RTX 3xxx performance speculation

[sethk]

I suspect pricing will be no higher than the previous generation at launch with the exception of the 3090 being a new higher RAM sku with the 3080ti coming in later at a lower price with less RAM.

 

[Aireoth]

I feel the opposite. A quiet AMD historically hasn't been met with an awesome product launch.

I have noticed the opposite, the weaker the amd product the more marketing and hype it gets.

 

[Snowdog]

Is there any explaination to why these cards have such a high TDP (320-350w) in these predicted specs? I haven't been following the 3XXX series that closely.

More performance takes more power. Also don't forget that Ray Tracing consumes a lot of power as well, and these cards are rumored to have grown RT performance more than Raster performance.

 

[reaper12]

Is there any explaination to why these cards have such a high TDP (320-350w) in these predicted specs? I haven't been following the 3XXX series that closely.

I read somewhere that the 320-350W figures weren't TDP but TGP.

 

[Mylex]

I read somewhere that the 320-350W figures weren't TDP but TGP.

Let's say it's TGP that's still going from 215 on the 2080 and 250 on the TI to now 320/350 respectively. Not a good look, but hey I'm ok with vehicles that get single digit mpg if it excels at what I want it to do so I will pay if the numbers add up.

 

[RamonGTP]

I have noticed the opposite, the weaker the amd product the more marketing and hype it gets.

You notice this from AMD or from their fans? Two very different things.

When was the last time AMD was real quiet and surprised us with awesomeness? The last truly great product was Zen 2 and they were anything but quiet.

 

[Snowdog]

Let's say it's TGP that's still going from 215 on the 2080 and 250 on the TI to now 320/350 respectively. Not a good look, but hey I'm ok with vehicles that get single digit mpg if it excels at what I want it to do so I will pay if the numbers add up.

Just for example, it's expected that at minimum 3090 is supposed to have at least +50% Raster performance.

250 Watts + 50% = 375 Watts

So 350 Watts would still be an improvement in Perf/Watt over 2080Ti.

And that's before you get to Ray Tracing. Ray Tracing uses a lot of power, and RT looks to have improved by MUCH more than 50%.

 

[Aireoth]

You notice this from AMD or from their fans? Two very different things.

When was the last time AMD was real quiet and surprised us with awesomeness? The last truly great product was Zen 2 and they were anything but quiet.

OG Ryzen was very sedate in comparison to their usual hullabaloo, and it turned out amazing.

 

[Snowdog]

I wouldn't read anything into the AMD silence either way.

I also expect we might hear more from AMD before NVidia ships to try to get people to wait for Big Navi.

 

[madpistol]

I wouldn't read anything into the AMD silence either way.

I also expect we might hear more from AMD before NVidia ships to try to get people to wait for Big Navi.

AMD is probably trying to get a bead on what Nvidia's focus is and see if there's any holes in their strategy. If they find one, that's when they strike. They're playing a long game; they're letting nvidia make the first move, and then they're countering.

 

[reaper12]

Let's say it's TGP that's still going from 215 on the 2080 and 250 on the TI to now 320/350 respectively. Not a good look, but hey I'm ok with vehicles that get single digit mpg if it excels at what I want it to do so I will pay if the numbers add up.

Well, TGP is max power that the PSU will need to supply. The 2080 and 2080Ti figures you mention are TDP. The 2080 uses a more than 215Watts and the 2080Ti uses more than 250W. So the actual power use Turing and Ampere might be a lot closer than we think.

Of course this all depends on those figures been TGP not TDP. So still speculation

 

[Mylex]

Well, TGP is max power that the PSU will need to supply. The 2080 and 2080Ti figures you mention are TDP. The 2080 uses a more than 215Watts and the 2080Ti uses more than 250W. So the actual power use Turing and Ampere might be a lot closer than we think.

Of course this all depends on those figures been TGP not TDP. So still speculation

Try again, its ok that you were mistaken. I understand the Difference between TDP,TGP and TBP. While my join date shows 2018 I've been around since the early 2000's under a different name.

 

[Zarathustra[H]]

One of the bases of the speculation that 3080 and 3090 are on 7nm, while 3070 and below are 8nm.

I'm squarely in the $600-800 camp so will likely be waiting to see what AMD has, but will be very interesting seeing the performance on the entire 30xx line up tomorrow. Actually, when do NDAs end... Tomorrow, or will it be like Turing where preorders can be done through nV, reviews don't appear until they are available for purchase?

Who even has an 8nm process?

Samsung?

 

[vegeta535]

They were pretty quiet with Ryzen. That crap came out of no where in terms of performance. For the last 3 years AMD has been under promising and over delivering.

Not in the gpu space.

 

[RJ1892]

I read somewhere that the 320-350W figures weren't TDP but TGP.

Ok makes sense now. Was expecting TDP to atleast maintain 250-300w since they die shrink of 12nm to 7nm.

I just hope we get some decent raster gains as well as RT, not many games using RT right now. I would like 50% more raster performance but, guessing its gonna be 20-30%.

 

[Algrim]

OG Ryzen was very sedate in comparison to their usual hullabaloo, and it turned out amazing.

Compared to Bulldozer it was... Now Zen2 is quite amazing as AMD finally mitigated the weaknesses of the original Zen architecture.

 

[Aireoth]

Compared to Bulldozer it was... Now Zen2 is quite amazing as AMD finally mitigated the weaknesses of the original Zen architecture.

getting too OT, all I am saying is that in my experience the quiter AMD is prior to launch, the more confident they are that the product itself will impress.

I doubt they'll hit the top end and compete with a 3090, but who knows? Also intel in the new year (enternity for some, but with my 2080ti I'm not in that much of a rush anyway).

I am excited to see what this launch will bring, tomorrows going to be fun.

 

[reaper12]

Try again, its ok that you were mistaken. I understand the Difference between TDP,TGP and TBP. While my join date shows 2018 I've been around since the early 2000's under a different name.

Other board partners and review sites all mention those figures as TDP.

If those figures are both TGP according to Gainward, then, yeah, it's looking like the 3090 will be using over 375watts.

 

[RamonGTP]

OG Ryzen was very sedate in comparison to their usual hullabaloo, and it turned out amazing.

In compared to BD maybe. It was still well behind intel in nearly all metrics. Not a good example if that’s the best you can think of.

 

[Aireoth]

In compared to BD maybe. It was still well behind intel in nearly all metrics. Not a good example if that’s the best you can think of.

wtf does Intel have to do with anything i said?

 

[madpistol]

In compared to BD maybe. It was still well behind intel in nearly all metrics. Not a good example if that’s the best you can think of.

It was amazing because it made Intel scramble to stay on top. Even now, Intel's best gaming CPU uses so much power compared to the AMD competition.

AMD innovated, and everyone is better for it.

 

[RamonGTP]

wtf does Intel have to do with anything i said?

when you say a product is amazing what are you comparing it to? You have two options

1) against their own product which was complete shit which means it wasn't all that amazing

2) against their main rival which would be Intel and again. Not that amazing

you realize zen is a CPU and Intel is the main competition right? I’m a little shocked this needed to be explained. I mean we are comparing RDNA2 to what nvidia is offering correct? Where exactly is your confusion here
When Intel is brought up during a zen discussion?

 

[Aireoth]

when you say a product is amazing what are you comparing it to? You have two options

1) against their own product which was complete shit which means it was t all that amazing

2) against their main rival which would be Intel and again. Not that amazing

you realize zen is a CPU and Intel is the main competition right? I’m a little shocked this needed to be explained. I mean we are comparing RDNA2 to what nvidia is offering correct? Where exactly is your confusion here?

I think your dancing and strawmanning to back up your position against a simple anecdotal observation.

The less hyped an amd product, the better the product in my experience. Full stop.

If you need more, Navi 2 hasn't been hyped near as much as prior AMD gpus, by either amd or fans, so they might have something, particularly with how balls out nVidia is going.

*even more, its a simple metric, and any company can be guilty of it. Over marketing an inferior product to increase initial sales before the market place realizes the product is an inferior one, vs less up front marketing and more reliance on the actual performance of said product to sell it. This can apply to anything and any company, but in my experience, AMD typically generates a lot of hype that they then rely on their fans to project and echo, when their product is inferior, and the only recent example of them not creating a lot of hype and launching something that actually shocked and shook up the market was Ryzen. Now I am seeing a similar trend with Navi 2, where it isn't hyped or marketed to the usual AMD extent. Could I be wrong, sure, I don't have a crystal ball or any insider knowledge. Just plan and simple critical thinking and a knowledge of how marketing functions in the sales cycle.

This has nothing to do with Intel, but admittedly has a bit to do with nVidia and how crazy they are going with this launch, reminds me of their sudden 980 Ti drop out of some concern that AMD's gpu would challenge them, IIRC AMD flubbed it and we ended up with the rather amazing 980 Ti. So maybe AMD will flub it again and we will end up with an amazing 3090?

 

[RamonGTP]

If saying the same thing is dancing around then I’m dancing around. I asked for an example. You gave me one of CPUs. I mentioned how that was a bad one becisse it really wasn’t amazing compared to its competitors and you got thoroughly confused by that comment.

I get what you’re saying, but it appears your experiences don’t actually exist. Full stop

 

[Aireoth]

If saying the same thing is dancing around then I’m dancing around. I asked for an example. You gave me one of CPUs. I mentioned how that was a bad one becisse it really wasn’t amazing compared to its competitors and you got thoroughly confused by that comment.

I get what you’re saying, but it appears your experiences don’t actually exist. Full stop

bye.

 

[noko]

I think your dancing and strawmanning to back up your position against a simple anecdotal observation.

The less hyped an amd product, the better the product in my experience. Full stop.

If you need more, Navi 2 hasn't been hyped near as much as prior AMD gpus, by either amd or fans, so they might have something, particularly with how balls out nVidia is going.

*even more, its a simple metric, and any company can be guilty of it. Over marketing an inferior product to increase initial sales before the market place realizes the product is an inferior one, vs less up front marketing and more reliance on the actual performance of said product to sell it. This can apply to anything and any company, but in my experience, AMD typically generates a lot of hype that they then rely on their fans to project and echo, when their product is inferior, and the only recent example of them not creating a lot of hype and launching something that actually shocked and shook up the market was Ryzen. Now I am seeing a similar trend with Navi 2, where it isn't hyped or marketed to the usual AMD extent. Could I be wrong, sure, I don't have a crystal ball or any insider knowledge. Just plan and simple critical thinking and a knowledge of how marketing functions in the sales cycle.

This has nothing to do with Intel, but admittedly has a bit to do with nVidia and how crazy they are going with this launch, reminds me of their sudden 980 Ti drop out of some concern that AMD's gpu would challenge them, IIRC AMD flubbed it and we ended up with the rather amazing 980 Ti. So maybe AMD will flub it again and we will end up with an amazing 3090?

I would not read much of anything in quietness. Seriously. AMD was rather quiet themselves about RNDA, it was decent but not spectacular, Nvidia did a few adjustments (Super Line) and things went back to normal.

I do question the performance choices on Nvidia upcoming design:
- DDR6X requiring better cooling ending up with dual sided configuration - added cost
- cooler has to support that configuration - added cost
- to support DDR6x the board has to be 12 layers with back drilling - Adding costs
- Resulting in a 3 slott configuration dumping some serious heat at stock settings in the case - limiting use cases

Seems like the trade offs vice just using HBM may not be worth it. HBM memory bandwidth, ease of cooling, power efficiency, board simplicity maybe the cheeper way to go. Hynix mass production of HBM2e as reported by AnAndTech July 2 might be very good for us. AMD and Hynix have a close relationship, Nvidia is using Samsung HBM2e for A100.

The performance may not be an issue with Amper cards, being able to use them in your confirguration may be.

 

[Snowdog]

I would not read much of anything in quietness. Seriously. AMD was rather quiet themselves about RNDA, it was decent but not spectacular, Nvidia did a few adjustments (Super Line) and things went back to normal.

Really this part of the discussion is off topic (this is not the AMD thread) and kind of pointless. NVidia was quiet too, until they put up the countdown timer for tomorrows event. Everything else was leaks, and I don't think the leaks are on purpose from NVidia. They would want to preserve the surprise.

I do question the performance choices on Nvidia upcoming design:
- DDR6X requiring better cooling ending up with dual sided configuration - added cost
- cooler has to support that configuration - added cost
- to support DDR6x the board has to be 12 layers with back drilling - Adding costs
- Resulting in a 3 slott configuration dumping some serious heat at stock settings in the case - limiting use cases

Seems like the trade offs vice just using HBM may not be worth it. HBM memory bandwidth, ease of cooling, power efficiency, board simplicity maybe the cheeper way to go. Hynix mass production of HBM2e as reported by AnAndTech July 2 might be very good for us. AMD and Hynix have a close relationship, Nvidia is using Samsung HBM2e for A100.

GDDR6X was almost certainly a better decision because than regular GDDR6, since they can get away with less memory controllers. Those are power hungry and waste die space.

I'm sure we would all like to be a fly on the wall, and see why they didn't go with HBM. Clearly they know how to use it, and it would have saved power, so it seems most likely that it's still more expensive.

The performance may not be an issue with Amper cards, being able to use them in your confirguration may be.

I'll be sure to shed tears for those who can't fit 3090 cards into their ITX boutique builds.

 

[Mylex]

Other board partners and review sites all mention those figures as TDP.

If those figures are both TGP according to Gainward, then, yeah, it's looking like the 3090 will be using over 375watts.

The first slide is from Gainward the second slide is directly from Nvidia when they did a presentation on the in efficiency of RDNA. The specs pages for both show the same numbers for the non founder's editions, those are 10 watts higher.

 

[noko]

Really this part of the discussion is off topic (this is not the AMD thread) and kind of pointless. NVidia was quiet too, until they put up the countdown timer for tomorrows event. Everything else was leaks, and I don't think the leaks are on purpose from NVidia. They would want to preserve the surprise.




GDDR6X was almost certainly a better decision because than regular GDDR6, since they can get away with less memory controllers. Those are power hungry and waste die space.

I'm sure we would all like to be a fly on the wall, and see why they didn't go with HBM. Clearly they know how to use it, and it would have saved power, so it seems most likely that it's still more expensive.



I'll be sure to shed tears for those who can't fit 3090 cards into their ITX boutique builds.

Just a reflection point on when DDR6x maybe the last useful iteration. Performance is probably not best compared in a vacuum, meaning it would be nice if Nvidia actually had some real competition not only in RT but rasterization. Might be very interesting to see the design choices of Nvidia and AMD to get to a certain performance level which will probably be all over the place with RT, DLSS and anything else.

Now what is wrong with an ITX boutique build with a 1000w P/S crammed in and a 3090?

 

[Chimpee]

Now what is wrong with an ITX boutique build with a 1000w P/S crammed in and a 3090?

Nothing wrong with that as long you can fit it and cool it. That being said, my NCase M1 will not be thrill if I try to shove a 3090 into the case unless if I water cool it.

 

[noko]

Nothing wrong with that as long you can fit it and cool it. That being said, my NCase M1 will not be thrill if I try to shove a 3090 into the case unless if I water cool it.

I wonder if EVGA will have a water cooled version? Since the board is rather small, it could actually be very good after water cooling, that is if a suitable block can be made or bought. So a huge beefy air cooler to a slick short water cooled version would be rather cool.

Actually for me, that is probably the only viable way I would really want one.

 

[johnnysd]

I am interested to see how loud they are. In terms of AMD I think they are going to be more competitive than people here think. NVIDIA will likely not show a single raster only benchmark tomorrow so probably at least a week before we have real world benchmarks.

 

[Zarathustra[H]]

The first slide is from Gainward the second slide is directly from Nvidia when they did a presentation on the in efficiency of RDNA. The specs pages for both show the same numbers for the non founder's editions, those are 10 watts higher.

This is the second time in as many weeks I've heard the name Gainward, since not hearing it since probably the 90's. Are they making a comeback or something?

 

[Mylex]

This is the second time in as many weeks I've heard the name Gainward, since not hearing it since probably the 90's. Are they making a comeback or something?

lol, I had a similar moment when I googled and those were the images I saw first. I think maybe they need a return the North American market.

 

[Snowdog]

I am interested to see how loud they are. In terms of AMD I think they are going to be more competitive than people here think. NVIDIA will likely not show a single raster only benchmark tomorrow so probably at least a week before we have real world benchmarks.

Yeah, totally. Poor Ampere.

 

[Gideon]

I seriously wonder how weak the raster gain is going to be on this generation. Seems like all the talk about these coming cards is all about Ray Tracing performance.

 

[noko]

Yeah, totally. Poor Ampere.

Well Ampere will be true to it's name, AMPS and more AMPS consumed . . .

 

[Snowdog]

I seriously wonder how weak the raster gain is going to be on this generation. Seems like all the talk about these coming cards is all about Ray Tracing performance.

Not sure where this is coming from, since raster performance has been talked about/rumored/leaked for ages.

The Raster numbers are fairly consistently rumored at: 3080 is 20-30% faster than 2080 ti, and 3090 is about 50% faster than 2080 Ti.

That also lines up with the memory bandwidth increases on the cards as well, and that is verified right off the pages on AIB cards.

 

[n370zed]

Not sure where this is coming from, since raster performance has been talked about/rumored/leaked for ages.

The Raster numbers are fairly consistently rumored at: 3080 is 20-30% faster than 2080 ti, and 3090 is about 50% faster than 2080 Ti.

That also lines up with the memory bandwidth increases on the cards as well, and that is verified right off the pages on AIB cards.

Hopefully we can squeeze out a few more % of performance with a mild OC I’m going with an RTX 3080 myself.

 

[Astral Abyss]

I wonder if EVGA will have a water cooled version? Since the board is rather small, it could actually be very good after water cooling, that is if a suitable block can be made or bought. So a huge beefy air cooler to a slick short water cooled version would be rather cool.

Actually for me, that is probably the only viable way I would really want one.

EVGA Hybrid for me. My Titan XP was a dud until I added it.