ada lovelace NVIDIA Project Beyond GTC Keynote with CEO Jensen Huang: RTX 4090 + RTX 4080 Revealed

[leadeater]

29 minutes ago, Dr_Whom said:

Just want to point out that nVidia in fact marketed them as gaming cards. From nVidia's own website and description of the card:

True and a lot of reviewers and tech media criticized them for that too. Since the review material and product guidance for them were semi-pro/pro applications yet they also did stuff like that. Especially since Nvidia themselves complained to these people about putting them in gaming systems and put gaming focus on them in their videos and builds. Something Jay complained about more than once in videos talking about Nvidia and Titan specifically, since he was one that copped complaints from Nvidia about it.

 

Nvidia at it's finest lol

 

Titan name made no sense because of how badly Nvidia executed on it, but oh well it's not like it matters. Titan name is dead and I don't see it coming back, not soon or maybe ever.

 

P.S. Titan Black is slower at gaming than GTX 780 Ti making it factually not the "Ultimate gaming GPU" 

[Dracarris]

2 hours ago, ewitte said:

There are some benefits to small places that aren't hot.  So many here have 2000,3000,4000+ square feet houses here and we had many days over 100F last summer.  AC is most of the electric bill.  I hear a bunch of people complaining about $500+ for just one month. Before solar I did have a few months hit $300 but the prices were lower.  Highest this summer was $80.

Well this summer we also had a lot of days and weeks above 100 Freedom Units. We do not have ACs at residential buildings (but properly insulated buildings).

[BiG StroOnZ]

2 hours ago, leadeater said:

P.S. Titan Black is slower at gaming than GTX 780 Ti making it factually not the "Ultimate gaming GPU" 

 

Titan Black is the same config as the 780 Ti except with 3GB more VRAM (Titan Black also has slightly higher boost clocks). In nearly all benchmarks the Titan Black comes out ahead. You're probably thinking about the OG Titan. 

[MageTank]

3 minutes ago, BiG StroOnZ said:

 

Titan Black is the same config as the 780 Ti except with 3GB more VRAM. In nearly all benchmarks the Titan Black comes out ahead. You're probably thinking about the OG Titan. 

The Titan Black wasn't really "faster" at gaming compared to the GTX 780 Ti. GTX 780 Ti boosted higher and could slightly outperform it. I actually still have these cards (love their heatsink design) but yeah, the Titan Black wasn't exactly worth the investment.

 

It is very important to remember that the Titan Black cards were all reference design only (with one exception for Gigabyte). Non-reference 780 Ti's had better power delivery, frequency-adjusted VBIOS's and better thermals for higher boosting. Reference against reference, the Titan Black should win, but against a non-reference 780 Ti, wouldn't be a fair fight. Here is a fun little article showing what happened when you slapped the Gigabyte Windforce cooler on the Titan Black: https://www.tomshardware.com/reviews/gigabyte-geforce-gtx-titan-black-ghz-edition,3821-8.html. 

 

That said, there was one game that the Titan Black utterly destroyed the GTX 780 Ti in which was GRID 2, that game was pretty unique though. Supported AVX2 over SSE and people often complained about how hot that game ran, despite it efficiently using hardware, lol.

 

Man you people have me really missing Kepler right now. Remember when overclocking GPUs was fun?

[BiG StroOnZ]

24 minutes ago, MageTank said:

The Titan Black wasn't really "faster" at gaming compared to the GTX 780 Ti. GTX 780 Ti boosted higher and could slightly outperform it. I actually still have these cards (love their heatsink design) but yeah, the Titan Black wasn't exactly worth the investment.

 

It is very important to remember that the Titan Black cards were all reference design only (with one exception for Gigabyte). Non-reference 780 Ti's had better power delivery, frequency-adjusted VBIOS's and better thermals for higher boosting. Reference against reference, the Titan Black should win, but against a non-reference 780 Ti, wouldn't be a fair fight. Here is a fun little article showing what happened when you slapped the Gigabyte Windforce cooler on the Titan Black: https://www.tomshardware.com/reviews/gigabyte-geforce-gtx-titan-black-ghz-edition,3821-8.html. 

 

That said, there was one game that the Titan Black utterly destroyed the GTX 780 Ti in which was GRID 2, that game was pretty unique though. Supported AVX2 over SSE and people often complained about how hot that game ran, despite it efficiently using hardware, lol.

 

Man you people have me really missing Kepler right now. Remember when overclocking GPUs was fun?

 

Yes, it was ever so slightly faster than a reference 780 Ti, which is what I was basing my comment on (should have specified):

 

Spoiler

 

 

 

 

''

 

 

 

 

 

 

https://www.guru3d.com/articles_pages/nvidia_geforce_gtx_titan_black_review,1.html

 

I'm not necessarily saying that it was a worthy investment, but there were situations where at higher resolutions the 3GB VRAM on the 780 Ti could become a bottleneck. For sure people who had the Titan and Titan Black were able to hold onto their cards longer than those with a 780 and 780 Ti.

 

I do remember the Windforce cooler for the Titan Black and that article, EVGA also had an ACX Cooler for the Titan Black:

 

https://asia.evga.com/products/product.aspx?pn=100-FS-3790-B9

 

There were definitely faster non-reference 780 Ti's, I agree, but if you wanted to play at higher resolutions the Titan Black made it much more possible at the time with its 6GB frame buffer. 

 

Yeah Kepler was great for overclocking, I had an EVGA GTX 780 SC. Wish I got the 6GB variant though, now that I look back on it. Even Maxwell was great at overclocking, remember the gains people were getting with the 980 Ti?

[MageTank]

18 minutes ago, BiG StroOnZ said:

Yeah Kepler was great for overclocking, I had an EVGA GTX 780 SC. Wish I got the 6GB variant though, now that I look back on it. Even Maxwell was great at overclocking, remember the gains people were getting with the 980 Ti?

I still have an MSI GTX 980 Ti with a massive hunk of copper for the cooler: https://www.msi.com/Graphics-Card/GTX-980-Ti-GAMING-6G-GOLDEN-EDITION/Gallery

Spoiler

This thing overclocked really well too, I just preferred Kepler since it actually scaled really well with voltage and you can do some crazy stuff with volt mods. Maxwell seemed to be Nvidia's shift towards a less robust hardware design and more focus on better precision FP performance and technologies that improved performance without needing massive hardware behind it. God forbid we bring up that hardware scheduler nonsense, lol.

 

Part of why I am excited for the RTX 4090 is because of all the talk around power consumption and cooling it. The harder it is to cool, the more fun I'll have playing with it. If this is the next generation of Fermi, I am ready to receive, lol.

[leadeater]

2 hours ago, BiG StroOnZ said:

 

Titan Black is the same config as the 780 Ti except with 3GB more VRAM (Titan Black also has slightly higher boost clocks). In nearly all benchmarks the Titan Black comes out ahead. You're probably thinking about the OG Titan. 

No I checked Titan Black, it's 2%-5% slower than the 780 Ti. It should be faster on spec paper but it's not, between different vram chips being used and different coolers the AIB 780 Ti's were faster. Edit: You have to remember that there were a lot of AIB cards factory OC to over 1000 MHz, often mid points between 1000-1100 and the Titan Black simply did not have this commonly.

 

Like today the Titan Black would now be faster simply due to more vram and the 780 Ti choking but back then wasn't an issue.

[BiG StroOnZ]

18 minutes ago, leadeater said:

No I checked Titan Black, it's 2%-5% slower than the 780 Ti. It should be faster on spec paper but it's not, between different vram chips being used and different coolers the AIB 780 Ti's were faster.

 

Like today the Titan Black would now be faster simply due to more vram and the 780 Ti chocking but back then wasn't an issue.

 

Not sure where you were looking, but the Titan Black consistently comes out ahead of the 780 Ti (reference vs reference);

 

 

 

 

 

''

 

 

 

 

 

 

https://www.guru3d.com/articles_pages/nvidia_geforce_gtx_titan_black_review,1.html

 

There are more reviews which show the same results, which I can post if you are still not sure (reference vs reference of course). Some AIB 780 Ti's were definitely faster, but reference vs reference the Titan Black is faster nearly all the time.

 

Can you show me an instance where the Titan Black is 2-5% slower than a 780 Ti? There aren't really many reviews of the Titan Black especially those that compare it to a non-reference 780 Ti. 

 

Anyway, yeah, today it would definitely excel due to VRAM. But there were many high resolution gamers at the time who simply were bottlenecked by the 3GB VRAM on the 780 Ti and as a result had to opt for the Titan Black.

[MageTank]

5 minutes ago, BiG StroOnZ said:

Can you show me an instance where the Titan Black is 2-5% slower than a 780 Ti? There aren't really many reviews of the Titan Black especially those that compare it to a non-reference 780 Ti.

The one I linked from Tomshardware did, though it also certainly favors your statement: https://www.tomshardware.com/reviews/gigabyte-geforce-gtx-titan-black-ghz-edition,3821-8.html

Spoiler

 

Reference 780 Ti averaged 125.9 FPS across their gaming test suite, while the reference Titan Black OC averaged 131.3. Now the Titan Black OC did have a factory OC, but I don't really count this much as Kepler was extremely thermal limited after 80C and these cards almost always stayed pegged at 80C, offsetting most OC's anyways. Still, that difference between the two was nearly 6fps, or at that framerate, 4.19% difference in favor of the Titan Black OC at 1080p.

 

If you compare the Windforce versions of both cards when OC'd, the 780 Ti actually came out ahead by 3%, which made sense to me. GK110B is GK110B after all, and having more VRAM than you actually need doesn't really do much for you in terms of performance, but having more power budget for core frequency did, if only by a little.

 

I personally wouldn't consider either card "faster" than one another. I've tested both on water blocks and I can say that when cooling is not a concern, both perform nearly identical. Even if we compare 4K performance, the extra VRAM did nothing when you still met the same ROP and bus limits imposed by both cards. Again, the only game I can recall there being a huge performance difference in was GRID 2, but that was a very niche situation.

[leadeater]

18 minutes ago, BiG StroOnZ said:

Not sure where you were looking, but the Titan Black consistently comes out ahead of the 780 Ti (reference vs reference);

Well yes reference vs refence but is that actually what is being purchased? 

 

780 Ti Lightning for example is 13% & 14% higher base and boost respectively.

[Quackers101]

seems like the 4000 GPUs have 10x more L2 cache compared to RTX 2000-3000 gpus.

but equal to slower in other things, a big focus on the lower end 4080. they really want to claim that its a big leap in transisters on a smaller die area, like 1.5-2x up to 3x more, than previous RTX cards. also having 8th gen NVENC, and 3000's 5th gen NVDEC. All top cards to have dual NVENC and I guess AV1 support.

 

geforce shadowplay might start to support 8K/60 streams/recordings? to 4 of 4K/60 video editing. ouch.

Also that the reordering unit offers 2x perf for raytracing, I guess similar that intel might have seen, unsure about AMD.

also a bit sus from nvidia

Quote

Nvidia innovations have democratized streaming so that more people can easily stream on their
PC

 

eliminated the need for a dedicated PC for video capture, allowing users to
play and stream from the same PC with good stream quality and high fps in games. Finally,
NVIDIA’s Broadcast suite, powered by AI, provides tools for noise and echo removal, virtual
backgrounds, video noise removal, and automatic camera tracking

 

NVIDIA collaborated with OBS Studio to add AV1 — on top of the recently released HEVC and
HDR support — within an upcoming software release, expected later this year. OBS is also
optimizing encoding pipelines to reduce overhead by 35% for all NVIDIA GPUs. The new release
will additionally feature updated NVIDIA Broadcast effects, including noise and room echo
removal, as well as improvements to virtual background.

We’ve also worked with Discord to enable end-to-end livestreams with AV1. In an update releasing
later this year, Discord will enable its users to use AV1 to dramatically improve screen sharing, be
it for game play, schoolwork, or hangouts with friends.

Edited October 3, 2022 by Quackers101

[BiG StroOnZ]

44 minutes ago, MageTank said:

The one I linked from Tomshardware did, though it also certainly favors your statement: https://www.tomshardware.com/reviews/gigabyte-geforce-gtx-titan-black-ghz-edition,3821-8.html

  Reveal hidden contents

 

Reference 780 Ti averaged 125.9 FPS across their gaming test suite, while the reference Titan Black OC averaged 131.3. Now the Titan Black OC did have a factory OC, but I don't really count this much as Kepler was extremely thermal limited after 80C and these cards almost always stayed pegged at 80C, offsetting most OC's anyways. Still, that difference between the two was nearly 6fps, or at that framerate, 4.19% difference in favor of the Titan Black OC at 1080p.

 

If you compare the Windforce versions of both cards when OC'd, the 780 Ti actually came out ahead by 3%, which made sense to me. GK110B is GK110B after all, and having more VRAM than you actually need doesn't really do much for you in terms of performance, but having more power budget for core frequency did, if only by a little.

 

I personally wouldn't consider either card "faster" than one another. I've tested both on water blocks and I can say that when cooling is not a concern, both perform nearly identical. Even if we compare 4K performance, the extra VRAM did nothing when you still met the same ROP and bus limits imposed by both cards. Again, the only game I can recall there being a huge performance difference in was GRID 2, but that was a very niche situation.

 

Yes, the Tom's article you linked earlier was one of the only instances I could find where a non-reference 780 Ti was used as a comparison against the Titan Black. And then of course this is after OC. But the Windforce 780 Ti vs the Windforce Titan Black, the Titan Black was only 0.14% slower (142 vs 141.8). I also found a PC Gamer article where they used a PNY XLR8 edition of the 780 Ti which has a 12% higher base clock and 13% higher boost clock. It scored 2 AVG frames more than the Titan Black but they had the same minimums. The XLR8 edition was still a reference designed cooler though.

Therefore, I'm trying to find the Titan Black being 2-5% slower and am not having much luck. Most of the results are reference vs reference, with having the Titan Black coming out slightly ahead.

 

Anyway, the cards perform very similar as you said. Many of the results I found with reference vs reference have the Titan Black coming out ahead by 1-3 fps most of the time. As you have experience testing the cards firsthand, I'll take your personal input as reliable. Although, I do remember when I was an avid member of Overclock.net that many of the people who were running triple monitor setups or trying to play at 4K or higher resolutions preferred the Titan Black. I do see how you say you will be limited by the cards performance before you hit a VRAM limitation. This has always been the case except in very specific scenarios. Like I remember Watch Dogs preferred 6GB of VRAM cards or it had a stuttering problem. Without heavy mods applied to remove the stuttering, simply running a 6GB card made the experience smoother. I believe the Overclock.net members claimed even though the raw fps might have been the same between the 780 Ti and Titan Black (or 780 and Titan OG), the experience was smoother, with less hitching/stuttering on the 6GB cards in certain games. Not sure what your experience was like. 

 

44 minutes ago, leadeater said:

Well yes reference vs refence but is that actually what is being purchased? 

 

Well I do remember a lot of people at the time preferred the reference design of those cards. Especially the early adopters. While later on AIB/AIC cards with custom coolers would probably be more widely available, before Founders Edition cards existed many AIB/AIC partners/suppliers sold the reference designs with maybe small tweaks like frequency bumps. I do get your point, but most of the reviews I found reviewed the reference designs (which is all I was pointing out).

[MageTank]

15 minutes ago, BiG StroOnZ said:

Anyway, the cards perform very similar as you said. Many of the results I found with reference vs reference have the Titan Black coming out ahead by 1-3 fps most of the time. As you have experience testing the cards firsthand, I'll take your personal input as reliable. Although, I do remember when I was an avid member of Overclock.net that many of the people who were running triple monitor setups or trying to play at 4K or higher resolutions preferred the Titan Black. I do see how you say you will be limited by the cards performance before you hit a VRAM limitation. This has always been the case except in very specific scenarios. Like I remember Watch Dogs preferred 6GB of VRAM cards or it had a stuttering problem. Without heavy mods applied to remove the stuttering, simply running a 6GB card made the experience smoother. I believe the Overclock.net members claimed even though the raw fps might have been the same between the 780 Ti and Titan Black (or 780 and Titan OG), the experience was smoother, with less hitching/stuttering on the 6GB cards in certain games. Not sure what your experience was like. 

Admittedly, until very recently, I used a Dell S2417DG 1440p 165hz monitor, so it is possible that 4K would have scaled differently, but I can't say for certain. Got a 4k 120hz OLED now, but removing a GPU from my loop to test an old card wouldn't be convenient, especially since most of my old GPUs are just art decorations now, lol.

 

I personally wouldn't call the 780 Ti faster than a Titan Black. For all intents and purposes, they were the same exact GPU just with a difference in VRAM capacity. Those that needed the VRAM could probably see a benefit from it, but I would never recommend that card if gaming was the goal. SLI 780/780 Ti's was just a much better value preposition back then.

 

Besides, if you could afford a Titan Black, you could also afford a water block. In which case, your Titan Black certainly wasn't slower than a 780 Ti, lol.

[Deadpool2onBlu-Ray]

On 9/29/2022 at 11:03 PM, BiG StroOnZ said:

Newegg has some AIB/AIC listings for the RTX 4090:

 

 

 

Fun fact. Did you guys know the "4080" 12gb won't have a founders edition? So we are gonna be stuck paying 1k plus for a x60ti tier card

[ewitte]

1 hour ago, CHICKSLAYA said:

 

Fun fact. Did you guys know the "4080" 12gb won't have a founders edition? So we are gonna be stuck paying 1k plus for a x60ti tier card

Just a 4080FE.  Fake Edition.

[starsmine]

1 hour ago, CHICKSLAYA said:

So we are gonna be stuck paying 1k plus for a x60ti tier card

Fun fact. no?
that's not how this works, that's not how any of this works. 
Why make up shit like x60ti tier when nothing supports that conclusion. 

Its a full fat 104 die, 2080 didnt even get the full fat 104 die. 
Also it costs more then every xx70 card to make, ever.

if your point is the memory bus
dont forget, MASSIVE archectural changes with memory

48MB of L2, let me repeat that 48MB of L2 cache

You know how much H100 has? 56MB

A100? 41MB

A102, the 3090, the big top dog of last gen for gamers? 6.... 6MB

aka, the amount of times the card has to hit even the ram is significantly less.
Who the fuck gives a shit about 192bit bus when the amount of cache misses forcing you into ram is literally decimated. 
The whole paradigm is different in how memory is managed. so to call it an x60ti class of card is a bad and dumb take.

Honestly, I personally think Nvidia will drop the price of it pretty damn fast from 900 to something like 800, but only because of RDNA3 

[BiG StroOnZ]

10 minutes ago, starsmine said:

Its a full fat 104 die, 2080 didnt even get the full fat 104 die. 
Also it costs more then every xx70 card to make, ever.

 

Knew I would be able to finally post this chart, signifying cost of a chip on a leading edge node:

 

[Quackers101]

ah yes, because newer cards is worth the added expenses, so let people miss the older cards and their prices. or buy two 3080s for one 4080

how the 1070ti, still able to push against some of the newer ones, when it was around 399 which might be 600-700 now.

https://www.youtube.com/watch?v=H11VBZFLRrI

[Sir Beregond]

On 10/2/2022 at 9:14 PM, leadeater said:

RTX 3090 and RTX 4090 are just as much Titans as the Titan RTX, Titan X and Titan Xp were....

How long has that been true? Titan had Titan drivers and GeForce did not and as such had specific limitations around the FP16 tensor ops and FP32 accumulate performance. We can even see this in the original LTT review of the 3090 at launch.

 

So news to me if that's not still true.

[leadeater]

26 minutes ago, Sir Beregond said:

How long has that been true? Titan had Titan drivers and GeForce did not and as such had specific limitations around the FP16 tensor ops and FP32 accumulate performance. We can even see this in the original LTT review of the 3090 at launch.

 

So news to me if that's not still true.

The Titan drivers are the Studio Drivers. The Studio Drivers are based on the Quadro Drivers just without the software and hardware validation. As the video explains, Nvidia are just being dicks about it and selectively not allowing functions for certain device class identifiers to push people over to more expensive products.

 

It's not a hardware limitation or a driver limitation, simply an artificial limitation. 

 

 

Since there is no Titan after the Titan RTX and likely will not be one and the affected applications and use cases is quite small I'm quite certain this has been done specifically by Nvidia either at the behest of themselves or their partners because of the impact to the more expensive Quadro sales and OEM/ODM systems that come with Quadros and carry things like SolidWorks certifications for application compatibility (yea these things exist heh).

 

So you're right there is a very select feature set not allowed however the implication here is simply that Nvidia doesn't have nor as I can see intend to release any new Titans or address this issue and I would be quite weary that if a new Titan does come it will still be allowed this feature set.

 

It's a pity really since I liked the idea of the Titan cards, Nvidia kept messing around with them though. Feels like they regret making them in the first place.

[CTR640]

Differences between the 4080/16 and 4080/12 in performance:

 

https://www.nvidia.com/en-us/geforce/news/overwatch-2-out-now-geforce-rtx-reflex-high-fps/

 

Yep, the 4080/12 is simply 4070 or even 4060Ti.

[starsmine]

1 hour ago, CTR640 said:

Differences between the 4080/16 and 4080/12 in performance:

 

https://www.nvidia.com/en-us/geforce/news/overwatch-2-out-now-geforce-rtx-reflex-high-fps/

 

Yep, the 4080/12 is simply 4070 or even 4060Ti.

find me any generation in the last 8 years where the x70/xx70 was the FULL 104/204 chip.

I know the answer to this question, so its rhetorical, but I want you to find it yourself. 
so no, its not simply a 4070, and it likely never was considered to be. 

there are full 104 chips used for the 70ti class cards, but never 70. so sure argue it could have been called the 4070ti. but again, it was likely never going to be a 4070

[Kilrah]

21 minutes ago, starsmine said:

find me any generation in the last 8 years where the x70/xx70 was the FULL 104/204 chip.

The argument is that the chip used/architecture is irrelevant to the user, the performance stack and relative performance between models is what they care about and what they want the names to reflect.

[leadeater]

27 minutes ago, starsmine said:

find me any generation in the last 8 years where the x70/xx70 was the FULL 104/204 chip.

 

27 minutes ago, starsmine said:

so no, its not simply a 4070, and it likely never was considered to be. 

 

You are aware that the RTX 3070 is a GA104-300 die with 46 active SM's, RTX 3070 Ti is a GA104-400 die with 48 active SM's , RTX 3080 is a GA102-200 die with 68 active SM's, RTX 3080 Ti is a GA102-225/250 die with 80 active SM's

 

Difference between RTX 3070 and RTX 3080 is a 48% increase.

Difference between RTX 3080 and RTX 3080 Ti is a 18% increase

 

RTX 4080 12GB is a AD104-400 die with 60 active SM's, RTX 4080 16GB is a AD103-300 die with 76 active SM's

 

Difference between RTX 4080 12GB and RTX 4080 16GB is a 27% increase.

 

RTX 4080 has a legitimate problem here, it's above the difference of a Ti and dangerously close to x70.

 

27 minutes ago, starsmine said:

so sure argue it could have been called the 4070ti. but again, it was likely never going to be a 4070

Difference between x70 and x70 Ti is miniscule, barely worth differentiating if talking about GPU dies, SM counts and CUDA cores. Main difference at least for 30 series was the memory bandwidth, same bus width just much different transfer rates of the memory modules.

 

The heart of the issue has nothing to do with what Nvidia calls the GPU dies, it's the performance difference between a RTX 4080 and a RTX 4080.....

[MageTank]

3 hours ago, starsmine said:

find me any generation in the last 8 years where the x70/xx70 was the FULL 104/204 chip.

I know the answer to this question, so its rhetorical, but I want you to find it yourself. 
so no, its not simply a 4070, and it likely never was considered to be. 

there are full 104 chips used for the 70ti class cards, but never 70. so sure argue it could have been called the 4070ti. but again, it was likely never going to be a 4070

GK104 (Kepler) with the GTX 770. Kepler as a generation is still within that 8 year timeframe (they technically released Kepler cards in 2016). What do I win?

 

Man, gotta love that we already had Kepler on the brain from earlier in this thread, lol.