Question 'Ampere'/Next-gen gaming uarch speculation thread

[Ottonomous]

How much is the Samsung 7nm EUV process expected to provide in terms of gains?
How will the RTX components be scaled/developed?
Any major architectural enhancements expected?
Will VRAM be bumped to 16/12/12 for the top three?
Will there be further fragmentation in the lineup? (Keeping turing at cheaper prices, while offering 'beefed up RTX' options at the top?)
Will the top card be capable of >4K60, at least 90?
Would Nvidia ever consider an HBM implementation in the gaming lineup?
Will Nvidia introduce new proprietary technologies again?

Sorry if imprudent/uncalled for, just interested in the forum member's thoughts.

 

[VirtualLarry]

Ampere, It's OVER 10,000!!!!

 

[SpaceBeer]

But wouldn't 2x3080 20GB be way better (and bit more expensive) than 3090?
https://techgage.com/article/nvidia-geforce-rtx-3090-proviz-performance/
https://www.pugetsystems.com/labs/a...-3080-10GB-RTX-3090-24GB-Review-Roundup-1899/

 

[Bouowmx]

If your work doesn't use more than 10 GB and can be parallelized to multiple non-bridged GPUs (the latter applying to distributed computing), makes more sense to get 2x GeForce RTX 3080 (17408 cores) than 1x 3090 (10496 cores).

 

[aleader]

I want one due to the over 10,000 cuda cores. For those that can afford it (I can) its worth it. I use it for DC and it seems that cuda cores are almost equal to performance.

I can 'afford it' too. No way I'd spend that much on a GPU for playing games, period.

This is likely exactly what Nvidia is going to do (release a 3060 ti). I'm going to wait for AMD regardless as it's too close now not to, but I still think we'll be able to get all of these cards by the end of November (and all of the benchmarks will be out by then):

https://www.pcgamer.com/will-nvidia-try-to-spoil-amds-big-navi-launch-with-a-geforce-rtx-3060-ti/

 

[CP5670]

The 3090 is really good for machine learning work because of its 24GB memory, while being much cheaper than a Tesla or Titan RTX. More memory lets you train complex models with more parameters and larger batches. It's only pointless for gaming.

 

[DooKey]

New drivers out from NV that are supposed to address some crashing problems.

 

[Mopetar]

Where are you getting the idea that it isn't random?

The people who respond self select. Go read the link I posted. Steam surveys are not a random sample.

Even if Valve truly does select a random subset of people to offer the survey to, the people who actually participate are those who choose to. If you can't think of any possible ways that might lead to some kind of bias you're not trying particularly hard

New drivers out from NV that are supposed to address some crashing problems.

Good news that drivers can fix the issue. Hopefully it isn't anything asinine like just lowering the boost, but I can't see NVidia doing something like that.

 

[Heartbreaker]

The people who respond self select. Go read the link I posted. Steam surveys are not a random sample.

Even if Valve truly does select a random subset of people to offer the survey to, the people who actually participate are those who choose to. If you can't think of any possible ways that might lead to some kind of bias you're not trying particularly hard

The link you posted had nothing to do with Steam Survey.

Generally, every survey, outside of things like official government info collection (Census) in some countries, allows people to opt out.

Just because a survey allows an opt out, does not make it non random.

 

[Markfw]

The link you posted had nothing to do with Steam Survey.

Generally, every survey, outside of things like official government info collection (Census) in some countries, allows people to opt out.

Just because a survey allows an opt out, does not make it non random.

Speaking as a statistician, you are wrong. Any REAL random population becomes corrupted and unusable for any real stats, the minute the population has been altered.

If you don't believe be, then take a statistics class, then talk to me.

And also notice that I said a real random population. The fact that a large number of people do not even use steam, and a lot of people don't even play games, there is no way the population is random.

 

[Stuka87]

Good news that drivers can fix the issue. Hopefully it isn't anything asinine like just lowering the boost, but I can't see NVidia doing something like that.

Except that the fix is most likely putting a cap on boost clocks.

 

[Heartbreaker]

Speaking as a statistician, you are wrong. Any REAL random population becomes corrupted and unusable for any real stats, the minute the population has been altered.

If you don't believe be, then take a statistics class, then talk to me.

By that logic, nearly all survey data is unusable, since with rare exceptions, they all allow opting out.

While I am not a statistician, I have taken university level statistics as part of a CS degree.

You will need a lot more than "I am a statistician and this is the way it is" rhetoric.

Point to some sources, that indicate there is an over-riding rule, that any survey that allows opting out is worthless.

You can't, because there isn't.

There will be appropriate caution on not constructing or delivering a survey in such a fashion, that it drives away certain groups, and thus tilting the Survey. But there is no blanket prohibition against allowing people to opt out.

 

[Qwertilot]

No, but there will inevitably be some biases introduced.

Not that I can easily think how the steam survey results could be systematically badly biased when used for measuring GPU market share.

 

[Hitman928]

By that logic, nearly all survey data is unusable, since with rare exceptions, they all allow opting out.

While I am not a statistician, I have taken university level statistics as part of a CS degree.

You will need a lot more than "I am a statistician and this is the way it is" rhetoric.

Point to some sources, that indicate there is an over-riding rule, that any survey that allows opting out is worthless.

You can't, because there isn't.

There will be appropriate caution on not constructing or delivering a survey in such a fashion, that it drives away certain groups, and thus tilting the Survey. But there is no blanket prohibition against allowing people to opt out.

Very few polls/surveys are truly random but the people administering them know this and base their statistic modeling on weighted factors taking into account the demographics/filtering data of the respondents. That's why trusting the Steam survey is difficult, because Steam doesn't publish any of this info so we have no idea how trustworthy the survey is or even how representative the numbers are for what country/group/demographic/market/etc.

Edit: I'm sure the Steam survey tells us something, it's just without any methodology information I can't tell you what that something is.

 

[Heartbreaker]

Very few polls/surveys are truly random but the people administering them know this and base their statistic modeling on weighted factors taking into account the demographics/filtering data of the respondents. That's why trusting the Steam survey is difficult, because Steam doesn't publish any of this info so we have no idea how trustworthy the survey is or even how representative the numbers are for what country/group/demographic/market/etc.

This really seems more like FUD to me. We don't have precise details of everything, so we must discount it, because there might be unknown biases.

The reality is, that until something better comes along, it's the only game in town for granular information on GPU sales, and in all likelyhood it is a fair representation of GPUs in the hands of gamers.

As I also said before, it does more or less correspond with other Supply side industry surveys if you find them more accurate:

Nvidia graphics cards now account for 80% of the PC market

Nvidia reportedly has 80% of the discrete PC GPU market, Jon Peddie Research claims.

www.pcgamer.com

They just don't have the same granularity.

 

[A///]

Jon Peddie is the industry standard of NEW units sold.

 

[Hitman928]

This really seems more like FUD to me. We don't have precise details of everything, so we must discount it, because there might be unknown biases.

The reality is, that until something better comes along, it's the only game in town for granular information on GPU sales, and in all likelyhood it is a fair representation of GPUs in the hands of gamers.

As I also said before, it does more or less correspond with other Supply side industry surveys if you find them more accurate:

Nvidia graphics cards now account for 80% of the PC market

Nvidia reportedly has 80% of the discrete PC GPU market, Jon Peddie Research claims.

www.pcgamer.com

They just don't have the same granularity.

Does is follow for every quarter, not just the latest report? What about when the market researchers showed AMD having over 30% of the discrete market sales, did Steam survey reflect that as well?

 

[Tup3x]

I want one due to the over 10,000 cuda cores. For those that can afford it (I can) its worth it. I use it for DC and it seems that cuda cores are almost equal to performance.

But for the price of 3090 you can get two 3080 cards... And much more CUDA cores!

 

[Heartbreaker]

Does is follow for every quarter, not just the latest report? What about when the market researchers showed AMD having over 30% of the discrete market sales, did Steam survey reflect that as well?

Steam shows the cumulative results over time. Industry snapshots of quarterly sales, obviously only represent that quarter. The latter can obviously fluctuate faster than cumulative results.

Overall, Steam and Industry sales are in the same 70/30 to 80/20 range over time.

People keep acting like unless Steam HW results are 100% perfect, they are unusable.

They are obviously in the ballpark, and of more than sufficient quality for something as trivial as forum discussions.

People routinely use anecdotes about what they see on reddit/forums as the basis of argument, but Steam HW survey is not rigorous enough?

 

[Markfw]

But for the price of 3090 you can get two 3080 cards... And much more CUDA cores!

The problem there is memory, and power. Each only has 10 gig ram, and consumes 320 watt, while the 3090 has 24 gig (more than twice) of ram, and only 360 wayy vs 640 watt.

Where am I wrong here ?

 

[Hitman928]

Steam shows the cumulative results over time. Industry snapshots of quarterly sales, obviously only represent that quarter. The latter can obviously fluctuate faster than cumulative results.

Overall, Steam and Industry sales are in the same 70/30 to 80/20 range over time.

People keep acting like unless Steam HW results are 100% perfect, they are unusable.

They are obviously in the ballpark, and of more than sufficient quality for something as trivial as forum discussions.

People routinely use anecdotes about what they see on reddit/forums as the basis of argument, but Steam HW survey is not rigorous enough?

You won't find me using anecdotes as evidence. As far as Steam goes, as I said, I'm sure it tells us something, but what is that something I can't say exactly and it has departed from quarter to quarter market research findings on many occasions, even over significant amounts of time with the Q/Q numbers showing very strong trends that don't get reflected in Steam numbers. It only gets worse when looking at individual models. If you want to trust the Steam numbers, that's fine, but I won't be trusting any numbers where next to nothing is known of methodology, no matter how big the platform is.

 

[Hitman928]

The problem there is memory, and power. Each only has 10 gig ram, and consumes 320 watt, while the 3090 has 24 gig (more than twice) of ram, and only 360 wayy vs 640 watt.

Where am I wrong here ?

As far as I know none of the DC projects require very large amounts of VRAM but correct me if I'm wrong.

Between the two options, I would say if PCIe slot space isn't an issue then 2x3080s would probably be the better buy but if slot space is an issue, then 3090 is probably better. Once the full system power is taken into account, the 3090 will probably be at best a wash in efficiency but at much lower performance. Even if it is a slight efficiency win it would be at much lower performance and I doubt the efficiency difference over the lifetime of the cards would make up for the big performance gap.

 

[Ajay]

Edit: I'm sure the Steam survey tells us something, it's just without any methodology information I can't tell you what that something is.

They tell us what a select group of users, who have opted in to the survey, are using as their primary graphics card. That's it, it has no statistical value - it's just a marketing tool.

 

[KentState]

Finally got it installed today, but had to undo half the system. Had to dig up my old black/red 8-pin to power it up. Hopefully EK or Alphacool start making a 3090 block for the EVGA card.

 

[CP5670]

Is the vertical mounting worth it? Most cases have the vertical slots pretty close to the side panel and seem to restrict the card's airflow a lot. It looks you're not using those slots and the card is placed further inward, so maybe this is less of a problem.

 

[KentState]

Is the vertical mounting worth it? Most cases have the vertical slots pretty close to the side panel and seem to restrict the card's airflow a lot. It looks you're not using those slots and the card is placed further inward, so maybe this is less of a problem.

The prior card was part of the loop so I just reused the vertical mount. The mount itself is from Cooler Master and puts the card about 1” from the motherboard and far from the glass. Only way I would do vertical and no issues with temps, even when sucking 450 watts.