Nobina
Posts: 4,509 +5,528
Why are you chasing 4K/120? You'd be better off chasing dragons.
Leaked Nvidia RTX 3090 benchmark score shows performance up to 26% faster than RTX 2080 Ti
- Thread starter mongeese
- Start date
MaxSmarties
Posts: 559 +326
Don't you know ? Nowadays if you don't reach 4K/120 you are no-one...
Adi6293
Posts: 931 +1,309
I'm married and I don't have to ask how much I can spend on my hobby, I made sure that my then fiancé knew now to get involved with it, we have separate banks and each pay for separate things this way we don't have to "argue" about money
VitalyT
Posts: 6,845 +8,136
And then?
kmo911
Posts: 352 +44
Yepp 1080 ti are still tha best single core. I lov 1 it. FE 1080 ti 11 gb. but the future tells it would be heavyer to run thos 4k-16 textures. we are at 8k gaming and cad 8k. now a good ati nvidia intel can run low-uhd as before you needed a upd cpu much ram and a good speedy optical line to get those big games downloaded. now playing at lga 1200 and 1080 ti fe. remember those 2080 ti are just 2x 1080 ti cards together. ati did it wit 2x gpus. sli and x sli amd red 2 western game needs 170 gb so tink about how lon you must sppend downloading it. and with more textures you gonna needing a faster pc to take down all info and a I got it pack sendt up. using adsl vdsl 1 2 for those games and big pak files. quake 3 4 5 pak files was small. but future games are up to 30 gb pak files and comes in 1-40 gb s.
you can now save whole games to youre 1 tb-8 tb drives. and hope you not needing downloading tose again ever. but window 10 older os bsod makes it hard to be safe. if win 10 could be a rom file it would always be in fit. other files to get youre gtx 9xx-1080 to to run into the future would be hard too futurproof it. if you play older games you can use a low end 1mb-32 64-128 mb vram gpu. if games needing more then that I would just tell you that. 1gb-4ggggg ok old games gta 1 2 3 some how 4 5 but needing that 4-6 gb vram and a lots of ram from 2-8 gb ram. more heavy textures 8-11 12 gb. heavyer textures like 4k-16k would be needing heavy duty 16-24 48 64 gb v ram (future) but to exspensive 1 st months. so you waiting 4 mont ..n ope 8 monts still same price getting that lovely one gtx rtx amd intel nvidia. and when you getting a new gpu cpu more ram thay you must use a whole year to save up to a 30xx getting more pricy then 2x or 4x 20xx ti gpu. then console comes wit low ram 4-6 gb and still running games better then a pc with gtx 900-1080 series.
if they could have set price lower and selling a lot of gpus it would be totable to get a better gpu before its getting older.
ask you're self getting new 2xxx or just wait to 2xxx ti getting low price.
it would taking ages.
crysis 1 2 gpu with xxx gbps and c 3-4 needing better ram.
it would never end.
you can now save whole games to youre 1 tb-8 tb drives. and hope you not needing downloading tose again ever. but window 10 older os bsod makes it hard to be safe. if win 10 could be a rom file it would always be in fit. other files to get youre gtx 9xx-1080 to to run into the future would be hard too futurproof it. if you play older games you can use a low end 1mb-32 64-128 mb vram gpu. if games needing more then that I would just tell you that. 1gb-4ggggg ok old games gta 1 2 3 some how 4 5 but needing that 4-6 gb vram and a lots of ram from 2-8 gb ram. more heavy textures 8-11 12 gb. heavyer textures like 4k-16k would be needing heavy duty 16-24 48 64 gb v ram (future) but to exspensive 1 st months. so you waiting 4 mont ..n ope 8 monts still same price getting that lovely one gtx rtx amd intel nvidia. and when you getting a new gpu cpu more ram thay you must use a whole year to save up to a 30xx getting more pricy then 2x or 4x 20xx ti gpu. then console comes wit low ram 4-6 gb and still running games better then a pc with gtx 900-1080 series.
if they could have set price lower and selling a lot of gpus it would be totable to get a better gpu before its getting older.
ask you're self getting new 2xxx or just wait to 2xxx ti getting low price.
it would taking ages.
crysis 1 2 gpu with xxx gbps and c 3-4 needing better ram.
it would never end.
Last edited:
If this is the 3090 (or 3080ti) even allowing for MUCH better ray tracing, this is objectively disappointing and opens a massive door for AMD. I had though Nvidia would remain 20% fps better than AMD Comparing top cards ... If this is accurate, I think at best they will be about tied, at worst, Nvidia may take the crown.
That 26% isn't in raytracing though, is it? Raytracing could be twice as fast for all we know.
Ha. My limit is probably $200. Seems like things aren't moving fast in the under $200 category.
I'm amazed that anyone would pay that much for a second hand product, especially when the next model is pretty near at hand.
Let's assume the score is completely genuine and it was generated using a consumer-grade Ampere model of some kind. We obviously have no idea of the system platform used, but take a look at one of Kingpin's top single Titan RTX results to one I've just done on my system (RTX 2080 Super):
www.3dmark.com
3DMark reports the memory clock rates, not the adjusted data rates, so with mine at 2050 MHz, that's equivalent to 8200 MHz (likewise with Kingpin's card running at 8676 MHz). In both cases, the reported core clocks are the Boost frequencies (2070 MHz for me, 2550 MHz).
Note how Kingpin's Graphics Test is more than double mine? 20969 vs a paltry 9408. The alleged Ampere result is 18257, a little under double mine. So even if this is the absolute top-end, best-of-the-best version of consumer Ampere, it's almost twice as good as a 2080 Super and about the same as a heavily overclocked Titan RTX.
The former sounds a lot more impressive than the latter, but once you go over the required shader throughput rate with Time Spy, it becomes far more ROP and bandwidth limited.
Result
www.3dmark.com
3DMark reports the memory clock rates, not the adjusted data rates, so with mine at 2050 MHz, that's equivalent to 8200 MHz (likewise with Kingpin's card running at 8676 MHz). In both cases, the reported core clocks are the Boost frequencies (2070 MHz for me, 2550 MHz).
Note how Kingpin's Graphics Test is more than double mine? 20969 vs a paltry 9408. The alleged Ampere result is 18257, a little under double mine. So even if this is the absolute top-end, best-of-the-best version of consumer Ampere, it's almost twice as good as a 2080 Super and about the same as a heavily overclocked Titan RTX.
The former sounds a lot more impressive than the latter, but once you go over the required shader throughput rate with Time Spy, it becomes far more ROP and bandwidth limited.
Evernessince
Posts: 5,469 +6,158
Performance per watt isn't some arbitrary designation like "20 times the ray tracing performance (over non RTX cards)". Marketing trick? How does one fake performance per watt? I have never seen Nvidia or AMD fake PPW.
1080ti SC2 ... At the time thought I was crazy for spending so much, now, so glad I did, it's just held up SO well, even for 4k/60 gaming.
JimboJoneson
Posts: 310 +507
I believe the 2080ti is only 31% faster than 1080ti, so if this is 3080ti, this improvement is roughly in line with previous precedence. If this is new Titan, then that seems not very spectacular (considering precedence).
I think AMD will be much more competitive with RDNA2, but I don't think they'll be toppling NVidia ... as long as they can get near Nvidia in the top end and provide good value in the mid and mid-high tiers, that's really all that matters for 99% of people. 1% of people are very noisy as well. That said, I don't have grand expectations, but I do expect them to close the gap some.
I just don't want to see the prices continuously going up and up ... its a bit crazy.
Last edited:
Evernessince
Posts: 5,469 +6,158
Both maxwell and pascal were more then double the jump the 2080 Ti gave. By all regards turing was a major disappointment.
yup I've said this before buying the 1080ti was a solid investment, and its the gpu that keeps on giving.
JimboJoneson
Posts: 310 +507
Well it wasn't more than double and certainly not between Maxwell vs prev. Yeah, the jump to 1080ti was impressive. No argument there.
Looking at AMD's recent gen over gen stuff, I don't know, maybe 25-35% is the new normal for what we can expect? I think the 1080ti was a bit of an outlier overall.
But this is all just rumour until we see reviews.
Evernessince
Posts: 5,469 +6,158
The 20xx series could have squeezed out more performance if it wasn't dedicating resources to tensor and turing cores. Same thing could have been said for vega, where the HBCC (which is only useful for professional applications) took up a good chunk of die space. That was clearly never a gaming card to begin with.
If you ask me, AMD actually releasing a high end card on a new architecture would be nice. When was they last time they did that where the card was actually designed for gamers? The R9 290? That wasn't even a new architecture. Fury and Fury X were essentially baby vega, not really a gaming card (plus still GCN). It's kind of sad.
JimboJoneson
Posts: 310 +507
Agreed on the specialized hardware taking up resources ... Let's see what RDNA2 will bring.
What might be cool is if AMD (or NVidia for that matter) could use the same cores for both render and RT, and let the user set the balance between the two -- that would be amazing.
krizby
Posts: 429 +286
Let's assume the score is completely genuine and it was generated using a consumer-grade Ampere model of some kind. We obviously have no idea of the system platform used, but take a look at one of Kingpin's top single Titan RTX results to one I've just done on my system (RTX 2080 Super):
Result
3DMark reports the memory clock rates, not the adjusted data rates, so with mine at 2050 MHz, that's equivalent to 8200 MHz (likewise with Kingpin's card running at 8676 MHz). In both cases, the reported core clocks are the Boost frequencies (2070 MHz for me, 2550 MHz).
Note how Kingpin's Graphics Test is more than double mine? 20969 vs a paltry 9408. The alleged Ampere result is 18257, a little under double mine. So even if this is the absolute top-end, best-of-the-best version of consumer Ampere, it's almost twice as good as a 2080 Super and about the same as a heavily overclocked Titan RTX.
The former sounds a lot more impressive than the latter, but once you go over the required shader throughput rate with Time Spy, it becomes far more ROP and bandwidth limited.
Pardon me but somehow your system is getting very low score in Timespy. A 2080 Super should get around 11k5 in graphic in Timespy stock, not 9k5. Even my 2060 Super could get 9k9 Graphic
And you shouldn't get variable scores that change with each reset either, perhaps you should look into it.
I'd accidently left a frame rate limiter on. Here's another two runs of mine (standard and overclocked GPU), without the limiter this time:
www.3dmark.com
Taking the higher one, with a graphics score of 12471, the alleged Ampere result (18257) is still 46% better. Of course, without any information about how genuine the score is and what platform it was tested on, it's not possible to really judge how good an improvement this is.
Result
www.3dmark.com
Taking the higher one, with a graphics score of 12471, the alleged Ampere result (18257) is still 46% better. Of course, without any information about how genuine the score is and what platform it was tested on, it's not possible to really judge how good an improvement this is.
krizby
Posts: 429 +286
I'd accidently left a frame rate limiter on. Here's another two runs of mine (standard and overclocked GPU), without the limiter this time:
Result
Taking the higher one, with a graphics score of 12471, the alleged Ampere result (18257) is still 46% better. Of course, without any information about how genuine the score is and what platform it was tested on, it's not possible to really judge how good an improvement this is.
If you compare stock vs stock then it's a 57% uplift in performance. Given how confident Nvidia is about these GA102 chips I suspect this 18257 score belongs to the alleged RTX 3080. If the 3080 is around 30% faster than the 2080 Ti (stock 2080 Ti score is 13k6), Nvidia can comfortably charge 800-900usd for these GPU while Big Navi is around 10% slower and cost just 650-700usd
Last edited:
Well let's assume it is completely genuine and also assume that Time Spy is completely compute/shader bound (it's not, but you have to start somewhere). The purported Ampere is clocked to 1935 MHz, whereas my overclocked 2080 Super averaged a core frequency of 2030 MHz during the tests - so the new chip was 4.7% lower, but achieved a graphics score 46% better.
So clock-for-clock, it's roughly 50% better. The TU104 in the 2080 Super has 6 GPCs, each with 4 TPCs, and each of those with 2 SMs, giving a total of 3072 CUDA cores. Sticking with the compute bound assumption, the Ampere chip would need to have 4608 cores to reach that 50% improvement - the same number as in the TU102.
I've taken an educated guess at the full consumer GA102 being a 6 to 8 GPC, retaining the 6:1 TPC ratio (same as Turing, same as the GA100). This would put the CUDA core count from 4608 up to 6144 - like this:
6 GPCs = 4608 CUDA cores > '3080'
7 GPCs = 5376 CUDA cores > '3080 Ti?'
8 GPCs = 6144 CUDA cores > '3090 / Titan A?'
My 2080 Super is roughly 18% better in Time Spy (even accounting for clock differences) than my previous Titan X Pascal, so if this is a 3080, then it would follow a similar pattern of generational increments - in that the xx80 model is as good or better than the previous top-end version.
Given that the GA100 is only 1.3% larger (die area) than the GV100 it has replaced, Nvidia/TSMC are clearly happy to be churning out very large chips again, so there's a good chance that a full GA102 is an 8 GPC chip.
krizby
Posts: 429 +286
Comparing 2080 Ti vs 1080 Ti spec:
Transistors: 18.6 vs 11.8 (billions)
Boost Clock: 1545mhz vs 1582mhz
FP32: 13.45 vs 11.34 (TFLOPS at normal boost clock) ~ 19% increase
Bandwidth: 616 vs 484.4 (GB/s) ~ 27% increase
Timespy graphic: 13610 vs 9521 ~ 43% increase (data from Guru3d)
So with only 19% and 27% more FP32 throughput and bandwidth, Turing has an additional ~20% IPC gain from improved cache and concurrent INT32+FP32 SM. With Ampere SM promised to bring even higher IPC than Turing (Nvidia said Ampere SM has twice the performance of Turing SM but who knows), I'm pretty sure with IPC gain alone a RTX 3080 can be 20-30% faster than 2080 Ti when they both have the same number of SMs and clock.
Last edited:
The cache improvement plays a big part in this - Turing has significantly more L1 and L2 cache than Pascal, 64+32 kB and 6 MB versus 48 kB and 3 MB, respectively. And the L1 cache bandwidth is up to double that in Pascal.
The SMs in the GA100 have 192 kB, although I don't expect the GA102 to be the same, but it retains the same L1/shared memory structure of Turing. Compute data is now compressed, and the L2 bandwidth is monstrously high - something like 3 times that of Volta. Again, I don't think the GA102 will follow suit and have anything like the L2 amount and partitioning seen in the GA100, but the compression and bandwidth improvements should come through.
It will be interesting to see just how different the consumer version of Ampere really is.
Similar threads
- Locked
