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.

 

[HurleyBird]

You can't just make wild statement like "GPU that costs three as much as shipping console and performs a third as good". A RTX 3070 is neither of those.

It's called a hypothetical. The point is to show that comparisons between console and PC hardware are not facially invalid as you claim. They can be difficult and muddy, but it's obvious enough that if the value proposition shifts far enough to one side, the other is affected.

Also, back in the PS4 XBone days, things are very different. The online services was no where near the money making capability as today, console had a bigger lock on game exclusivity, and the hardware at the time was not that great.

You seem to be arguing that the new consoles are selling at a greater loss this time. I haven't seen any evidence to support such an assertion. I've seen some speculation on news sites that the XSX is around $500 per unit, but take that with a pinch of salt. Most likely, they're selling at a bit higher loss than launch PS4, but a much lower loss than launch Xbone.

AMD was in a bad shape with both their CPU and GPU.

Only the total loss per unit is relevant here, but no, AMD's graphics division was in pretty good shape when they were awarded the last gen console contracts. GCN started faltering a bit against Kepler, even though the actual hardware was superior with proper drivers, and the technical decline only began in earnest against Maxwell.

On the other hand, when AMD would have been awarded the contracts for the next gen, their graphics business had been thoroughly bruised and battered.

If you're going to complain about a $500 NV GPU, then what about a Ryzen 3700x that's $300 when just a slightly slower version is in the $300 series S?

False equivalency. We aren't comparing the price of the new consoles to Turing, we're comparing them to Ampere. By the same token it's not Zen 2 parts that we need compare the new consoles against, it's Zen 3.

 

[Kenmitch]

Reviews delayed to the 16th?

Nvidia GeForce RTX 3080 Founders Edition Review: A Huge Generational Leap in Performance

The GeForce RTX 3080 delivers big gains over Turing and the RTX 2080 Ti, but at a lower price.

www.tomshardware.com

 

[DooKey]

Reviews delayed to the 16th?

Nvidia GeForce RTX 3080 Founders Edition Review: A Huge Generational Leap in Performance

The GeForce RTX 3080 delivers big gains over Turing and the RTX 2080 Ti, but at a lower price.

www.tomshardware.com

I guess they didn't get enough out to the reviewers? As long as it's before availability date I'm not too concerned.

Would you buy Nvidia anyway?

 

[Stuka87]

I guess they didn't get enough out to the reviewers? As long as it's before availability date I'm not too concerned.

Would you buy Nvidia anyway?

Typically delays like this means they are doing an emergency driver patch because of an issue found. Which sucks for reviewers that have already completed their performance testing.

 

[DooKey]

Typically delays like this means they are doing an emergency driver patch because of an issue found. Which sucks for reviewers that have already completed their performance testing.

If true that's a legitimate reason to extend the review embargo. Guess we'll find out in a few days.

 

[Konan]

If true that's a legitimate reason to extend the review embargo. Guess we'll find out in a few days.

Apparently the feedback was the reviewers wanted more time to test. So that could mean a new driver version (rumor was a new v2 one was out just very recently )

 

[Kenmitch]

I guess they didn't get enough out to the reviewers? As long as it's before availability date I'm not too concerned.

Would you buy Nvidia anyway?

Get enough what out of the reviewers? Isn't it a day to think about it once you see the reviews.

I buy whatever I perceive as the best bang for the buck in time of need. I currently have no plans to upgrade my video card. So I guess that'a no for now.

 

[DJinPrime]

It's called a hypothetical. The point is to show that comparisons between console and PC hardware are not facially invalid as you claim. They can be difficult and muddy, but it's obvious enough that if the value proposition shifts far enough to one side, the other is affected.

If your hypothetical is so far from reality, how can there be a discussion on that..

You seem to be arguing that the new consoles are selling at a greater loss this time. I haven't seen any evidence to support such an assertion. I've seen some speculation on news sites that the XSX is around $500 per unit, but take that with a pinch of salt. Most likely, they're selling at a bit higher loss than launch PS4, but a much lower loss than launch Xbone.

I guess you missed MS confirming the price and specs for both series X and S, $500 and $300 confirmed. Just doing the math on components shows that MS is eating the cost. 8 core zen2, 16 GB GDDR6, 1TB custom nvme, new RDNA2 GPU, and all the other regular stuff that goes into a console. Just do the math, retail equivalent for those parts are much higher than $500. Just the SSD is rumored to be around $200 if buying as expansion.

Only the total loss per unit is relevant here, but no, AMD's graphics division was in pretty good shape when they were awarded the last gen console contracts. GCN started faltering a bit against Kepler, even though the actual hardware was superior with proper drivers, and the technical decline only began in earnest against Maxwell.

I don't think so, seems the xbox one was using something equivalent to a 7790, and NV should be at the Kepler generation by then. I just remember not being impressed with the specs of last generation consoles.

False equivalency. We aren't comparing the price of the new consoles to Turing, we're comparing them to Ampere. By the same token it's not Zen 2 parts that we need compare the new consoles against, it's Zen 3.

I am comparing against Ampere, RTX 3070. You really think Zen 3 will be significantly better at price performance? A 2700x is still $240 atm at amazon. So, even after Zen 3, don't expect to get a 3700x much cheaper than that. And you think AMD will price big Navi significantly less than $500?

It seems you're not up to date with the new console specs and current price of things.

 

[mohit9206]

Not sure if this was already posted

 

[A///]

Don't post his trash here.

 

[Tup3x]

Inno3d has 20GB filter ready. Judging by all rumours it would seem very likely that 20GB variant is coming (be it just double VRAM variant or Ti/SUPER).

 

[Shmee]

Yeah, I would hope there is a 20GB version, I feel that would be the optimal amount of RAM for a card of that power and price range, and I don't want to pay for a 3090.

I would likely get a 20GB 3080, or a competing AMD card with at least 16GB. I kinda don't want to go less than 12 GB when I get a new card.

 

[A///]

20 GB should be great for VFX artists.

 

[Tup3x]

Yeah, I would hope there is a 20GB version, I feel that would be the optimal amount of RAM for a card of that power and price range, and I don't want to pay for a 3090.

I would likely get a 20GB 3080, or a competing AMD card with at least 16GB. I kinda don't want to go less than 12 GB when I get a new card.

20GB would be very reassuring amount. Pricing would be something that I would likely be willing to pay. RTX 3090 isn't worth it unless you really need all that VRAM (i.e. do some other stuff than gaming).

Soo.... It's looking so obvious that these are coming that I guess I'll wait a month or two before making any buying decisions.

 

[sze5003]

Yeah, I would hope there is a 20GB version, I feel that would be the optimal amount of RAM for a card of that power and price range, and I don't want to pay for a 3090.

I would likely get a 20GB 3080, or a competing AMD card with at least 16GB. I kinda don't want to go less than 12 GB when I get a new card.

Yea but at what price? $1200? If so I'd rather just get the 3090. If it's only $100 more than the 3080 that would make sense.

 

[mohit9206]

Yea but at what price? $1200? If so I'd rather just get the 3090. If it's only $100 more than the 3080 that would make sense.

Considering that most 3080s will be around $800 hence my guess would be $1k for 20GB model.

 

[tviceman]

I know I'm somewhat beating a dead horse here, but the performance gaps between the 3090 to 3080 and then 3080 to 3070 makes no sense. The gap between the 3080 and 3070 is massive. The 3070 already needs a $100 price drop and a 3070 TI to be slotted in at $550 that cuts the gap. It will be interesting to see how the 3070 (and 3080) stacks up to big Navi, especially given the ridiculous performance gap between the 3080 and 3070.

Other random commentary no one cares about: Nvidia hasn't used their big die for x80 card since the GTX 780. The 980, 1080, and 2080 were all the second tier dies. The big die has been the x80 TI model since Maxwell.

Edit hottake:

AMD hasn't price bombed on Nvidia since the HD4000 series. I don't consider their RX 5700 series jebait to have been of any significant consequence since Nvidia made their move with the supers prior to Navi's release and stuck to it. The HD4000 series, on the other hand, forced immediate price drops of existing products and refreshes (GTX 285 & GTX 260-216) a few months later.

It would be nice to see big Navi come in with twice the ram and 10% more performance than the 3070 for $500-550 BUT that would require big Navi to be 70-75% faster than the 3070 and the 3070 is already ~55% faster than Navi 10 with the same power draw sooooo.... who knows.

 

[A///]

Remind me what happened during the HD4000 era?

 

[Kuiva maa]

Remind me what happened during the HD4000 era?

4850/4870 came out cheap and so fast that forced nvidia to reduce prices of GTX260/280 like right away if my memory serves me well. There were even cases of money returns unless mistaken.

 

[.vodka]

Remind me what happened during the HD4000 era?

ATI/AMD came out of the blue with a 800sp, 250mm² 55nm GPU, RV770.

Paired with GDDR3, it was the 4850 and it destroyed G92 based cards, 324mm² big, 65nm. Paired with GDDR5 (first!) it was the 4870 and it performed the same or better than the GT200 flagships... that chip was 576mm² big, 65nm. Both AMD cards were cheaper, $/perf was awesome.

It was bad for nV. Perf/mm² difference was insane. They had to cut prices almost nearly in half in some cases.

 

[VirtualLarry]

4850/4870 came out cheap and so fast that forced nvidia to reduce prices of GTX260/280 like right away if my memory serves me well. There were even cases of money returns unless mistaken.

Not only that, but vendors were forced to offer cash-equivalent refunds to early purchasers of those cards!

 

[VirtualLarry]

ATI/AMD came out of the blue with a 800sp, 250mm² 55nm GPU, RV770.

Paired with GDDR3, it was the 4850 and it destroyed G92 based cards, 324mm² big, 65nm. Paired with GDDR5 (first!) it was the 4870 and it performed the same or better than the GT200 flagships... that chip was 576mm² big, 65nm. Both AMD cards were cheaper, $/perf was awesome.

I actually had FOUR HD4850 VisionTek reference single-slot (COPPER cooler) cards. Bought them all at two different BestBuy stores BEFORE official release (by a few days), for $150 (+tax) ea. B&M.

But I just let them languish in their boxes for a year, since I wasn't PC gaming much at the time. I used them for F@H (Dec. race) one year, two of them, I think.

If I had used them all to mine Bitcoin, back in the REALLY early days, I too, could have been a Bitcoin Billionare. Sigh. Missed opportunities. Like investing in Tesla.

 

[DJinPrime]

I know I'm somewhat beating a dead horse here, but the performance gaps between the 3090 to 3080 and then 3080 to 3070 makes no sense. The gap between the 3080 and 3070 is massive. The 3070 already needs a $100 price drop and a 3070 TI to be slotted in at $550 that cuts the gap. It will be interesting to see how the 3070 (and 3080) stacks up to big Navi, especially given the ridiculous performance gap between the 3080 and 3070.

$200 over $500 is a 40% increase, so aren't the 3070 and 3080 basically inline with price/performance? Maybe they'll do something like the Super in the last generation, just release a better version at the same price to combat big Navi if needed.
Something else to consider with this new generation, the power requirement jump is pretty big from 3070 to 3080. I'm leaning toward getting the 3070 partly because I'm not sure if my current 600 watts ps is enough, it's old. I will need to buy a new PSU for sure if I get a 3080, so that's a potential hidden cost with 3080. Also, I'm on a 1440 ultrawide monitor, do I really need the extra power if I'm not going to game at 4k? Definately will wait until 3070 benchmarks are released.

 

[maddie]

ATI/AMD came out of the blue with a 800sp, 250mm² 55nm GPU, RV770.

Paired with GDDR3, it was the 4850 and it destroyed G92 based cards, 324mm² big, 65nm. Paired with GDDR5 (first!) it was the 4870 and it performed the same or better than the GT200 flagships... that chip was 576mm² big, 65nm. Both AMD cards were cheaper, $/perf was awesome.

It was bad for nV. Perf/mm² difference was insane. They had to cut prices almost nearly in half in some cases.

The unintended consequence of that move was Nvidia radically rethinking their GPU designs. I firmly believe the Kepler line, a big break from earlier generations, was a reaction to AMD's simpler, highly packed shaders. Two great generations from AMD and then back to mediocrity. Nvidia reacted strongly and the events since are solely what most gamers know today.

The lesson here is if you shake up the market, be prepared to have "as good" follow-on designs for at least several generations. Their CPU division, I think, appears to have learnt that lesson. Their messaging on RDNA & CDNA is promising.

 

[HurleyBird]

It was bad for nV. Perf/mm² difference was insane. They had to cut prices almost nearly in half in some cases.

Terrascale was the most frustrating instance of AMD shooting themselves in the foot. They had a massive technological advantage and used that to... only produce small chips so as to surrender the performance crown to Nvidia.

And they bragged non-stop about how awesome their "sweet spot" strategy was. It wasn't. AMD could have had a G80 or R300 moment if they hadn't totally squandered it.