-
Ever wanted an RSS feed of all your favorite gaming news sites? Go check out our new Gaming Headlines feed! Read more about it here.
What happen to actual Nvidia graphics card upgrades?
- Thread starter Deleted member 40102
- Start date
It isn't.
For me, HDMI 2.1 VRR support and OpenGL performance matters, so an AMD card has poor value for me.
Sure, I could get more frames in some games with an RX 5700, but do I really want to deal with screen tearing and judder in 2019? No.
With the RTX and tensor hardware using such a small percentage of the die.. it wouldn't be a significant benefit. Few people are going to pay a significant amount of extra money just for ray tracing before software is available. Essentially what you are saying is that RTX shouldn't have happened at all. This stuff has to start somewhere.
You're not wrong, but just like the GeForce 3 cards, I feel like waiting for the next generation is a good idea. I want the card that is to ray tracing what the Radeon 9700 Pro was to shaders. Until then my 980 Ti will do.
Well, they're competing against non RT parts that offer better dollar to framerate metrics (but missing those RT features) anyway. Nvidia only put out a GTX based on this chipset at the low end. They could have done one at the high end and priced it competitively against AMDs cards.
Personally, I'm not sure just how viable RT is on the lower RTX cards. It certainly able to hit decent performance and IQ on the 2080 Ti, but below that I'm not necessarily convinced. Those of who bought the RTX 2080 Ti would have bought it anyways. Perhaps you'd have lost 2080 and 2070 buyers though. But as I mentioned, they're competition with non RT parts already, and people who aren't yet convinced RT is worth paying more money for are going to look outside RTX either way.
I don't think RTX doesn't happen. Ray tracing is very much the near future of rendering. It is and was inevitable. Fully rasterized games only offer so much, and only go so far. As we get the hardware that can handle RT, it's going to take off. Control with RTX is clearly closer to what Remedy wanted to do than without it. It's an astoundingly good looking game, that hints at the type of dynamism we might see in larger environments as we move into the next generation.
Just as people who think Nvidia might have killed RT by over pricing these cards... It's not going to matter in the long run what they were priced. Nvidia are going to still be around in a few years and ray tracing will become common place before you know it, no matter which vendors are making GPUs come the middle of next decade.
As someone with a 980 ti in one system and an RTX 2070 in another system, you'll be just fine with your 980 ti.
Removing RTX would free up 8-10% die space and performance isn't going to scale linearly. You would wind up with a card so close to the regular 2080 Ti that it'd be within overclock range, but lack ray tracing.
I'm not sure there's a place for that at the high end in terms of product segmentation, plus it would undercut ray tracing development as has been mentioned a number of times. This is the future of rendering. Nvidia can't dump all the R&D burden onto a few cards that nobody buys.
And it would still be a MASSIVE die.
I upgraded to a 1080Ti in early 2018 and I still don't have an upgrade path that's actually worth the money. Absurdly good card.
Well my whole thought process was that the card would be the same as the 2080 Ti in terms of rasterization performance, but offered at a cheaper price.
I understand where you're coming from, I'm just pointing out that the AMD cards exist no matter what Nvidia do. That cheaper card that offers equivalent rasterization performance exists now, and all I'm saying is if Nvidia had put one out, they'd get more of that money that's currently going to AMD, and ray tracing isn't going to fail eitherway.
HDMI 2.1 VRR support is coming to AMD cards as well.
Pretty sure nobody is even offering VRR right now?
Between my Freesync monitor and Enhanced Sync I'm not dealing with screen tearing or judder, so I don't know what that remark was for?
Sadly, I can't use future features right now. There isn't even an ETA for it.
The remark was for this:
"But Nvidia is the bigger brand name that a lot of folks are just going to keep flocking to, regardless of how much they get bent over."
I just gave you two reasons why I prefer an Nvidia card.
Can I use freesync with my LG C9 and an AMD card? No, so it's useless to me. It has nothing to do with "getting bent over".
Sorry, but I don't follow?
You prefer an Nvidia Card for VRR, which you admitted is a future feature?
What's useless about the AMD card that the Nvidia Card is offering for your TV?
I was referring to price/performance. It's a thing whether you want to admit it or not.
I have no problem admitting it, it's a fact. The problem is this part here:
It's just plain wrong. Price-performance ratio is a part of value, but not all of it.
For AMD, it isn't even close to be released. For Nvidia, it is available right now.
If I had bought an AMD card, I'd have two options:
1 - play with traditional v-sync and waste the extra power from the card.
2 - play with judder and/or tearing.
And you are telling me that's the best value, lol.
Last edited:
No, no we have not, not by a country mile. And, more to the point, employing the tricks currently used to fake lighting/shadows/reflections constitutes a very significant amount of development time. Hence a huge reason RT is the future: it will drastically reduce the work load for developers, while achieving superior results.
Wasting more than half of the die size for ray tracing crap instead of raster performance is the stupidest thing they could do
Again, where are people getting these crazy estimations of wasted die size?
There is a ton of misinformation being handed out on this front.
The Nvidia Turing GPUs already support VRR via HDMI, and even provide 4k/120hz via HDMI now too.
I expect 2020's new line of cards to be a pretty significant improvement in performance, plus with the really nice added bonuses from HDMI 2.1 support
However without proper competition I fully expect them to overprice them and try to pull another trick with the naming convention to mask that
However without proper competition I fully expect them to overprice them and try to pull another trick with the naming convention to mask that
Same here, want to upgrade, but the cost of doing it isn't justified with the current lineup of cards.
At the unveiling nvidia showed a die where half of it was only rt cores and more soace wasted for tensor cores, regardless of how accurate it was just look at how enormous the die size for turing is while still having more or less the same number of stream processors/tmu/rops
That's awesome. I didn't know that Nvidia already had it, until this thread.
So you does that mean you don't need Gsync anymore?
There was no useful data in the overview they initially provided. Initially there was a lot of misinformation that people have hug onto. It turns out that most of the additional space you're describing comes from cache and increased instruction set.
Eventually someone was able to get their hands on an actual image of TU106 and TU116 dies several months ago. It was reverse "engineered" to discern exactly how much of the die space was RTX and tensor.. 8-10%. I saw it in multiple places months back but at the moment can only find an analysis that is uncontested on Reddit.
Not all of the die space is used, so the percentage of used space is larger but my math seems ~18% for TU106 but it's possible I'm missing something. I've seen others claim 22%.
Source:
Nope. https://www.techpowerup.com/254452/...ses-tpc-area-by-22-compared-to-non-rtx-turing
So even without any raytracing hardware Turing would be much bigger than Pascal.
The 22% number is as a fraction of the die space used by the cores.
The 10% number is as a fraction of the total die space.
The total number is a lot smaller because a significant amount of the die is devoted to caches and interconnect.
Ok my bad but still... that cache doesn't seem to do much and more instruction sets... well, let's just say i and i think many others would have appreciated much more either a cheaper video card without the rt baggage or a full fat die of rasterizing performance.
But nvidia gotta nvidia so we have to hope on AMD to deliver (lol they are on a more advanced node and still can't grasp the power or performance per watt of nvidia) or Intel to breakthrough immediately.
Yeah HDMI VRR output on Turing removes the need for G-Sync on the C9. It does the same thing. But the C9 is receiving a G-Sync update so it can also support pre-Turing cards.
You wanted nothing changed, just a bigger Pascal die. Ok, got it.
Regardless, your comments are now getting into criticism of the Pascal -> Turing architecture changes without the knowledge to have the discussion. This isn't a criticism of your knowledge, as I am in that same boat. I just know enough than to avoid that particular conversation.
Except people go gaga whenever they see raytracing they can actually play (Minecraft PTGI videos). And they get sour grapes when it costs $1000.
The difference is clear. Everyone sees it. Raytracing is absolutely critical and inevitable, and when more people can afford it they'll stop acting like children.
NVIDIA shouldn't be pricing their cards like they are. Though the reason there is not a huge difference is counterintuitive. NV ain't trying to scam you on the chip. It's as good as it can be.
The issue is that we've hit the zone where engineering gets pretty tough. We're talking Quantum Effects, quantum entanglement, electrons doing weird as shit things. This is engineering at scales we've never engineered before. It's going to slow down, and it's going to slow down massively. That's why NV is introducing ray tracing. They're hoping it's something drastic they can improve year over year. Because the rest of the chip? That's not happening anymore. Period. Physics says no.
The issue is that we've hit the zone where engineering gets pretty tough. We're talking Quantum Effects, quantum entanglement, electrons doing weird as shit things. This is engineering at scales we've never engineered before. It's going to slow down, and it's going to slow down massively. That's why NV is introducing ray tracing. They're hoping it's something drastic they can improve year over year. Because the rest of the chip? That's not happening anymore. Period. Physics says no.
nvidia is still on 12nm (16nm). they're jumping to 7nm next year
That is precisely my point. NV is still at 12nm because jumps are getting harder and harder to make. And even after making the jump, there are further and further diminishing returns. And once we jump to 7nm, that's it for awhile. The returns we used to get every year are gone. Back then folks were upgrading their process size constantly, and that's why it was so much performance for each upgrade.
More upsetting is there are no value upgrades to be made... My 580 is just a slightly overclocked 480... and there is nothing in the 200-300 range that is worth it. The 2060 Super is almost there, but still quite a bit pricey for about 50% improvement over the 580 (with RT mostly not being worth it as a feature since it's got so few tensor cores)
Right, I had posted the same data. It turns out I was just looking at it differently.
The percent of the TPC that is RTX is ~18%. That's what I was stuck on.
The percent increase in TPC size due to RTX is ~22%.
Of course both are greater than looking at it from a die perspective, but in the end it's just different ways of viewing the same thing. To me it makes more sense to consider what percentage of the TPC and die are used for RTX (ray tracing and tensor).
Freaking THANK YOU. Nvidia literally took one for the team and the reaction to the RTX 20 generation is incessantly parroted as a move done due to lack of competition. Can you believe this madness? "Lazy Nvidia" would've said screw the RT noise and just release okay cards at not good prices. Instead they sunk crazy R&D into RT, ate a an ocean of bad PR because they released when their was no software support, oh yeah and shifted the entire freaking industry for the foreseeable future. Now we have AMD, Sony, and Microsoft on board, meaning ray tracing is completely legitimized and here to stay, likely forever.
But Nvidia gets no props for this because we didn't get to upgrade this year...
The gamer's inability to think past its nose irritates the fuck out of me in cases like this.
Well if anything… AMD is to blame here. Nvidia is the market leader and people go and willingly choose their graphics cards. The marketplace has voted with their wallets and NVIDIA won so far. Its AMDs job to put pressure on NVIDIA by releasing new graphics cards at a faster pace. I want to upgrade my GTX 760 but only if I can get a card for 200 USD that can grant me at least GTX 1080 performance.
6 and a half years after the GTX 760 release (a mid range card by the way) and the lowest end gaming card you can have is a RX 560 for around 120 USD a card that just trades blows with the 760.
I am really looking forward for the RX 5500. Why the interest for a low end card? We need the new low end to be at least at RX 590 level so that tech at the high end follows suit. Polaris tech is more than 3 years old by now and those GPUs still sell like hotcakes. I understand that we had the mining craze and also lets be honest: most gamers do not NEED more graphics performance than a RX 570 to play modern games at 1080p. Pop one of those bad boys on an old PC and you can game just fine, at least until the new consoles are released.
Last but not least, I would like NVIDIA and AMD to find a workaround for dual GPU setups. This gen was just terrible at that. I loved to see those enthusiast rigs with 2 way 3 way SLI configs driving multi monitor configs at high framerates. As modern (NVIDIA) GPUs have low power consumption, a 2 way SLI could be a viable upgrade path if they solve the inherent issues with the technology.
6 and a half years after the GTX 760 release (a mid range card by the way) and the lowest end gaming card you can have is a RX 560 for around 120 USD a card that just trades blows with the 760.
I am really looking forward for the RX 5500. Why the interest for a low end card? We need the new low end to be at least at RX 590 level so that tech at the high end follows suit. Polaris tech is more than 3 years old by now and those GPUs still sell like hotcakes. I understand that we had the mining craze and also lets be honest: most gamers do not NEED more graphics performance than a RX 570 to play modern games at 1080p. Pop one of those bad boys on an old PC and you can game just fine, at least until the new consoles are released.
Last but not least, I would like NVIDIA and AMD to find a workaround for dual GPU setups. This gen was just terrible at that. I loved to see those enthusiast rigs with 2 way 3 way SLI configs driving multi monitor configs at high framerates. As modern (NVIDIA) GPUs have low power consumption, a 2 way SLI could be a viable upgrade path if they solve the inherent issues with the technology.
They released the 1660Ti. Why not offer the full spectrum non-RTX cards as a complement and let consumers decide whether they value ray tracing or not? At some point, it has to happen. That can even be now. But there are ways to make the transition less bumpy.
What you describe is a good way to stall adoption. This is a well understood princple of the technological adoption curve. This is actually very bad for developers to do. When you start offering less advanced, cheaper technical solutions during the infancy of a new technology's life, it halts their adoption. Adoption happening over time isn't an accident or forgone conclusion. It's actually a cycle - developers need an audience with hardware to start pumping out content, and audiences need content to buy the hardware. If nobody has the hardware, developers won't pump out the content to support it, as this is a rendering process very far removed from conventional rasterization.
It's expensive right now, because it's new, because only enthusiasts will pick it up currently, with the expectation that that audience will grow. Offering those same enthusiasts a non-RT option stalls the adoption.
Don't put words into my mouth and don't go around assuming things.
I said that given the results of how Turing turned out on the bigger chips i would have preferred a bigger Pascal chip at the same price (or a Turing without the baggage), that doesn't mean i don't want nothing changed.
Second: more cache is always good, reduced latency by keeping more stuff on the on die memory rather than sending it outside but we have not seen that many improvements from it so far; last but not the least the wider instruction set are good if you do compute and ML stuff aka more Tensor/RT baggage when talking pure "traditional" gaming performance because they just added native low precision integer pipes rather than having it reserved for more FP32.
maybe i'm in the minority but i really enjoy what ray-tracing brings to the table. That said for me (i have a 1070 gtx) the new GPU don't yet offer enough performance to wander a buy. I'd really enjoy ray tracing in strategy games like total war and battlefleet gothic.
How am I putting words in your mouth when I simply restated your position? You want a big Pascal.
Regarding the second part, you had no clue how much of the die was "baggage" yet you do feel knowledgeable enough to critique the evolution of their architecture.
At what color bit? HDMI 2.0 doesn't have the necessary bandwidth for this. If the above statement is factually correct, how are people doing this? And how much of a hit does IQ take?
All of these cards are kind of crappy investments for this very reason. You want full 4k@120hz with VRR, we need to wait for cards that ship with full 2.1 silicon.