NVidia Ada Speculation, Rumours and Discussion

Status
Not open for further replies.
Do we even know what they have improved with these "new" cores ?
The Ada RT Core has been enhanced to deliver 2x faster ray-triangle
intersection testing and includes two important new hardware units. An Opacity Micromap
Engine speeds up ray tracing of alpha-tested geometry by a factor of 2x, and a Displaced
Micro-Mesh Engine generates Displaced Micro-Triangles on-the-fly to create additional
geometry. The Micro-Mesh Engine provides the benefit of increased geometric complexity
without the traditional performance and storage costs of complex geometries.
● Shader Execution Reordering: NVIDIA Ada GPUs support Shader Execution Reordering
which dynamically organizes and reorders shading workloads to improve RT shading

ada white paper
 
The Ada RT Core has been enhanced to deliver 2x faster ray-triangle
intersection testing and includes two important new hardware units. An Opacity Micromap
Engine speeds up ray tracing of alpha-tested geometry by a factor of 2x, and a Displaced
Micro-Mesh Engine generates Displaced Micro-Triangles on-the-fly to create additional
geometry. The Micro-Mesh Engine provides the benefit of increased geometric complexity
without the traditional performance and storage costs of complex geometries.
● Shader Execution Reordering: NVIDIA Ada GPUs support Shader Execution Reordering
which dynamically organizes and reorders shading workloads to improve RT shading

ada white paper
Shader Execution Reordering need to be exploited by the devs, no ?
 
Port Royal is from beginning of 2019. Was one of the first Raytraycing application. I hope that UL will release the new test in a few weeks.
 
Shader Execution Reordering need to be exploited by the devs, no ?
Yes, Micro-Mesh engine as well as Shader Execution Reordering must be implementedby the developer via nvidia's proprietary api

But the Opacity Micromap Engine seems to work without any api
 
The RT performance in Port Royal doesn't look particularly impressive though. I was expecting to see the largest difference there given the fact that these are 3rd generation RT cores.
Port Royal has single bounce reflections and shadows, not exactly heavy ray tracing wise. The heavier test is 3DMark Ray Tracing feature test.
 
But the Opacity Micromap Engine seems to work without any api
Afaict, opacity is implemented on top of DMM: The triangle is teasselated go a grid mesh, and each face of the mesh is either fully opaque or transparent, or partial. Only the partial faces then need a texture lookup.

So i guess some developer work is needed to enable the propriatary feature, which is no obstacle for new games ofc. But not sure.
 
I don’t see why anyone would go for 4080 12/16 gig over the 4090. They are already priced out of most people’s budget but the performance difference between 4080 and 4090 might the biggest one in recent years. Last gen the 3090 was niche and definitely not worth it over the 3080, I think this gen is the reverse.

It’s just hard to get excited about 4080 given the price. It’s actually worse perf/$ than the 3080 was at MSRP (obviously you couldn’t find one at MSRP so maybe that’s irrelevant).
 
The 4090 at 40% fan speed and 450w power consumption operates at 65 degrees temperature, and 2730MHz clocks.

 
The 4090 at 40% fan speed and 450w power consumption operates at 65 degrees temperature, and 2730MHz clocks.


Over engineerd and large cooling solution, but very quiet and cool even at those clocks and performances. Predicted that before in the topic.
 
I have to wonder what will Nvidia ask for the full AD106, when the full AD104 is already $899.
AD106 is supposedly ~190mm2 while AD104 is 294.5mm2.

For comparison:
AmpereDie Size [mm2]SMCudaTMUROPCache [MB]Memory bus [bit]Vram [GB]
GA107?202560803221284
GA1062763038401204831928
GA10439248614419296425612

I think Ada based full chips could look like this.
Ada LovelaceDie Size [mm2]SMCudaTMUROPCache [MB]Memory bus [bit]Vram [GB]
AD104294.5607680240804819212
AD106~19040512016040-48 ?321288
AD107?243072963216-24 ?96-128 ?6-8 ?

The full AD106 should be RTX 4060Ti or maybe RTX 4070?
I expect 3070-3070Ti level of performance with ~200W TDP, and their official MSRP was $499-599. I think, Nvidia will price It at $499, maybe $449 If we are lucky, in other words $50-100 more than RTX 3060Ti. The problem will be the big gap in price between full AD106 and AD104. In my opinion, only one cut down SKU won't be enough to fill the gap.

The fight AD106 vs N33 will be very interesting.
 
Last edited:
I expect 3070-3070Ti level of performance with ~200-225W TDP
Doesn't make much sense to me considering that 3070 is already a 220W card with 5888 SPs.
If AD106 is 5120 SPs and is clocked above 3GHz (which seems possible) then we're looking at about +40% of performance to 3070 from clocks alone.
If they won't push it this high on frequency then why would it be 200-225W?
 
Doesn't make much sense to me considering that 3070 is already a 220W card with 5888 SPs.
If AD106 is 5120 SPs and is clocked above 3GHz (which seems possible) then we're looking at about +40% of performance to 3070 from clocks alone.
If they won't push it this high on frequency then why would it be 200-225W?
I changed It to 200W in my first post, because 225W looked a bit much, even before reading your post.
My performance prediction was based on the leak from kopite7kimi about <7000 in TSE, but It's too low for 5120SP.

AD106: 5120SP * 2 FLOPs * 2610 MHz = 26,726 GFLOPs
GA104: 6144SP * 2 * FLOPs * 1770 MHz = 21,750 GFLOPs
Difference in FLOPs is already 23% without any architectural improvement or higher clocks than AD102-104.

For this level of performance, I don't think 200W is out of the question, but 175-185W is also possible.
In comparison, the full GA104+GDDR6x -> RTX 3070 Ti has 290W.
RTX 4080 12GB has 285W.

If the performance is only at the level of 3070Ti then ~4096SP or 32SM is more likely and of course <<200W TDP. So maybe he was talking about a cut down AD106.
 
Last edited:
I mean it's half of the price...

I think the point is that even the 4080 12GB is in the ultra-enthusiast pricing tier (it'll likely cost >£1000 in the UK), so if you're already committed to that kind of outlay then you obviously care more for raw performance that you do for value. At that point the 4090 may seem more attractive at £1800-£1900. In terms of performance the 3080Ti is likely to offer much better value that the 4080 12GB over here as you can pick those up now for <£800 and I expect it to be pretty similar in general performance where Ada's unique features aren't used.

Personally, I'm not really convinced of the value proposition of even the 4090 when you consider in the UK you can pick up a 3090 (the closest equivalent chip from the Ampere generation) for £1100 vs the 4090 which is likely to be >£1800. You're essentially paying 60-70% more for 60-70% more performance. So the value proposition there is equal to the 3090 which wasn't considered great value when it launched, but we're two years on from that now. The additional features do add to that somewhat of course but we haven't even considered how the 3080Ti impacts the value proposition there given it's only around 40-45% the price of the 4090.

I'm starting to think about getting a 3080Ti myself on the Nvidia side. That is unless AMD come out with something that offers sufficiently superior value to make up for the lack of DLSS (it'd have to be a LOT!)
 
I think the point is that even the 4080 12GB is in the ultra-enthusiast pricing tier (it'll likely cost >£1000 in the UK), so if you're already committed to that kind of outlay then you obviously care more for raw performance that you do for value.
If that would be the case then everyone would only be buying 3060s and 3090/Tis. Which is obviously not the case. The difference between $900 and $1600 is almost two times while the performance difference will likely be smaller than that meaning that 4080/12 will still have higher value for those who count money. It is also very likely to not loose as much of value with the launch of "Super" or 50 series down the road as any of the top most product.

Also, yeah, did I say that it's like half the price already?

Sometimes I wonder if people genuinely can't understand that there are products below the fastest one and you don't really have to buy the fastest one to get the new features and performance which would be enough for you. Why would I buy a 4090 if I plan to game on a 1080p TV for the foreseeable future? Rhetorical question.
 
Personally, I'm not really convinced of the value proposition of even the 4090 when you consider in the UK you can pick up a 3090 (the closest equivalent chip from the Ampere generation) for £1100 vs the 4090 which is likely to be >£1800.
Come on, be fair, you can't objectively compare a brand new product that is being launched in the market to a last gen end-of-life SKU that is heavily discounted in order to clear the stock...
 
Status
Not open for further replies.
Back
Top