• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD Navi 23 ‘NVIDIA Killer’ GPU Rumored to Support Hardware Ray Tracing, Coming Next Year

Status
Not open for further replies.
I'll just put this here (from a post I wrote a couple days ago):

Based on this:
"With RDNA 2 we get basically a 25 percent performance uplift over GCN with no work by developers at all"
Link:
https://www.theverge.com/2020/9/17/...xt-gen-gaming-performance-gpu-memory-hardware

Based on (some) evidence:
I think it's safe to assume that a 60 CU RDNA 2 GPU would get a +25% performance improvement, vs GCN GPUs, like the Radeon VII.

If clocked 23.9% higher than the Radeon VII at 2233mhz (PS5 GPU frequency), we would probably get another 12% (at least) of performance (5700 XT performance gains from higher frequencies are about equal to the percentage overclocked, divided by 2. This is based on 3d Mark:Time Spy GPU scores).

So, a minimum of 37% faster than the (GCN based) Radeon VII if RDNA 2 is clocked at 2233mhz, with 60 CUs.

That would put it ahead of the RTX 2080 TI.

Speculation and observations:
An 80 CU RDNA 2 GPU could be upto 33.3% faster than a 60 CU RDNA 2 GPU (similarly, a 90 CU GPU could be upto 50% faster).

So, a maximum of 70.3% increased performance for an 80 CU RDNA 2 GPU, vs the Radeon VII.

Comparing the 5700 to the 5700 XT, the performance is about 12.7% higher in 3D Mark (GPU score), the clocks are about 10% higher and the CU amount is 11% higher. The point is, I think AMD would need to combine the extra 20 CUs with a ~33.3% higher GPU clocks (2233mhz + 33.3%) to see a proportionate increase in performance.

One of the main benefits of RDNA 2 seems to be lower energy consumption. The RX 5700 XT used 45w more power, just for higher clocks and 4 extra CUs vs the RX 5700. The new consoles have significantly lower TDP per CU.

What do you guys think, is this too optimistic, or not optimistic enough?
 
Zeed said:
STOCK is what the problem will be after all ****** off people with preorders taht sold 2080tis cant get cards will go to buy AMD card instead and sell out in first Minute.
Click to expand...

No no no no no!!!!!

These guys that wanted AMD to compete so they can get cheaper nvidia cards wasnt it???

Stick to the queue and let us people that appreciate AMD for what they are get them first, in fact we should be telling them the drivers are awful - dont bother with them!! :D

Sargatanas2511 said:
Sleeves are for people with things to hide :p.
Click to expand...

LOL. Jensen crook.
 
g67575 said:
What do you guys think, is this too optimistic, or not optimistic enough?
Click to expand...
I think your basis for comparison is wrong, if I'm honest. I don't see how you can take an interview talking about custom GCN vs custom RDNA 2 in consoles and then jump into using Radeon VII as your reference point, especially as it's just a MI50 compute card underneath. And from there all of your numbers just seem to be off, arguably overthinking it too.

GCN just doesn't matter any more, it's gone. Why aren't you using RDNA 1 as your reference point? That is a known quantity and a more accurate base for comparison. humbug's already done a lot of number crunching using RDNA 1 as the reference point, I've already done some best-case numbers using AMD's 50% performance per watt uplift over RDNA 1 too. So I don't understand why Radeon VII and GCN even comes into it.
 
g67575 said:
I'll just put this here (from a post I wrote a couple days ago):

What do you guys think, is this too optimistic, or not optimistic enough?
Click to expand...

I think you overthink stuff buddy. Keep it simple and the credit card ready when Lisa has them available to order £££.
 
Sargatanas2511 said:
If Nvidia lose enough sales he might have to downgrade to a denim jacket.
Click to expand...

PMSL. Love it.

Would go well with this underneath:

ssrco,classic_tee,womens,fafafa:ca443f4786,front_alt,square_product,600x600.jpg
 
g67575 said:
What do you guys think, is this too optimistic, or not optimistic enough?
Click to expand...

its all in the realms of possibility as far as theoretical performance if the above math checks out.

BUUUUUUUUUUUUUUT

factor in how far amd want to push the boundaries with speed verses power efficiency, something amd keep banging on about and with ampere they could really go all out with claiming power savings in the 100's of watts. yes pc users dont generally car i bet we get a marketing campaign based around performance per watt and the idea amd's better as you dont need a new psu. and yes due to ampere being a hungry gpu it does leave amd that extra room to push their power levels up a bit more.

so who knows. amd hopefully will push the limits of the gpus or its going to be another mid tier refresh once again rather than a balls to the wall flagship gpu pushed as far as they can. but dont be surprised if they arnt clocked as high as you'd hope just to keep that power level down a touch.
 
5700XT: 100%
Radeon VII: 102%

5700XT: 2560 Shaders @ 1900Mhz / 448GB/s Memory Bandwidth
Radeon VII 3840 Shaders @ 1800Mhz / 820GB/s Memory Bandwidth

5% lower clock speed, 50% more shaders (+45%) and over 80% higher Memory Bandwidth..... its hard to know what that Memory Bandwidth difference would make but i think i'm safe in saying that RDNA1 has 60% higher IPC vs GCN.

GCN is junk compared with RDNA, its not even funny how they compare.

Now add 50% Shaders to the 5700XT, add 10% IPC to it, add 20% clock speeds to it (+80% Total) take off 20% for scaling loss. so +60% actual performance gain, where are you now? You're at about 170 FPS on this slide.

Now add 100% Shaders, add 10% IPC, add 20% clocks, take off 30% for scaling loss, 100% actual performance gain, now where? 210 FPS.

eK4Y1E6.png
 
Well we wont get carried away, but I am excited to believe that if this is a genuine challenge to the nvidia 3000 series, we should have good cards at good prices and it drills down to how available are they when November hits.
 
LePhuronn said:
GCN just doesn't matter any more, it's gone. Why aren't you using RDNA 1 as your reference point?
Click to expand...

You are right to point that out. Unfortunately, we don't know how well RDNA gen 1 compares with RDNA gen 2. No one from AMD, Microsoft, Sony or any game developers (that I'm aware of) have given us any idea of the performance benefit.

We only have limited information about the performance benefit from GCN to RDNA 2.

Therefore, comparing RDNA 2 to a Radeon VII is still a valid comparison.
 
We have been using 5700XT (rdna1) for our fag packet calculations for a while now. Even though some of us dont understand that, or think thats a valid metric.
 
g67575 said:
You are right to point that out. Unfortunately, we don't know how well RDNA gen 1 compares with RDNA gen 2. No one from AMD, Microsoft, Sony or any game developers (that I'm aware of) have given us any idea of the performance benefit.

We only have limited information about the performance benefit from GCN to RDNA 2.

Therefore, comparing RDNA 2 to a Radeon VII is still a valid comparison.
Click to expand...

RDNA1 is vastly better than GCN, its not going to be anything like that uplift, AMD said 10%+ higher IPC, we know RDNA2 clocks about 350Mhz higher that RDNA1, about ~20%, so take that and add the 10 IPC (+30% Total) and then you have a formula for an RDNA1 vs RDNA2 GPU comparison, a 2560 Shader RDNA2 GPU is 30% more GPU than a 2560 Shader RDNA1 GPU. Beyond that you add the shaders in whatever the RUMOURS are and take some performance off for scaling, with that you can get a good idea what a 2560 Shader RDNA2 GPU would look like, or a 3840 Shader RDNA2 or a 5120 Shader RDNA2 GPU.
 
Status
Not open for further replies.
Back
Top Bottom