Jump to content
Search In
  • More options...
Find results that contain...
Find results in...

Cost of RT and Tensor Cores

AlfaProto
 Share

Just going to share the cost of the RT and Tensor Cores on the GeForce 20 series, removing the traditional cores, ROPs, TMUs. Do note that the price are based on MSRP, and not the FE pricing, unless otherwise stated with 'FE'.

 

PRICE-TO-PRICE

 

RTX 2080 Ti ($999) - Titan X Pascal/p (FE $1200) - Cost of RT/T Cores: $201

RTX 2080 ($699) - 1080 Ti ($699) - Cost of RT/T Cores: $0

RTX 2070 ($499) - 1080 ($549) - Cost of RT/T Cores: $51

RTX 2060 (FE $349) - 1070 Ti ($449) - Cost of RT/T Cores: $100

RTX 2060 (FE $349) - 1070 ($379) - Cost of RT/T Cores: $30

 

If we removed the RT/T Cores, and adjusts its pricing we have this:

 

RTX 2080 Ti - $799

RTX 2080 - $699

RTX 2070 - $449

RTX 2060 - FE $319

 

Now, I really can't get all the benchmarks in the world, but it's up to you to do the math, perf-per-dollar, and decide whether you are paying more or less for the new GeForce 20 series without the RT/T Cores. Generally, I see 10-15% improvement when compared the 20 series against the 10 series, except the RTX 2080 Ti, which can run at 4K90 in certain titles.

 

Let's move to the tier-to-tier comparison.

 

Titan RTX (FE $2499) - Titan X Pascal/p (FE $1200) - Cost of RT/T Cores: $1299

RTX 2080 Ti ($999) - 1080 Ti ($699) - Cost of RT/T Cores: $300

RTX 2080 ($699) - 1080 ($549) - Cost of RT/T Cores: $150

RTX 2070 ($499) - 1070 Ti ($449) - Cost of RT/T Cores: $50

RTX 2070 ($499) - 1070 ($379) - Cost of RT/T Cores: $120

RTX 2060 (FE $349) - 1060 6GB ($249) - Cost of RT/T Cores: $100

RTX 2060 (FE $349) - 1060 3GB ($199) - Cost of RT/T Cores: $150

 

There's nothing to remove on the RTX series cards, at all, but this is the cost if RT/T cores are added to their respective previous generation cards. I haven't seen, or found a reputable BF V benchmark with ray tracing capabilities tested on the Pascal line-up.

 

Even if we go with the Ray Tracing performance, there's no RT benchmark for the Pascal line-up at all, the tech doesn't exist for those line-up. Plus, we can't go simply with RT performance, because, it doesn't account for other factors like deep learning AI stuff, use case scenario: DLSS, the tensor cores are used, but not the RT cores. Maybe the Volta architecture could, since it does have tensor cores, but not RT cores.

 

In conclusion, this discussion is just to show how much you actually pay for the RTX cards, without its RT/T cores, and its performance loss/gains. There's limited to no data on BF V ray tracing benchmarks on Pascal line-up, but if you find the data I've shown useful if you performed a benchmark of BF V ray tracing on a Pascal line-up, and do the perf-per-dollar.

Link to comment
Share on other sites

Link to post
Share on other sites

This comparison to me doesn't make any sense. For one the launch MSRP of the GTX 1080 was $650. And as time goes on, this comparison makes less and less sense as the price changes.

 

It's much better if you want to compare the cost of RT cores is by figuring out how much it cost to manufacture the GPU itself, which you could probably use something like transistors per dollar. However, that's assuming everything else used the same amount of transistors to make per unit between Pascal and Turing and the only thing that was added was the RT and tensor cores. I don't think this assumption holds very well especially since that Turing has more L2 cache per SM than Pascal, possibly among other things.

Link to comment
Share on other sites

Link to post
Share on other sites

2 hours ago, Mira Yurizaki said:

It's much better if you want to compare the cost of RT cores is by figuring out how much it cost to manufacture the GPU itself, which you could probably use something like transistors per dollar. However, that's assuming everything else used the same amount of transistors to make per unit between Pascal and Turing and the only thing that was added was the RT and tensor cores. I don't think this assumption holds very well especially since that Turing has more L2 cache per SM than Pascal, possibly among other things.

Not to mention R&D cost needs to be considered (and technically marketing.... but I don't think it had one based on the presentation given)

Link to comment
Share on other sites

Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
 Share


×