AMD Reveals More Details Around The Radeon RX 7900 Series / RDNA3

[Danny3]

Well, I'm disappointed that they call these expensive GPUs gaming GPUs, while they don't support SR-IOV and many games can't even run because of that.
Wake me up when AMD will really care about gaming and game compatibility!

[WannaBeOCer]

That's a joke right? ATX12VO is basically dead except of some niche OEM stuff and you haven't seen all those burnt HPWR connectors some even without those crappy nvidia adapters?

ATX12VO is much more efficient than the current implementation. It shocks me how slow the progression is to embrace it. I’m going to be holding out upgrading my rig until ATX12VO is adopted by some higher end motherboard manufacturers. We’re already seeing them in lower end 600 series consumer boards. Won’t be long until they hit a few higher end boards.

We’ve already been informed months ago by PCI-SIG about potential issues when using non-ATX 3.0 compliant power supplies when using 12VHPWR. Using ATX 2.0 adapters/cables to 12VHPWR isn’t recommended. All PCIe gen 5.0 GPUs will be required to use 12VHPWR to meet PCI-SIG’s specifications.

[skeevy420]

Of course, I'll forego sharing all of the Windows-specific information and focus on some of the key highlights and commentary from the Linux perspective. No Linux-specific information was shared at the event so you'll need to wait for my reviews to find out about these RDNA3 graphics cards on the open-source driver.

You're becoming quite the comedy writer, Michael.

[oleid]

Well, I'm disappointed that they call these expensive GPUs gaming GPUs, while they don't support SR-IOV and many games can't even run because of that.

Huh? What are you talking about? What games need SR-IOV?

If you want that feature, get a workstation card.

[shmerl]

I hope radv developers will iron out that pipeline callables feature that breaks ray tracing in Cyberpunk 2077 so far.

[artivision]

First i'm hearing DUAL ISSUE then i'm hearing OUT OF ORDER. For those who know, the first one doesn't mean double execution (because we had that already) as Navi 2 had 2FP+1INT compressed together, but double in the same logic thread instead, like a small Cpu. Then i wonder "why the second" on such a low issue? Small Arm processors are In_Order for example, this is usually happens if they do want to achieve something like FOUR ISSUE by Software / Compute Compiler. Also each Issue probably can compress 256bit Multiple Data equal to a Single Instruction Vector Register (or 8 shaders) and like Cpu Vector Instructions. So if those facts are actually real, then is it possible that we can build normal C and C++ apps native for Navi 3 HW or IR? How much is the potential that NV dropped in the trap of the century? Is it possible that Navi 4 will speak an AVX4 subset instead?

[Hibbelharry]

Well, I'm disappointed that they call these expensive GPUs gaming GPUs, while they don't support SR-IOV and many games can't even run because of that.
Wake me up when AMD will really care about gaming and game compatibility!

Can you please name any game that doesn't run because SR-IOV is missing?

I guess thats plain nonsense.

ATX12VO is much more efficient than the current implementation. It shocks me how slow the progression is to embrace it. I’m going to be holding out upgrading my rig until ATX12VO is adopted by some higher end motherboard manufacturers. We’re already seeing them in lower end 600 series consumer boards. Won’t be long until they hit a few higher end boards.

It's dead Jim. You will literally wait forever.

We’ve already been informed months ago by PCI-SIG about potential issues when using non-ATX 3.0 compliant power supplies when using 12VHPWR. Using ATX 2.0 adapters/cables to 12VHPWR isn’t recommended. All PCIe gen 5.0 GPUs will be required to use 12VHPWR to meet PCI-SIG’s specifications.

Thats wrong. Where did you get that from? See the new AMD Cards. Two 8 pin jacks and PCIe5.0, compliant to standards.

[WannaBeOCer]

Can you please name any game that doesn't run because SR-IOV is missing?

I guess thats plain nonsense.



It's dead Jim. You will literally wait forever.



Thats wrong. Where did you get that from? See the new AMD Cards. Two 8 pin jacks and PCIe5.0, compliant to standards.

How’s it dead? Intel just started advising motherboard manufacturers to use it in 2021 and we’re already seeing lower end consumer boards with it.

The RX 7000 series is still using PCIe Gen 4. The alternative name for 12VHPWR is PCIe Gen 5 connector.

https://www.techpowerup.com/287682/pcie-gen5-12vhpwr-connector-to-deliver-up-to-600-watts-of-power-for-next-generation-graphics-cards?amp

[Hibbelharry]

The RX 7000 series is still using PCIe Gen 4. The alternative name for 12VHPWR is PCIe Gen 5 connector.

I stand corrected, you're right. The AMD Cards are PCIe4.0, so same as nvidia.

Nontheless those HPWR Connectors are problematic, I don't want to see those used in my PC. There are already enough examples of them melted down.

How’s it dead? Intel just started advising motherboard manufacturers to use it in 2021 and we’re already seeing lower end consumer boards with it.

ATX12VO was standardized in 2019, so 3 years ago. Adoption is still scarce, next to noone needs its, so no components get produced. Also see this article:

https://overclock3d.net/news/power_supply/what_journalists_got_wrong_about_intel_s_atx12vo_standard/1

[Paradigm Shifter]

AMD talking up RDNA3's compute and machine learning capabilities is terrific. This also provides hope that they will be providing ROCm support for RDNA3 sooner rather than later, unlike the delays seen for RDNA and RDNA2 consumer GPUs with belated ROCm Linux compatibility.

I really hope you're right, Michael, but the proof is in the pudding. Frankly, I've grown accustomed to disappointment from AMD's GPU division regarding GPGPU support. But I'd love them to prove my cynicism unfounded this time.

...

edit: ATX12VO, or HPWR, or whatever that 16-pin connector is on the RTX 4000 series has, has - rightly - had a lot of poor press because of the melting/burning connectors on the 4090. That alone may do for it in the long term because if issues akin to that keep cropping up - whether it is the fault of the design itself or the implementation thereof - people will likely avoid it. And while Intel are pushing people to use it, all it'll take is a datacentre fire or a house fire well publicised and that will be the end of that.