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

[cainer]

Don't know about dollars but I paid around £300 for one

iirc i paid just over £400 for a Hercules x800xt Platinum at launch, i googled launch price and found this..

"As far as price goes, ATI has announced an SRP of $399 for the Pro and $499 for the XT part.
Going out on a limb, I'd imagine that this will equate to British pricing of £299 and £399 respectively, although this is pure speculation."

 

[HeX]

Don't know about dollars but I paid around £300 for one

Dollar rate was a lot higher back then, I remember getting almost $2:£1, and VAT was only 17.5%!

iirc i paid just over £400 for a Hercules x800xt Platinum at launch, i googled launch price and found this..

"As far as price goes, ATI has announced an SRP of $399 for the Pro and $499 for the XT part.
Going out on a limb, I'd imagine that this will equate to British pricing of £299 and £399 respectively, although this is pure speculation."

Yeah, think I paid about £319 for a Club3D X800 Pro on launch.

 

[GAC]

Dollar rate was a lot higher back then, I remember getting almost $2:£1,

yeah the exchange rate was a lot better back then, but also we didnt have the questionable use of it as a stick to beat up the prices. just look at the last few big releases its been a total farce.

 

[Dontrocktheboat]

yeah the exchange rate was a lot better back then, but also we didnt have the questionable use of it as a stick to beat up the prices. just look at the last few big releases its been a total farce.

You can't possibly be refering to OCUK and gibbo's price gauging on 3080 series on launch... Follower by the lamest excuse of exchange rate fluctuations

 

[GAC]

You can't possibly be refering to OCUK and gibbo's price gauging on 3080 series on launch... Follower by the lamest excuse of exchange rate fluctuations

no no no no nooooooooo never that

 

[Nablaoperator]

Yeah thats in line with what I thought. At best it will have around two thirds of the RT performance but thats still better than the 2080ti. and if it delivers on none RT grunt and priced right then I will buy.

I can move over to a future card once both sides have RT cards which can deliver 60+ fps on 4k with all the graphics settings turned up to max (the useful ones of course not silly ones like the grass which crippled every framerate)

There are some implicit assumptions I have made to stop short of building a full scale model

• All rays hit the scene
• BVH hits mutually exclusive, which means search is aborted at first hit.. there are no diverging queries

The first assumption makes sense cuz Nvidia would have built an appropriate scene to churn out these theoretical numbers.

However, the second assumption is an approximation but untrue in theory.. right now with the denoiser they can probably afford to do this without discernible loss of visual quality.

Also we don't know if diverging queries are being treated as new rays (should be if you are aiming a scalable arch).. in which case the difference can be compressed further..

So I would revise my estimates to round about 38% slower instead of atleast 38% slower than RTX 3080 in RT.

 

[Rroff]

There are some implicit assumptions I have made to stop short of building a full scale model

• All rays hit the scene
• BVH hits mutually exclusive, which means search is aborted at first hit.. there are no diverging queries

The first assumption makes sense cuz Nvidia would have built an appropriate scene to churn out these theoretical numbers.

However, the second assumption is an approximation but untrue in theory.. right now with the denoiser they can probably afford to do this without discernible loss of visual quality.

Also we don't know if diverging queries are being treated as new rays (should be if you are aiming a scalable arch).. in which case the difference can be compressed further..

So I would revise my estimates to round about 38% slower instead of atleast 38% slower than RTX 3080 in RT.

Problem is there are just too many unknowns right now - there will be implications for moving results around via VRAM and caches, we don't know how efficient shading of results is when that has to look up other data, etc. and we don't know how effective AMD's attempts to work around or remove the reasons why utilising general compute for ray tracing is ultimately very slow as can be seen with Pascal (even when you add specialised functionality for certain parts of it).

 

[Chrisc]

I'm beginning to believe we got bots posting in this thread.

Its Lisa Su just trolling us

 

[Chrisc]

I just found out that the RTX 3080 Founders edition GPUs brought in the US has an RRP of $699.99 dollars, which is tax free if brought from Nvidia's online store. That's equivalent to £537. Easy to see why they are selling out sold out so quickly.

Question - If I went on Holiday to the US (if there's a Covid-19 Vaccine, was planning a trip anyway sometime next year), am I allowed to take items like graphics cards back to the UK, brought while I was there?

EDIT - the US customs rule is apparently "You are allowed to take home $800 worth of items per person duty-free in your luggage, once every 30 days"
Link here: https://www.ricksteves.com/travel-tips/packing-light/clearing-customs

Brought or bought?

 

[bigfatron]

I just found out that the RTX 3080 Founders edition GPUs brought in the US has an RRP of $699.99 dollars, which is tax free if brought from Nvidia's online store. That's equivalent to £537. Easy to see why they are selling out sold out so quickly.

Question - If I went on Holiday to the US (if there's a Covid-19 Vaccine, was planning a trip anyway sometime next year), am I allowed to take items like graphics cards back to the UK, brought while I was there?

EDIT - the US customs rule is apparently "You are allowed to take home $800 worth of items per person duty-free in your luggage, once every 30 days"
Link here: https://www.ricksteves.com/travel-tips/packing-light/clearing-customs

Thats the limit taking stuff into the US you're quoting, not the limit for bringing stuff into the UK from the US (which Google says is £390 currently).

 

[Nablaoperator]

Problem is there are just too many unknowns right now - there will be implications for moving results around via VRAM and caches, we don't know how efficient shading of results is when that has to look up other data, etc. and we don't know how effective AMD's attempts to work around or remove the reasons why utilising general compute for ray tracing is ultimately very slow as can be seen with Pascal (even when you add specialised functionality for certain parts of it).

True.. but can only do synthetics with available info.. just trying tog get a speculative range so that folks stay grounded when RT benchmarks are published.. meanwhile i have also changed my mind on the estimates once again

 

[Nablaoperator]

Here are the scenarios:

RTX 3080
450 giga ray-box intersections per sec
225 giga ray-tri intersections per sec

Top Navi
280 giga ray-box intersections per sec
70 giga ray-tri intersections per sec
38% slower RT at least

I have changed my mind again.. but this is the best i have now.
..believe there are 2 scenarios:

• Mutually exclusive BV hits.. search aborted on first hit: Big Navi 38% slower at RT
• Exhaustive BV hits.. all BVs searched for hit, but nvidia used a scene/camera with 0 divergence. Big Navi 65% slower. For 4 wide BVH, Big Navi 30% slower

Overall, Big Navi (80 CU) 30-65% slower than RTX 3080 at RT, depending on nvidia's implementation.
Midpoint 53% slower at RT.
Eager to see some benchies leaking

 

[humbug]

I have changed my mind again.. but this is the best i have now.
..believe there are 2 scenarios:

• Mutually exclusive BV hits.. search aborted on first hit: Big Navi 38% slower at RT
• Exhaustive BV hits.. all BVs searched for hit, but nvidia used a scene/camera with 0 divergence. Big Navi 65% slower

Overall, Big Navi (80 CU) 38-65% slower than RTX 3080 at RT, depending on nvidia's implementation.
Midpoint 55% slower at RT.
Eager to see some benchies leaking

I'm looking forward to those reviews too but i'm going to stick my neck on the line and say that Nvidia's RT performance will be better than AMD's.

And this is how Nvidia will justify a higher price, that's fine, its all good and proper but for someone like me for who RT is completely irrelevant..... i win. Check my signature.

 

[Troezar]

I have changed my mind again.. but this is the best i have now.
..believe there are 2 scenarios:

• Mutually exclusive BV hits.. search aborted on first hit: Big Navi 38% slower at RT
• Exhaustive BV hits.. all BVs searched for hit, but nvidia used a scene/camera with 0 divergence. Big Navi 65% slower. For 4 wide BVH, Big Navi 30% slower

Overall, Big Navi (80 CU) 30-65% slower than RTX 3080 at RT, depending on nvidia's implementation.
Midpoint 48% slower at RT.
Eager to see some benchies leaking

I think this will become a deciding factor for buyers once one vendor can hit high frame rates with ray tracing. At the moment I see it as a bonus rather than a feature. One or two more generations and I think we'll be there.

 

[Jacky60]

Pretty sure the X800XT back in the early 2000's was $500 too...

The X800XT PE was $499 at launch, great card although it looks pathetic and puny compared to today's monsters.

 

[Nablaoperator]

I'm looking forward to those reviews too but i'm going to stick my neck on the line and say that Nvidia's RT performance will be better than AMD's.

And this is how Nvidia will justify a higher price, that's fine, its all good and proper but for someone like me for who RT is completely irrelevant..... i win. Check my signature.

Yes the price should be sweet
though i dont do RTX either..
..but you know how the broader market works, all those $$$ for trademarked marketing jargons

 

[Nablaoperator]

I think this will become a deciding factor for buyers once one vendor can hit high frame rates with ray tracing. At the moment I see it as a bonus rather than a feature. One or two more generations and I think we'll be there.

Sorry i changed my post again..
but the price will be sweet, they have got to price for the broader market which also includes those brainwashed by nvidia

 

[Jacky60]

iirc i paid just over £400 for a Hercules x800xt Platinum at launch, i googled launch price and found this..

"As far as price goes, ATI has announced an SRP of $399 for the Pro and $499 for the XT part.
Going out on a limb, I'd imagine that this will equate to British pricing of £299 and £399 respectively, although this is pure speculation."

I paid £380 for mine but that was way more than $500 iirc about £:$ was 1:1.8 back then so well over $600.

 

[Rroff]

I have changed my mind again.. but this is the best i have now.
..believe there are 2 scenarios:

• Mutually exclusive BV hits.. search aborted on first hit: Big Navi 38% slower at RT
• Exhaustive BV hits.. all BVs searched for hit, but nvidia used a scene/camera with 0 divergence. Big Navi 65% slower. For 4 wide BVH, Big Navi 30% slower

Overall, Big Navi (80 CU) 30-65% slower than RTX 3080 at RT, depending on nvidia's implementation.
Midpoint 53% slower at RT.
Eager to see some benchies leaking

I'll be surprised if it is slower than the 2080ti - worst case between 2080S and 2080ti - but still I don't want to see AMD aiming at "hybrid" or heavily constrained ray tracing which doesn't dramatically do better than what we have now given what ray tracing can bring to the table especially if you have the performance budget to collect reflected light from 1-2 bounces, etc. which is where it really starts to shine.

 

[Nablaoperator]

I'll be surprised if it is slower than the 2080ti - worst case between 2080S and 2080ti - but still I don't want to see AMD aiming at "hybrid" or heavily constrained ray tracing which doesn't dramatically do better than what we have now given what ray tracing can bring to the table especially if you have the performance budget to collect reflected light from 1-2 bounces, etc. which is where it really starts to shine.

if i were the marketing guy, i would be somewhat concerned about positioning the product..
how to go abt overplaying strengths and leveraging weaknesses?
either that or theres something big we dont know about Big Navi
is there any possibility of atleast 1 amd optimised RT game releasing before/at 28th oct for benchmarking purposes?