compatible?

On paper their all DXR games, they should run on rdna2.
 
Back
Top