Computer GPU/CPU/tech Mega Thread Based on new NVidia/AMD launches in 2025 and beyond. (Page 3 of 47)
|
Originally Posted By Flammenwerfer35: I'm guessing the artifacting is a driver issue and that will get sorted out quickly. Early drivers are always buggy. Originally Posted By Flammenwerfer35: Originally Posted By Dam17er: I didn't watch the LTT video because I think LTT is trash, but Digital Foundry just put out a video on DLSS 4 and how it compares to current gen. There are definitely noticeable artifacts in CP2077, but the DF vid also shows some improvements in graphical fidelity. I'm hopeful that there will continue to be improvements in overall image quality, as there should be if people are going to drop $2k+ on a new card.  https://www.youtube.com/watch?v=xpzufsxtZpAI'm guessing the artifacting is a driver issue and that will get sorted out quickly. Early drivers are always buggy. The main issue is with RT, pop-in, and culling vs new data/pixels. Not all of that can be solved with new drivers since some of it is just inherent to the current engine/technology and its limitations. |
|
Originally Posted By -daddy: I 'member when a $600 card was considered "insane." https://i.imgur.com/Fu5nevA.gif Originally Posted By -daddy: Originally Posted By Ayada: Originally Posted By Electricview: Damn.. I swear I just paid 599... Or was it 699? for a 3080 like 3 years ago? Now its a grand for the newer series similar 5080? sheesh.. Bidenomics .. MSRP was $699 for the 3080 but the founders edition at that price was unobtanium. The brands were selling 3080s damn near past $1k and the 3080ti was a $1600 card up until the supply caught up in a massive way and the prices fell through the roof in preparation for the 4000 series launch. Don't get me wrong, I do wish we were still back in the days of $600 cards being the envy of pc builders and not the 'well it's kind of like a 4090 if you use DLSS in these two games' option. I 'member when a $600 card was considered "insane." https://i.imgur.com/Fu5nevA.gif Pretty much what I said to my buddies last night. "I miss the days of $500 top of the line cards." Hell I remember when top of the line was $250-$300. |
|
Originally Posted By klinc: Pretty much what I said to my buddies last night. "I miss the days of $500 top of the line cards." Hell I remember when top of the line was $250-$300. Originally Posted By klinc: Originally Posted By -daddy: Originally Posted By Ayada: Originally Posted By Electricview: Damn.. I swear I just paid 599... Or was it 699? for a 3080 like 3 years ago? Now its a grand for the newer series similar 5080? sheesh.. Bidenomics .. MSRP was $699 for the 3080 but the founders edition at that price was unobtanium. The brands were selling 3080s damn near past $1k and the 3080ti was a $1600 card up until the supply caught up in a massive way and the prices fell through the roof in preparation for the 4000 series launch. Don't get me wrong, I do wish we were still back in the days of $600 cards being the envy of pc builders and not the 'well it's kind of like a 4090 if you use DLSS in these two games' option. I 'member when a $600 card was considered "insane." https://i.imgur.com/Fu5nevA.gif Pretty much what I said to my buddies last night. "I miss the days of $500 top of the line cards." Hell I remember when top of the line was $250-$300. Voodoo2 was $300, and you needed two of them for top of the line performance. That's about $1200 in today's dollars. They were also hopelessly obsolete in just a few years. A pair of GTX 980s was $1100 in 2014, or $1500 today. |
Member
Joined:
Jun 2022
Posts:
8188
EE: 100% (3)
IA, USA
|
Originally Posted By LittlePony: The 5070 has my attention. Mild but more intelligent improvement over a 4070 architecture. 12GB of VRAM will have issues on current titles, let alone titles over the next 3 years. 4090 performance sounds nice until you realize its *in supported games using DLSS4 and Frame Generation 4x on the 5070. |
|
Originally Posted By eurotrash: 12GB of VRAM will have issues on current titles, let alone titles over the next 3 years. 4090 performance sounds nice until you realize its *in supported games using DLSS4 and Frame Generation 4x on the 5070. Not for what I'm playing, I'm still rocking a 2080 My bottleneck is cpu, none on the market are strong enough in standard desktop architecture for what I need. |
Twitter ID: @GattoSeh
|
Originally Posted By JustinU235: RTX 5090: $1999. Twice the power of 4090. RTX 5080: $999 RTX 5070Ti: $749 RTX 5070: $549. 4090 performance.  The 5070 is NOT as fast as the 4090...marketing BS explained |
|
Yeah always look at hardware first Honestly, the biggest thing that has annoyed me about the current video card market for the past 10 years is that they have always under performed with Vram deliverance in their products. None of the cards I would say is acceptable amounts of vram A 5090 should have 64 gb of vram for its price, if not 128 if just to gloat We game in an era of very poorly optimized engines that eat vram like candy so why they always scimp is beyond me. |
Twitter ID: @GattoSeh
|
Originally Posted By LordEC911: The main issue is with RT, pop-in, and culling vs new data/pixels. Not all of that can be solved with new drivers since some of it is just inherent to the current engine/technology and its limitations. This is why the best settings I've found for gaming, for example, don't include using DLSS. |
Become prompt critical.
|
Originally Posted By JustinU235: This is why the best settings I've found for gaming, for example, don't include using DLSS. I avoid it where possible. Having an ultra wide 1440p does mean I can stretch settings out further than those pushing 4k at least. However since the press release was really focused on 'double the performance (with DLSS)' I'm certainly more interested in what the actual performance uplift will be. |
|
Originally Posted By JustinU235: This is why the best settings I've found for gaming, for example, don't include using DLSS. At least the tech has no where to go but up, I'm hopeful for the tech, but it won't be great i think until dlss 6 comes out |
Twitter ID: @GattoSeh
|
Originally Posted By castlebravo84: Voodoo2 was $300, and you needed two of them for top of the line performance. That's about $1200 in today's dollars. They were also hopelessly obsolete in just a few years. A pair of GTX 980s was $1100 in 2014, or $1500 today. I miss the days of Ti 4200 value. |
|
Originally Posted By Skoal7: I have never had a GPU die. I have stacks and stacks of GPUs going back to Pentium 100mhz cpu days. I’m sitting this one out and sticking with my 4080 Super that only eats 300 watts. Would you possibly be interested in selling an old one? |
|
Originally Posted By LittlePony: Yeah always look at hardware first Honestly, the biggest thing that has annoyed me about the current video card market for the past 10 years is that they have always under performed with Vram deliverance in their products. None of the cards I would say is acceptable amounts of vram A 5090 should have 64 gb of vram for its price, if not 128 if just to gloat We game in an era of very poorly optimized engines that eat vram like candy so why they always scimp is beyond me. Originally Posted By LittlePony: Yeah always look at hardware first Honestly, the biggest thing that has annoyed me about the current video card market for the past 10 years is that they have always under performed with Vram deliverance in their products. None of the cards I would say is acceptable amounts of vram A 5090 should have 64 gb of vram for its price, if not 128 if just to gloat We game in an era of very poorly optimized engines that eat vram like candy so why they always scimp is beyond me. Ummmm.... what? For what benefit? That makes zero sense... you are basically asking for increased VRAM for no reason while either significantly decreasing raw GPU performance or increasing the cost by a crazy amount. |
| What ever happened to just rendering the actual frames the game is generating, instead of retarded AI upscaled fake resolution hijnks and fake frame generation voodoo bullshit. Have these people actually played games heavily relying on DLSS? It looks like shit and feels nowhere near as smooth as actual rendered frames. 75fps native looks and feels better than 120fps frame generation |
|
Originally Posted By Hking: What ever happened to just rendering the actual frames the game is generating, instead of retarded AI upscaled fake resolution hijnks and fake frame generation voodoo bullshit. Have these people actually played games heavily relying on DLSS? It looks like shit and feels nowhere near as smooth as actual rendered frames. 75fps native looks and feels better than 120fps frame generation Native rendering doesn't make huge numbers and we can't sell market buzzwords with only 25% uplift on a generation. We need FIVE TIMES THE FRAMES AT SIX TIMES THE RESOLUTION (upscaled from 480p). |
|
Originally Posted By castlebravo84: Voodoo2 was $300, and you needed two of them for top of the line performance. That's about $1200 in today's dollars. They were also hopelessly obsolete in just a few years. A pair of GTX 980s was $1100 in 2014, or $1500 today. Think mine was $279 back in the late 90s IIRC. Cost almost three months jump pay 
|
Joined:
Mar 2023
Posts:
462
EE: 0% (0)
MD, USA
|
Originally Posted By Sir_Sparhawk: At least AMD still cares about regular people and offers GPU's at reasonable prices. Nvidia only seems to want to sell to rich people and corporations. I leave influencers out of that statement because they usually shill hard enough to get cards for free. Let's see if they get their heads out of their asses and come up with a better ROCm to compete with CUDA. |
Member
Joined:
May 2024
Posts:
483
EE: 0% (0)
WI, USA
|
Originally Posted By Hking: What ever happened to just rendering the actual frames the game is generating, instead of retarded AI upscaled fake resolution hijnks and fake frame generation voodoo bullshit. Have these people actually played games heavily relying on DLSS? It looks like shit and feels nowhere near as smooth as actual rendered frames. 75fps native looks and feels better than 120fps frame generation They figured out they can make more money off optimizing for their AI customers and figuring out how to use that hardware to give us fake AI frames. How many people that you know buy a new tv and don't turn off that smoothing shit? None of those people are even going to know it looks like shit, they just see number go up. |
|
Originally Posted By FreefallRet: I had a Voodoo2, man my tribes and quake 2 FPS for killer. Think mine was $279 back in the late 90s IIRC. Cost almost three months jump pay Originally Posted By FreefallRet: Originally Posted By castlebravo84: Voodoo2 was $300, and you needed two of them for top of the line performance. That's about $1200 in today's dollars. They were also hopelessly obsolete in just a few years. A pair of GTX 980s was $1100 in 2014, or $1500 today. Think mine was $279 back in the late 90s IIRC. Cost almost three months jump pay Lucky! I played Half Life on software renderer at 320x240 until I was able to make enough money on a high school summer break job to get a TNT2. |
|
Originally Posted By Hking: What ever happened to just rendering the actual frames the game is generating, instead of retarded AI upscaled fake resolution hijnks and fake frame generation voodoo bullshit. Have these people actually played games heavily relying on DLSS? It looks like shit and feels nowhere near as smooth as actual rendered frames. 75fps native looks and feels better than 120fps frame generation Easy enough to turn that off and render entirely natively. |
I'm not the one REEING, motherfucker! -FCSD2162
Member
Joined:
May 2024
Posts:
484
EE: 0% (0)
WI, USA
|
Originally Posted By Flammenwerfer35: Twice the power? Lol, they say that every time. Probably more like 30%. Just like every other time. I've been watching these launches for 20 years and they always claim the cards are 75-100% faster, and it's always been 30%.  PC Gamers Learn the truth about RTX 5090's True Performance |
|
Originally Posted By LittlePony: Inflation and powercreep are a bitch Originally Posted By LittlePony: Originally Posted By klinc: Pretty much what I said to my buddies last night. "I miss the days of $500 top of the line cards." Hell I remember when top of the line was $250-$300. Inflation and powercreep are a bitch I agree* GPUs have been out of sync in the price/performance game compared to other components of the build. CPU, RAM, MOBO, PSU and storage have maintained solid price/performance scale with storage specifically being incredibly cheap for what the end user gets. |
|
Originally Posted By LordEC911: Ummmm.... what? For what benefit? That makes zero sense... you are basically asking for increased VRAM for no reason while either significantly decreasing raw GPU performance or increasing the cost by a crazy amount. for no reason? dude its easy to max 16gb with many games now, and its only getting worse. |
Twitter ID: @GattoSeh
|
Only one of these new cards that will beat the 4090 is the 5090. 3 AI generated frames for every 1 rendered? That's going to probably be a smeary, artifacty, input laggy mess. Lol 5070 claimed to meet 4090 performance. Right, generate 75% fake frames to "meet" the 4090 output. Ok Jensen.... Current Frame Gen is render 2 frames, use AI to predict and create a "middle" frame and insert between both rendered frames. Looks pretty good and next to no artifacting unless you wildly move around the camera view. 5090 has a 30% bigger core, more and faster RAM on a faster buss. It should truck the 4090. Everything else has less RAM, slower busses and less cores. Guarantee the 50 series driver and feature software is artificially locked to make it "better". 40 would probably run it just fine. Wonder if someone will jail break it for 40 series? Waiting to see how the 9950X3D performs in gaming vs the 9800X3D, and I might jump to AM5. Just hope my RAM doesn't come flat pack shipped....  |
|
Originally Posted By Lexustech48: Only one of these new cards that will beat the 4090 is the 5090. 3 AI generated frames for every 1 rendered? That's going to probably be a smeary, artifacty, input laggy mess. Lol 5070 claimed to meet 4090 performance. Right, generate 75% fake frames to "meet" the 4090 output. Ok Jensen.... Current Frame Gen is render 2 frames, use AI to predict and create a "middle" frame and insert between both rendered frames. Looks pretty good and next to no artifacting unless you wildly move around the camera view. 5090 has a 30% bigger core, more and faster RAM on a faster buss. It should truck the 4090. Everything else has less RAM, slower busses and less cores. Guarantee the 50 series driver and feature software is artificially locked to make it "better". 40 would probably run it just fine. Wonder if someone will jail break it for 40 series? Waiting to see how the 9950X3D performs in gaming vs the 9800X3D, and I might jump to AM5. Just hope my RAM doesn't come flat pack shipped.... I understood that reference. |
Become prompt critical.
|
Originally Posted By castlebravo84: There will be 3GB GDDR7 modules soon, so I would expect Nvidia to release a 24GB 5080 Ti/Super in maybe a year. I could go for that to replace my 3090Ti in my main pc. |
A tough man can take a bullet, but a wise man can dodge one. Stay focused my brothers.
|
Originally Posted By buckstrucks: My buddy decided he wanted a 5090, and offered me his 4090 for $1000. I won't be able to get it until he gets a new one, and I won't be home to enjoy it for 7 weeks or so. Do I want it? Or pretend I want it for a while until we see what happens to 4090 prices? There are second hand 3090's for 7-$800 on ebay right now. 4090's are gonna bring a premium for a while i think. Still offer him $800 and see if he bites. |
|
Originally Posted By LittlePony: for no reason? dude its easy to max 16gb with many games now, and its only getting worse. Originally Posted By LittlePony: Originally Posted By LordEC911: Ummmm.... what? For what benefit? That makes zero sense... you are basically asking for increased VRAM for no reason while either significantly decreasing raw GPU performance or increasing the cost by a crazy amount. for no reason? dude its easy to max 16gb with many games now, and its only getting worse. Well... to be fair you were talking about how 32GB wasn't enough... but would love to see realworld examples of 16GB of VRAM causing performance issues. But yes, in most cases increasing the VRAM capacity, especially 2-4x the current amount, only decreases the overall performance due to the increased power consumption of the memory subsystem. Most modern engines just load and leave everything in the VRAM if there is enough capacity but leave enough available to not cause any issues. There is a very big difference between VRAM allocation and real world usage. That is part of the reason bigger caches started becoming a thing over the last decade. |
|
Originally Posted By buckstrucks: My buddy decided he wanted a 5090, and offered me his 4090 for $1000. I won't be able to get it until he gets a new one, and I won't be home to enjoy it for 7 weeks or so. Do I want it? Or pretend I want it for a while until we see what happens to 4090 prices? 5080 should be roughly equivalent to the 4090 in gaming for the same $1k, but would be a new card with a warranty, and would likely be more power efficient. The AI workstation performance of the 4090's higher vram capacity will probably keep it's used value at or over $1k for now. If you aren't doing non-gaming AI stuff, both you and your buddy might be better off of he ebays his 4090 and you get a 5080. |
|
Originally Posted By LordEC911: Well... to be fair you were talking about how 32GB wasn't enough... but would love to see realworld examples of 16GB of VRAM causing performance issues. Most modern engines just load and leave everything in the VRAM if there is enough capacity but leave enough available to not cause any issues. There is a very big difference between VRAM allocation and real world usage. That is part of the reason bigger caches started becoming a thing over the last decade. what i am saying that like motherboard ram it should never be a issue, like ever. Hunt Showdown is my most intimate example, my 2080 on low constantly maxes my 8 most of each match, good luck trying to get a 4090 to get 120 fps if doing absurd 4k or 8k impractical resolutions on max, i barley get 100fps at 2k with my 2080 on low and that is a impressive feat. it would not be so bad if the current era of game engines were not messy. really though, issue i have anymore is the game type i generally play is not even gpu based. its cpu based, another but far more understandable annoyance. i play paradox games, far beyond their intended time frame, say Victoria 3, its a nation building game that like most paradox games, like most paradox games of their classic format it is a massive tabulating machine legendary for their cpu melting taxation, game set intends you to play from 1836 to 1936, but you can close the end game score card and keep going, however the calculations eventually become so immense that not even the top end cpu in standard desktop format can run without the game slowing to a crawl, engine limitations and the sheer size of the numbers involved can cause 5 second imput delays if i had a 14900 and had 64gb of board ram i could perhaps play up to the year 2250 but alas im stuck only going to the year 1980 at most, not like they make 15GHz processors for desktops like i want anyway. |
Twitter ID: @GattoSeh
|
Originally Posted By LittlePony: what i am saying that like motherboard ram it should never be a issue, like ever. Hunt Showdown is my most intimate example, my 2080 on low constantly maxes my 8 most of each match, good luck trying to get a 4090 to get 120 fps if doing absurd 4k or 8k impractical resolutions on max, i barley get 100fps at 2k with my 2080 on low and that is a impressive feat. it would not be so bad if the current era of game engines were not messy. really though, issue i have anymore is the game type i generally play is not even gpu based. its cpu based, another but far more understandable annoyance. i play paradox games, far beyond their intended time frame, say Victoria 3, its a nation building game that like most paradox games, like most paradox games of their classic format it is a massive tabulating machine legendary for their cpu melting taxation, game set intends you to play from 1836 to 1936, but you can close the end game score card and keep going, however the calculations eventually become so immense that not even the top end cpu in standard desktop format can run without the game slowing to a crawl, engine limitations and the sheer size of the numbers involved can cause 5 second imput delays if i had a 14900 and had 64gb of board ram i could perhaps play up to the year 2250 but alas im stuck only going to the year 1980 at most, not like they make 15GHz processors for desktops like i want anyway. There's a reason 8k is basically on life support, and even 4k isn't as normalized as it should be, considering it's been around for over a decade. You have to spend a shit ton of money just to have a truly great 4k experience, and even then the goal post for "truly great" is ever moving. Lots of people want frame rates well above 60 now, because they want high refresh rates as well (and that's ignoring the issues with other 4 and 8k media, like movies, but I don't want to go on too long about it). That simply isn't feasible with native rendering at 4k without AI, so while Jensen's comments may be a bit misleading, he's not really wrong in judging performance of future gens based on AI. I can only hope that frame gen gets better at visual fidelity, because it's legitimately the only way to get your 120s and 240s, or however many hertz tickles your fancy, and it will be the only way for a long time. |
|
Originally Posted By JustinU235: Teaser vids coming out. https://www.youtube.com/watch?v=FDRR4pt_UjM https://www.youtube.com/watch?v=37yHH3YR7jM https://www.youtube.com/watch?v=ELzHzkKceXY https://www.youtube.com/watch?v=qguE6JDYXmQ https://www.youtube.com/watch?v=PKUjMoxfYp8Interesting that all of the partner models are chonky blocks compared to Nvidia's versions. I think massive GPUs are more aesthetically pleasing. 
|
Hardware Unboxed just dropped a vid where they go hands on with a bunch of different 50 series cards at CES. 42 Graphics Cards! Hands-On With RTX 5090, RTX 5080, RX 9070 XT, RTX 5070 and More A few things that stood out on the 5090 models: The air cooled variants are absolutely massive, which was expected given the TDP. The AIO variants come with a 360mm radiator (happy about this). The AIO variants now have the tubing coming out of the end of the card, which should provide more flexibility when mounting the rad (very happy about this). |
|
Originally Posted By Lexustech48: Only one of these new cards that will beat the 4090 is the 5090. 3 AI generated frames for every 1 rendered? That's going to probably be a smeary, artifacty, input laggy mess. Lol 5070 claimed to meet 4090 performance. Right, generate 75% fake frames to "meet" the 4090 output. Ok Jensen.... Current Frame Gen is render 2 frames, use AI to predict and create a "middle" frame and insert between both rendered frames. Looks pretty good and next to no artifacting unless you wildly move around the camera view. 5090 has a 30% bigger core, more and faster RAM on a faster buss. It should truck the 4090. Everything else has less RAM, slower busses and less cores. Guarantee the 50 series driver and feature software is artificially locked to make it "better". 40 would probably run it just fine. Wonder if someone will jail break it for 40 series? Waiting to see how the 9950X3D performs in gaming vs the 9800X3D, and I might jump to AM5. Just hope my RAM doesn't come flat pack shipped.... @Lexustech48 I don't fully agree with you on the raw power of the 5090 "trucking" the 4090. One mysterious thing that has received far too limited attention IMO is the abysmal clocks of the 5090, which are actually slower than the 4090. According to NVIDIA's own website, the 5090 has a base clock of 2010 Mhz and boost clock of 2410 Mhz. Compare this with the 5080 of the same generation's improved base clock of 2300 Mhz and boost clock of 2620 Mhz. The 4090 had a base clock of 2230 Mhz and boost clock of 2520 Mhz. While the 5090 does have more CUDA cores than the 4090, 21760 versus 16384, a ~33% improvement, the clocks are really holding this back from being a better card. A net 25% to 30% improvement in raw performance is meh in my opinion. With improved clocks, it would have been closer to the ~50% improvement of last generation. I don't think that memory speed will have much effect, or at least it has not in my experience. There has thus far been no explanation as to why the 5090 base and boost clocks are down ~10% for the base and ~5% for the boost from the 5090 to 4090. It is particularly mysterious when the 5080 has clocks that are up ~8% on the base and up ~4% on the boost from the 4090. Notably on launch, the 4080 had clocks almost identical to the 4090 (with very small improvements made by the later-released 4080 Super). Has the 5090 been deliberately nerfed by clock-throttling? It seems so, but it is hard to tell. If it is instead thermally limited, why does the founder's edition use a two slot design? Is this an effort to appease board partners like Gigabyte and Asus by letting them stand out in performance with better cooling? I am sure the board partners were upset with NVIDIA being a direct competitor when the 4090 performed similarly across all variants because the 4090 founder's edition had overengineered cooling. I would note that the maximum power draw for a card using these connectors is 675w, 600 from the power cable and 75w from the PCIe rail. That means there is additional power available even with the TDP of 575w. Perhaps the plan, as hinted above, is to place NVIDIA's own card into more of a small form factor PC niche while giving board partners the more lucrative extreme performance users through improved cooling. Or maybe NVIDIA is throttling the card to offer a Titan variant in the near future. In any event, the 5090 founders edition is not as impressive as it should be in performance IMO. I also concur with your disdain for increased frame generation and other visual tricks. |
|
Originally Posted By Dam17er: Hardware Unboxed just dropped a vid where they go hands on with a bunch of different 50 series cards at CES. https://www.youtube.com/watch?v=3-s4KrPlIx4A few things that stood out on the 5090 models: The air cooled variants are absolutely massive, which was expected given the TDP. The AIO variants come with a 360mm radiator (happy about this). The AIO variants now have the tubing coming out of the end of the card, which should provide more flexibility when mounting the rad (very happy about this). While the 4090 founder's edition was always the smallest card of that iteration, the difference between the 5090 founder's edition cards and other 5090s is now mind-boggling. |
|
Originally Posted By FLAJD11: @Lexustech48 I don't fully agree with you on the raw power of the 5090 "trucking" the 4090. One mysterious thing that has received far too limited attention IMO is the abysmal clocks of the 5090, which are actually slower than the 4090. According to NVIDIA's own website, the 5090 has a base clock of 2010 Mhz and boost clock of 2410 Mhz. Compare this with the 5080 of the same generation's improved base clock of 2300 Mhz and boost clock of 2620 Mhz. The 4090 had a base clock of 2230 Mhz and boost clock of 2520 Mhz. While the 5090 does have more CUDA cores than the 4090, 21760 versus 16384, a ~33% improvement, the clocks are really holding this back from being a better card. A net 25% to 30% improvement in raw performance is meh in my opinion. With improved clocks, it would have been closer to the ~50% improvement of last generation. I don't think that memory speed will have much effect, or at least it has not in my experience. There has thus far been no explanation as to why the 5090 base and boost clocks are down ~10% for the base and ~5% for the boost from the 5090 to 4090. It is particularly mysterious when the 5080 has clocks that are up ~8% on the base and up ~4% on the boost from the 4090. Notably on launch, the 4080 had clocks almost identical to the 4090 (with very small improvements made by the later-released 4080 Super). Has the 5090 been deliberately nerfed by clock-throttling? It seems so, but it is hard to tell. If it is instead thermally limited, why does the founder's edition use a two slot design? Is this an effort to appease board partners like Gigabyte and Asus by letting them stand out in performance with better cooling? I am sure the board partners were upset with NVIDIA being a direct competitor when the 4090 performed similarly across all variants because the 4090 founder's edition had overengineered cooling. I would note that the maximum power draw for a card using these connectors is 675w, 600 from the power cable and 75w from the PCIe rail. That means there is additional power available even with the TDP of 575w. Perhaps the plan, as hinted above, is to place NVIDIA's own card into more of a small form factor PC niche while giving board partners the more lucrative extreme performance users through improved cooling. Or maybe NVIDIA is throttling the card to offer a Titan variant in the near future. In any event, the 5090 founders edition is not as impressive as it should be in performance IMO. I also concur with your disdain for increased frame generation and other visual tricks. Respectfully, I don't think we can look at the clock speeds between two different architectures/generations and derive performance results. At this point, we don't really have any solid numbers that I'm aware of, probably because of NDA's, and I'm not relying on Nvidia's Even though 675w is the max in theory, I'm guessing in practice most cards will require a bios update to get anywhere near that, similar to 4090 owners updating to the Galax 666w bios. I hope I'm wrong on this. Originally Posted By FLAJD11: While the 4090 founder's edition was always the smallest card of that iteration, the difference between the 5090 founder's edition cards and other 5090s is now mind-boggling. Yeah the disparity in size between the 5090 FE and other air cooled cards is...uh...substantial. I don't really understand why Nvidia came up with a relatively slim design for the FE but the board partners went the opposite direction. Guess we'll figure that out as more info comes out. |