• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

Rumor: Radeon 6900 XT to feature 2.4ghz clockspeed

RoboFu

One of the green rats
that’s pretty neat if true. I’ll was as Keats getting one for the 16 gigs of vram and I always like to have a gpu from each camp for dev work.
 
Last edited:

Rentahamster

Rodent Whores
How so, dude? PS5 GPU (technically an APU) is based off RDNA2. Yes, a pc GPU part has dedicated cooling. But it is still based off the same architecture. We can make an assumption based on these rumors that RDNA2 was designed to run at higher clocks.
A discrete GPU has a lot of different needs and purposes than an APU. Just saying that they're both RDNA2 as if that means something is way oversimplifying the situation.
 

sendit

Member
A discrete GPU has a lot of different needs and purposes than an APU. Just saying that they're both RDNA2 as if that means something is way oversimplifying the situation.

You're over complicating it. A APU is simply a combined GPU+CPU chip made to solve the same problem as a discreet GPU + CPU. The end goal is the same depending on the use case.
 

LordOfChaos

Member
That's exactly between 3070 and 3080..

Ampere Tflops are weird. Counted the old way, say a 3080 would be 21.6Tflops, and this is how we'd get back to the traditional situation where for a given performance Nvidia has less flops on paper than AMD. Nvidia reversed that rule by way of Ampere. So don't get too quick to compare them basically.


Also ignore any CGI artists that tell you flops are the only relevant metric in all of graphics pipelines.
 
Last edited:

Eliciel

Member
Ampere Tflops are weird

Oh wow, okay. It's going to be an exciting end of October then I guess for anyone waiting for his decision which Card to buy
 

01011001

Banned
If true, so much for the PS5's GPU being overclocked. It looks like it was downclocked. :messenger_open_mouth:

you can't compare APU clocks with graphics card clocks.

these chips are housing a CPU cluster and a GPU cluster.
the cooling solution has to keep them both under control.

in a PC the CPU and the GPU are far apart from eachother so one doesn't influence the other nearly as much when it comes to heat. and both can have their own dedicated heat sink and even fan
 
Ampere Tflops are weird. Counted the old way, say a 3080 would be 21.6Tflops, and this is how we'd get back to the traditional situation where for a given performance Nvidia has less flops on paper than AMD. Nvidia reversed that rule by way of Ampere. So don't get too quick to compare them basically.


Also ignore any CGI artists that tell you flops are the only relevant metric in all of graphics pipelines.

Ampere TFLOPS are not "weird" and they are not counted in a different way now than any other TFLOPS were or are. Ampere TFLOPS are calculated in the EXACT same way that RDNA2 TFLOPS or Xbox TFLOPS or Playstation TFLOPS are calculated. There is no "counting the old way" because there is no "new way".

The problem you're having is trying to compare different architectures and believing the TFLOP count should tell you how powerful they are relative to each other. That ONLY works when the 2 GPU's you're comparing are the same or very similar architectures.
 
Last edited:

LordOfChaos

Member
Ampere TFLOPS are not "weird" and they are not counted in a different way now than any other TFLOPS were or are. Ampere TFLOPS are calculated in the EXACT same way that RDNA2 TFLOPS or Xbox TFLOPS or Playstation TFLOPS are calculated.

The problem you're having is trying to compare different architectures and believing the TFLOP count should tell you how powerful they are relative to each other. That ONLY works when the 2 GPU's you're comparing are the same or very similar architectures.

I understand they're not calculated another way. But the mix of INT and FP in the pipeline compared to how current games operate blows up the Tflop numbers compared to Turing, it's just an easier way to describe it like one might say RDNA 1 is 1.5x the performance per flop of GCN. The method of calculating the number doesn't change.

I should have said "compared to average Turning performance" rather than "counted the old way", in fairness
 
Last edited:

skneogaf

Member
If these new cards can clock this high then why is the series x clocked so low if its the same architecture?
 
D

Deleted member 17706

Unconfirmed Member
OCTOBER 28TH I believe.

Expect launch of these cards to be end of November into first week of December. RYzen 5000 launches Nov.5

Awesome. Was going to get an RTX 3080, but it's impossible to order one, so I'm looking forward to seeing what AMD can put out. Hopefully we'll learn about their answer to ray tracing and DLSS.
 

Papacheeks

Banned
If these new cards can clock this high then why is the series x clocked so low if its the same architecture?

Think about it. The cards are around 255w and thats keeping them at stock. Imagine overclocking or having a AIB that has a slight overclock. That will take the power draw up. Consoles are designed around a certain power range. Xbox series x currently while playing current gen games is around an impressive 165 watts. Thats insane when you calculate thats at the wall so that includes what the cpu and gpu is drawing. And other things like memory, and the ssd.

Super impressive thermal, and power design.

And that would be because of 7nm TSMC process.

Why do you think nividia is switching next year? Samsung cant keep up.
 

Kumomeme

Member
hopefully there no temperature issue. when they finally release new high end capable gpu that can toe to toe with nvidia, there usually some of caveats like high tdp usage or temperature that would hold back the card potential and buyer interest.
 
Last edited:

CrustyBritches

Gold Member
giphy.gif
 
If these new cards can clock this high then why is the series x clocked so low if its the same architecture?

We don't know the exact answer to that yet but there seems to be two reasons.

1. A lower CU count (PS5 GPU) makes higher clock speeds easier to achieve. At least RDNA 2 leaks suggest that anyway.

2. The PS5s GPU clock only "boosts up to" 2.2GHz and doesn't seem to be a clock speed meant to be maintained. Where as the higher CU XSX GPU is able to maintain that 1.8GHZ indefinitely.

The PS5 acts like a laptop where everything happens within a strictly controlled power envelope. For example, if you're really pushing the GPU then the CPU has to be downclocked or if you push the CPU then the GPU can't reach it's max clocks anymore. Both at the same time can't happen because that would exceed its power envelope. The XSX behaves more like a PC where both the CPU and GPU can be pushed to their rated specs and maintain those numbers simultaneously and indefinitely.

I hope someone (maybe Digital Foundry) can figure out a way to monitor the clocks on both consoles (kind of like MSI afterburner) as it would be interesting to see how each approach compares. This would be especially true in CPU heavy games.
 
Last edited:

sendit

Member
you can't compare APU clocks with graphics card clocks.

these chips are housing a CPU cluster and a GPU cluster.
the cooling solution has to keep them both under control.

in a PC the CPU and the GPU are far apart from eachother so one doesn't influence the other nearly as much when it comes to heat. and both can have their own dedicated heat sink and even fan

Thanks for explaining the obvious (which I already pointed out). Regardless of the solution, both are regulated and limited to power draw and thermal output. Either solution can achieve the same goal (not talking about how effective they’re in achieving said goal).

And once again, they share the same underlying GPU architect regardless of the solution.
 

waylo

Banned
If they have something similar to DLSS I'd consider jumping ship. However, I don't see that being the case and instead I'm going to be stuck waiting out a 3080.
 
Nvidia has more brute force TFLOPS while AMD relies more on technical with RDNA2.

Not sure why you think so. As things stand right now it's definitely Nvidia GPUs that have the more impressive technology under the hood (both hardware and software)

Nvidia GPUs now have both dedicated hardware (Raytracing cores and Tensor Cores). I guess we'll have to wait for the 28th to learn more about RDNA 2.
 

Marlenus

Member
If these new cards can clock this high then why is the series x clocked so low if its the same architecture?

To keep power consumption and thermals in an acceptable envelope for a console sized device.

It also helps yields as more chips will hit the required clockspeed at acceptable thermal and power levels than a higher clockspeed would.
 

Marlenus

Member
255W TGP not TDP, for the TDP you need to add watts for the rest of the board and not just the GPU. So around 300W TDP for the AIB cards. Reference might be around 280W.

No. TDP is the wattage of the chip. TGP/TBP is wattage of the entire card. NV use TGP and AMD use TBP.

The 5700XT has a TDP of 180W and a TBP/TGP of 225W.

Edit to add: The TDP is the minimum spec for the cooler as memory and VRMs etc can be passively cooled.
 
Last edited:
Top Bottom