Speculation: GPU Performance Comparisons of 2020 *Spawn*

[pjbliverpool]

I read today that a 3080 is going to be 50% faster then a 2080, 20% faster then a 2080Ti. Sounds too good to be true?

Fingers crossed for that as it should mean a 3070 (which is likely what I'll be getting) is roughly as fast as 2080Ti and hopefully even faster in RT and Tensor/DLSS performance.

Given I'll be getting a WQHD monitor to go with it which is only about 60% the resolution of 4K, and given that DLSS in quality mode at that resolution uses an internal resolution that's roughly equivalent to 1080p in terms of pixels then in DLSS enabled games I should basically have 2080Ti level performance to throw at 1080p resolution while still maintaining image quality at native or above 3440x1440. Ultra high frame rates here I come!

 

[Bondrewd]

Fingers crossed for that as it should mean a 3070 (which is likely what I'll be getting) is roughly as fast as 2080Ti and hopefully even faster in RT and Tensor/DLSS performance.

It is.

 

[techuse]

I read today that a 3080 is going to be 50% faster then a 2080, 20% faster then a 2080Ti. Sounds too good to be true?

50% has always seemed like a reasonable upper bound on general performance increase tier on tier IMO.

 

[PSman1700]

50% has always seemed like a reasonable upper bound on general performance increase tier on tier IMO.

Yes, but still, it seems alot. That would mean all things considered that a 3070 would be as fast as a 2080Ti in raw raster performance. Which is awesome. Then we also get more advanced features, faster ray tracing, upgraded tensor hardware. I think NV is really going all in, they know they got abit more competition from AMD.

 

[techuse]

Yes, but still, it seems alot. That would mean all things considered that a 3070 would be as fast as a 2080Ti in raw raster performance. Which is awesome. Then we also get more advanced features, faster ray tracing, upgraded tensor hardware. I think NV is really going all in, they know they got abit more competition from AMD.

It could and may fall lower of course.

 

[Bondrewd]

Yes, but still, it seems alot

Nah.
Prep your popcorn for this and the next year.

they know they got abit more competition from AMD

Abit is uh, a very harsh understatement to put it rather bluntly.

It could and may fall lower of course.

Nah.
But you pay the power tax.

 

[PSman1700]

Abit is uh, a very harsh understatement to put it rather bluntly.

Well, we don't know anything about RDNA2 yet, how that stacks up to Ampere. It seems that RDNA2 will be more of a match to Turing or an in between. This is also bad, as prices wont come down from NV still.

 

[Bondrewd]

Well, we don't know anything about RDNA2 yet

Yes we do.

how that stacks up to Ampere

Well enough for beancounters at AMD to not be bored.

It seems that RDNA2 will be more of a match to Turing or an in between

Oh noes not even close.

as prices wont come down from NV still.

See, bad mentality.
You will be reeducated.

 

[PSman1700]

See, bad mentality.
You will be reeducated.

Calm down lol, i want AMD to be competitive again in the gaming segment. As a PC gamer at heart, as things are now, it's way to expensive with the GPU pricing going on. AMD has promised alot now, but still not really making a comeback.
If RDNA2 can compete with Ampere, prices will come down. Give me that 18+TF RDNA2 full gpu, being competitive with whatever NV is going to come with, on all levels including RT.

 

[Bondrewd]

Oh shite, Intel announced 6-12 months delay on their 7nm.
Can I finally bury them?

i want AMD to be competitive again in the gaming segment

Then expect the best of them; for they only desire to deliver the best.

Give me that 18+TF RDNA2 full gpu, being competitive with whatever NV is going to come with, on all levels including RT.

Wish granted, and then some.

 

[trinibwoy]

Wish granted, and then some.

I don't know about you guys but I'm officially HYPED!

 

[Bondrewd]

I don't know about you guys but I'm officially HYPED!

Good, prep popcorn for the next 2-3 years where we'll see intense GPU titanomachy between two vendors.
And Intel dying, but that's par for the course.

 

[P_EQUALS_NP]

Oh shite, Intel announced 6-12 months delay on their 7nm.
Can I finally bury them?


Then expect the best of them; for they only desire to deliver the best.

Wish granted, and then some.

i wonder what this means for ponte vecchio and their hpc contracts, will aurora go cuda instead?

 

[P_EQUALS_NP]

I work as a system programmer for a small hpc sever system, when we upgraded our system to use accelerators early in the decade we went with cuda because opencl at the time required a lot of proprietary extensions in order to vector-ize anything defeating the purpose of being open source in the first place. so to be fair i haven't looked at opencl in years has it gotten any better?

 

[Bondrewd]

i wonder what this means for ponte vecchio and their hpc contracts, will aurora go cuda instead?

No lol, it's getting delayed.

has it gotten any better?

No, use HIP/SYCL.
OCL is going places you'd rather not see, courtesy of Nvidia.

 

[CarstenS]

OCL is going places you'd rather not see, courtesy of Nvidia.

How so? Isn't it an open industry standard? Or do you refer to Nvidia not giving it the same amout of love they do with Cuda?

 

[Bondrewd]

Isn't it an open industry standard?

That's more of a poison pill than a panacea.

Or do you refer to Nvidia not giving it the same amout of love they do with Cuda?

Yeah a third class citizen even on *the* GPGPU vendor along with Apple moving to Metal doomed it.
Also OCL2.x kinda went absolutely nowhere, especially with nV refusing to implement the important bits.
But hey, SYCL lives!
Even more than that, just look at Codeplay or what Intel does with it or hipSYCL and all.

 

[w0lfram]

Navi2 has more "IPC" and also faster clocks, than Navi1. It is also doesn't have a hybrid design and thus frees up die space.

Given the same amount of transistors, How much more faster is navi2, over navi1..?
I'd say... 30%+ faster.

 

[Deleted member 13524]

Navi2 has more "IPC" and also faster clocks, than Navi1. It is also doesn't have a hybrid design and thus frees up die space.

Given the same amount of transistors, How much more faster is navi2, over navi1..?
I'd say... 30%+ faster.

What do you mean with hybrid design?

 

[iroboto]

I can only suspect he is referring to this:
https://www.sweclockers.com/nyhet/2...med-inslag-av-gcn-renodlad-rdna-forst-ar-2020

quick stolen translated summary from wccftech:
However, while Navi sounds like a completely new chip design from the ground up, Sweclockers reports that it still has many things that it carries over from GCN. According to them, GCN GPU architecture has been here for almost a decade and AMD has made lots of optimizations for the gaming side for GCN which if they wanted to build a new GPU would just wash away. So there are still some design aspects being used by Navi GPUs that make it a hybrid of RDNA and GCN but for those looking for a purely new GPU, the next iteration of Navi which is rumored as Navi 20 will be the first completely RDNA design with AMD shifting all of their optimizations for gaming to the new enthusiast-grade chips.