a little thought experiment with pricing on recent nvidia cards

[VeganJoy]

i buy used a lot cause im half broke lol, so prices are generally pretty relative to real performance rather than silly retail and ebay prices. seeing today's 3080 reviews has had me thinking: what if turing had been normally priced? how good would ampere be considered to be? if we pretend that the original 20 series were launched with 20xx super performance and retailed at the usual prices, and the 2080ti was released a year later also at 700, would ampere be that gRoUnDbReAkInGlY iNsAnE? 

[RadiatingLight]

Probably not, but it would still be a huge leap. Reviews are comparing against the 2080 Super, not the 2080, and still finding that the GPUs are a huge step up.

[SolarNova]

23 minutes ago, VeganJoy said:

i buy used a lot cause im half broke lol, so prices are generally pretty relative to real performance rather than silly retail and ebay prices. seeing today's 3080 reviews has had me thinking: what if turing had been normally priced? how good would ampere be considered to be? if we pretend that the original 20 series were launched with 20xx super performance and retailed at the usual prices, and the 2080ti was released a year later also at 700, would ampere be that gRoUnDbReAkInGlY iNsAnE? 

Easy answer really.

 

You compare based on price.

So if the 2080ti released at the usual price of $700 (and thus the 2080 below that), then u''d compare as follows

 

1080ti vs 2080ti vs 3080 . At which point u look at the current benchmarks of the 3080 vs the 2080ti and see a more modest increase. (~30%) This is closer to a 'normal 'single generation increase. The 1080ti to 2080ti was roughly 30% as well.

 

This will likely change when the 3080ti is released, if its anything like the 3090 we can expect a 15% to 20% improvement over the 3080. Should it take the 3080's price point as it should (as in previous generations), then at that point u can compare the 1080ti vs 2080ti vs 3080ti and say the 3080ti is above the 'average' generational increase.

 

ofc the 2080ti wasn't a $700 card, and the 3080ti may end up being a $1000 card if Nvidia pull a 'Fk you' like they did with the 2080ti ..so in terms of comparisons at set price points ..its a different story.

[xg32]

imho  the comparison should be 2080S->3080 at 700 and 2080ti->3090 at whatever (fastest single cards)

 

Now Turing on release was weird, the 2080 was the 700dollar card and the 2080 ti was the fastest single, the 1080 ti was both, but seeing as there was no performance increase from the 1080 ti to the 2080, i'll compare it to the 2080 ti.

 

1080 ti ->2080 ti was a 30% improvement, 2080S->3080 is a 60% improvement. And the jump from 2080 ti -> 3090 looks to be ~50%. So it's a 50-60% generational improvement.

 

Nvidia isn't fooling anyone though by calling the 3080 "flagship" when the 3090 exists and the 20G is coming out later, that said, it's still a big jump.

[VeganJoy]

i guess my thing is that ampere doesnt seem super crazy as i'm mostly comparing it against the 1080ti and considering the fact that its been 2 generations. so whatever its capable of kinda needs to make up for turing being shit, and my idea with the alternate reality pricing would be that ampere is pretty in-line with standard generational gains. great but not specifically noteworthy aside from turing being exceptionally not so. maybe if we compare 1080 -> 2080 -> 3080 its more fair? i dunno, not sure how a 3080ti will look as the 3080 uses the ga102 die with very little locked out. 

[IkeaGnome]

Wouldn't it make more sense to do cost per frame for comparing? Then you could, in theory, compare them all?

[VeganJoy]

1 minute ago, Voluspa said:

Wouldn't it make more sense to do cost per frame for comparing? Then you could, in theory, compare them all?

that sounds reasonable but i aint a fancy maths person lol. youd have to average it out across a variety of games at different resolutions i suppose. 

[xg32]

7 minutes ago, Voluspa said:

Wouldn't it make more sense to do cost per frame for comparing? Then you could, in theory, compare them all?

 

4 minutes ago, VeganJoy said:

that sounds reasonable but i aint a fancy maths person lol. youd have to average it out across a variety of games at different resolutions i suppose. 

the aussies got u covered, i expect the 3070 to be more efficient here, but not by much.

 

 

[VeganJoy]

2 minutes ago, xg32 said:

 

the aussies got u covered, i expect the 3070 to be more efficient here, but not by much.

 

snip

wait how is the 2070 super less efficient than the og 2070??

[xg32]

1 minute ago, VeganJoy said:

wait how is the 2070 super less efficient than the og 2070??

good eye, i'd guess an old price with old data or something.

[VeganJoy]

9 minutes ago, xg32 said:

good eye, i'd guess an old price with old data or something.

as an aside, with the many architectural improvements from turing to ampere, it doesnt seem like that much of a jump considering the power draw. it uses 50% more power than the 2080/2080 super and is just a bit over 50% faster than those, and 30% higher tdp than the 2080ti for just over 30% gains on average. considering the staggering number of transistors, new 7/8nm process, the new cuda core arrangement, the gddr6x, and all those sorta things, id kinda expect a little more tbh. the well-engineered founders cooler is perfectly capable of dissipating the heat and its at a normal price point but it just seems weird to me

[IkeaGnome]

32 minutes ago, xg32 said:

good eye, i'd guess an old price with old data or something.

Looking on PCPP, prices are pretty close. At least in USD. It actually favors the 2060 over the 2060 super slightly more. 2060 S has gone up about $20ish, while the 2060 is down $50ish from their pricing on the video.

 

Couldn't see the 2070S buried way at the bottom so I assumed you were talking about the 2060. You know what they say about assumptions, except it was just me. I'm almost wondering if they got the 2070 and the 2070S mixed on that graph(2070 with 2070S data)