@@wahidtrynaheghugh260 Not true! Steve's cat has most definitely been in his videos! ... just not so much after they got an actual studio, and later the current school-studio 😅
This is also the first card in Nvidias lineup to have tensor cores. It's a very interesting card that was quite a bit faster than those titan X small and big p's.
I thought maybe you could use DLSS on one of these, which is one of the "modern features" worth upgrading for, but a quick Google search tells me Volta has an older tensor core design that is good for AI research, but can't leverage DLSS.
True, the HBM is what always most interested me, MCD or cache dies outside the core, link with a one layer hbm2/3/4 feels like it'd be the most impressive advancement we could have for feeding P Cores moving forward, while keeping chiplet latency low. @@ChristianStout
Anyone happen to know what its eth hashrate was? Just for posterity? Gonna look up if it beat the radeon 7 just for infos sake, but I have a feeling it might be hard to find that info
Titan X Pascal and newer generation of Titan XP never received a black badge like Titan Black. I had 4x Titan X Maxwell in SLI and world first 16K video resolution gamer and pioneer on RUclips and 2x Titan X Pascal in SLI and updated 4x Titan XP in SLI. You maybe talking about color but every Titan were black before Titan V or Titan RTX.
Thank you for this, I'm owner of one and was thinking how it stack. Few things to note : 1) I always undervolt mine, since default power/temp limits force lower clock in long test than with it. In My case, I settled for 0.8V @ 1620MHz via Afterburner V/F curve (going too low on voltage prevents card from reaching 3D clocks). 2) Power usage gets solved with undervolt 3) Some programs will be locked to 1350-ish MHz core clock, and that's "CUDA Compute Mode" thing. To disable throttling under it, you need to go into NV profiler and manually diasable that (this setting may be reset after driver update) 3) You don't have to remove the whole heatsink to see/repaste GV100. Window part of the cooler can be removed with 4 screws to clean heatsink (the same as on Pascal/Maxwell/Kepler "FE" cards that use similar design), while heatsink removal only needs 4 more.
@@theduck17 I hope it's about the cat. After all, cats that are capable of running Nvidia's proprietary software, and also can be overclocked and undervolted, are something worth noting. AMD and Intel really need to take lessons from the cat, if what I said was true.
I hear they also make use of their large VRAM buffer for intensive workloads paired with their wide memory bandwidth just means users have a lot of options when feeding it FISH (Float Integer String Handouts) @@jamesbrendan5170
Volta was never meant for gaming and that's why it never saw mass release in consumer market. I remember one of my friends had procured it for Machine learning(Artificial neural network training) and they sure found the price justified and his firm recovered the money within a year or so of purchase.
ChaitanyaShukla2503 Quadro Bridge, why he called it consumer card ? Was he not around back then ? Only Direct sale card from Nvidia, tot in German shops. You friend did College, you ?
It depends on your use case, if you are doing FP64 work loads than this card is about 5x faster than a 4090. Also you could find SLI boards with 4 slots back then so theoretically you could have 20x the FP64 performance in one rig for 600$ more than a 4090. So for people that are doing that kind of work at 600$ it is actually a really great price.
Very much so. If my problems didn't need more VRAM and ECC memory wasn't a must it would have been a steal even at full price. The Quadro GV100 I use is still available new today at even higher than original MRSP - if you need active cooling and display outputs there's no better GPU than the Titan V/GV100 for FP64 CUDA, otherwise you're onto the A100
this is assuming SLI actually works the way the marketing purported it to. which it patently did not. which is why it failed and was forgotten. this comment is laughable.
@@lucasrem SLi doesn't originally stack Ram nor is it meant to, it is to increase the output resolution of the viewport, not the scene complexity or texture ram. The texture has to be replicated across to every GPU for it's TMU to access and map correctly, the compositer of the output comes last in the chain and has nothing to do with the textures other than compositing the stream as it comes out from the TMU. That's what used to irritate me to no end when I read magazines in the "SLi era" because they used to add the TRam if the cards up as if that number had a meaning - it never did. If you SLi connect 4x4Gb cards you still have a 4Gb card, not a 16Gb card. You're adding together the rasterizing and resolution "power" of the GPU's but not their texture ram. There were multiple modes used but actual SLi does not "stack" anything other than output resolution, that's it. I ran a quad SLi set for a while, two 7950Gx2's bridged, and ended up having to try to explain what it is/does and isn't/doesn't do many times to people mostly when I upgraded to a single 8800Ultra I think it was, it was so long ago I forget now. My 4x 7950's with 512mb each did not result in a 2Gb GPU bridged, it resulted in a 3.5 times faster 512mb GPU, even that is kinda 'wrong' - it resulted in a 7950 with 3.5 times the output resolution capability of a single card - mostly. It's not 100% apples to apples that way either, but lightyears closer to ram stacking them pretending it's a 2Gb card in some form.
the main selling point of this gpu back then (and depending of use case, still is) is it's FP64 precision, since anything that can do these type of calculations are in the stupidly high mark price range, at least so far, not even the ADA RTX 6000 is even close to it in these type of calculations, since it's going directly for a tesla gpu, and amd tends to have better rations on FP64 (XTX has 1.9 FP64 Tflops, while the 4090 has 1.2 tlops, titan V 7.6 Tflops), but their support is limited since a lot of apps are cuda exclusive, while hip and zluda are still on the works. It's a very specific use case, but for those that need it, so far only the A100 tier of gpu's can satisfy that need, or getting and old titan V does wonders.
Nowadays there are GTX Titans 6gb that can be found for 80-100$. Its msrp was 999$ in 2013. I've been thinking about getting one of these just to put it on a shelf. Great stuff for collectirs considering its current price
I got both a Z (retired now) and a V (still in daily use). For FP64 CUDA compute without the need for ECC, even at full retail price they were really good value compared with the alternative quadro / tesla cars.The marketing for them was a bit more problematic, but if you actually needed their unique features the price actually made a whole lot of sense imo.
HBM still exists, but it has been proven to not be sturdy enough for consumer graphics cards and would make maintenance and repairs very expensive, since any vram issues would require replacing the GPU instead of just the RAM chips like on regular cards.
Legendary card...came on the market and everyone forgot about it as it was an oddball and insanely expensive :D I think the Volta core itself found it's way into datacenters instead.
yep. the titan v was nvidia's way to dump the dies with only 3/4 the memory bus functional, the datacenter versions had the full 4096bit bus and 16 or 32GB of HBM
just saw one for sale near me for a good price. already have a founders 3080 ti so definitely not a necessity but would love one for the collection. perfect timing with this being uploaded 3 minutes ago haha
Kitty is so cute. It would be pretty cool for you to upgrade all the thermals and see how well it can OC. See if there's more in the tank. A good show case for Kryosheet too :p
Battlefield 2042 has an in game FPS monitor, complete with CPU & GPU frame time & sim time. As well as graphs. You can turn them on in game console by pressing "~". You can stumble your way through by type "fps" & seeing all the options. You can find the proper commands online as well.
I have Titan V with waterblock and shuntmodded with 2x5mohms resistors for both power plugs. It results with about 525W max power limit instead of stock 300W (75*2 + 150*2 + 75W). At 2100@1.093v card can draw 450W in gaming. Highest power draw I saw was in GPUPI in CUDA mode (PL2 state unlocked in nvinspector) and it was 500-525W with slight power throotle. I did direct comparison with 3090 in metro exodus DX11 at 5k. 3090@1900mhz was about 40% faster than Volta@2100Mhz and 1000mhz HBM2 (58 vs 41FPS).
I've got the exact same setup, bought it with shunt mods already installed. I'm running a custom fitted 280mm AIO to keep temps under control (40-50C under full load, hotspot around 70-75C) What sort of temps do you get? I wish I could find a waterblock available somewhere.. love this card, apparently the only one in my country. Super rare!
Honestly the increase in performance on the 4070 is amazing in some games like cyberpunk, but relatively minimal in games that were around the same time this was released. It goes to show how well developers optimize games for the current gen of hardware.
WOW! I still have 3 of them. I wad crazy mining back then. They paid for themselves back then. 😅 Oh, and its performance gets way better when you use liquid metal on it and replace the junk thermal pads with Thermalright pads. 👍🏾
Would be cool to see if you can find a waterblock (or try fit an AIO if the HBM height isn't horrific) and shunt mod it, it was a financial mistake, but maybe you can try limit the "damage" by getting some more fun from it :) Clock was in the 1500 range a lot of the time, if it's like Pascal/Turing then should be able to get around 2GHz maybe with good cooling/power limit.
This is really interesting. It’s amazing how capable it is so long after release. Accounting for inflation the $3k price in 2017 would be $3.8k today. I wonder how current $2k cards will compare to $600 cards in 6 years from now.
Still have one, Pretty sure the cooler is the exact same heatsink (inside) as 1070/1080/1080Ti. They become temperature limited pretty fast, you really have to crank the fans up to 70% or so to get the most out of it without going totally deaf.... That said if you think that's loud go and compare to a Tesla V100 in a datacenter rack. I briefly had it watercooled which makes all the difference, but went back to air cooler for more compatability (so if anyone needs a spare waterblock..) In gaming they're about the same as a 2080Ti which isn't surprising since Turing was based on Volta (they took away the compute and added a token amount of ray tracing). rtx2080Ti: 4352sp x2 x2Ghz = *17.5Tflops* FP32 TitanV: 5120sp x2 x1.7Ghz = *17.4Tflops* FP32 (in single precision and lightweight tasks like games you'll run at higher clocks, unless you're temp/power limited like in the video, I have seen 2Ghz on mine under water and at the start of air cooled tests) However the FP64 double precision is miles different. rtx4090: 16384 x2 x2.6Ghz /64 = *1.3* Tflops TitanV: 5120 x2 x1.41Ghz /2 = *7.2* Tflops I've sold my rtx3090 but keeping the TitanV, with current prices I nearly bought another for NVlink (back when nvidia would let you buy multiple cards rather than force you to buy the new rtx5090 Ti super extreme for £4000). Edit: also with a 815mmSq die (plus HBM2) it cost Nvidia a fortune to make these cards. Now Nvidia is back to the far cheaper ~600mmSq dies using GDDR memory but still charge a fortune.
I won't lie... I really like seeing Der Cat. He's so fluffy and loveable, and he looks at GPU like he wants to do some serious gaming. Anyway, speaking of GPUs... Using a 2080 right now and it's surprisingly good even today. Sure, it struggles a little with shadows sometimes, and ray tracing... But still a solid card for 1080p.
New games and new cards usually benefit from driver optimizations. Whereas new games don't get optimized drivers with old cards so much. Games that have been out for awhile will most likely still show optimizations, even if they haven't been updated for these old cards in awhile. Comparing unoptimized, but compatible games, does give insight into how much the raw performance is.
I was surprised to hear this described as "aging well" with performance like that. My 5700xt is only 1 year newer than this thing (based on how old you said it was), and it seems to have aged much much better; especially for the $400 I paid for it on launch. It's still kicking and they go for like, $100 used now. I'd say my $400 1070 aged better too, but it died after 2 years and don't seem to exist anymore. My 290X is still kicking somehow, too. And don't get me started with the amount of 580s my friends have that are still running the newest games somehow. I still remember when the older Titans were shown to me as a reason to regret buying the 290X lol. I can't believe anyone ever bought these things. Even the 4090 somehow manages to be a better value proposition..
I saw one in a shop last year and was curious, but I'm not into collecting so the price was too much for me. I don't recall exactly how it was marketed at the time but I don't think it was ever targeted at gaming. It would be more interesting to see it in workloads it was intended for vs gaming GPUs of the time and today.
I still have my 1080 TI water cooled GTX card and it still kick ass but it is starting to show its age which for 6 years is normal and actually good because it's still kicking butt!
I'm curious to see more about the Titan RTX nowadays, being the only Titan released since RTX became a thing. From memory it was basically just a 24gb version of the 2080Ti with a little bit more performance, but I don't think many people bought it.
It also had the workstation-specific optimizations that the Titan series was known for. Sadly Nvidia then discontinued the Titans and replaced it with the RTX 90 series, which completely lack the workstation features that made the Titans such a good prosumer option.
Yea, a friend got one for cheap(at the time) at his job and it still runs most new fine at 1080p if you lower the settings a bit. It was truly a beast of its time
for viewers: if you're into older card, may I suggest getting the 1080ti 11gb which still valid today (and perhaps another few more years to come) use the almost latest graphics drivers and install the software to adjust the card to always run at 50% fan speed on idle, and boost up on higher clock speeds.
1:14 One of the largest consumer tech retailers in my country had it in stock so I'm sure at least a few people bought them with the intention to just game.
This card is HIGHLY desired for distributed computing. Smokes just about everything released so far at much lower wattages. My teammates are snapping up every one they can find on Ebay. The FP64 performance and Tensor cores get put to great use and out performs every current and older consumer Nvidia gpu.
Now, for around $200 to $250, you get the GTX 1080 Ti which is almost identical in terms of power to the Titan V and I think it's a good deal for an X99 motherboard. I think a lot of people still have the Nvidia 10 series for its ability to run modern games without too much problem, even if you can't set all the settings to full.
Every 1000 series I have encountered broke =\ I got a 1070 and it died in 2 years. My coworker got a 1070 and it died in 2 years (exact same model from MSI. He bought his a month later, and it broke a month after mine). My sort-of-coworker/friend got a 1080ti and it broke after crashing in Gears 5. I am surprised any 1000 series are left. In my house we have AMD cards going back to the ATI days, but for some reason all these Nvidia cards keep breaking. Not sure what it is, but I always have to buy a new AMD after an Nvidia breaks. It's been a constant plague for 12+ years of PC building.
I enjoyed that retrospective. If you want to run it more quietly, mount it in a vertical orientation in a case like the Silverstone FT05. I'd be interested to see your perspective on the Titan X (being a glorified 980ti).
I bought one of these a few years ago for $700 on ebay because I never had $3k to buy a new one. Coil whine was INSANE but performance was on par with the results you got. I’m a sucker for LEDs and since the Titan V logo didn’t light up, I swapped back to my Titan Xp which was only slightly worse in performance.
A head to head with some of the Radeons would be interesting. Instinct mi25, mi50, mi60 and Radeon Pro VII. Radeon's from the vega 20 generation are cheaper than the Volta card so that's why I'm suggesting them.
I got a small collection of cards! This is the 1 like to add but only at a way cheaper price so it'll probably be years before i can add! My favorite out of all the cards i own is probably evga 980ti kingpin! Its such a beautiful card
Awesome video, love it. As others in comments mentioned this card might not great for gaming but for compute especially FP64 its still a king, as such if it comes to it would be awesome to see some tests for compute work that benefits from it. Especially as it is the last relatively consumer oriented card that has FP64 performance, the only other cards with double precision unlocked are older Nvidia teslas or the the new A100 which is just incomparable for the price reasons. I do wish nvidia started doing again the FP64 unlocked cards, with them putting it on top of what used to be a quadro lineup, that way we get great cards for compute, without the data center oriented shrouds.
What really gets me with that GPU is the die size. 815mm^2. This thing is a similar size to the H100. I used to own one. It was in a system that also had a Titan Xp installed. Work provided the Volta and I sadly had to give it back, but I bought the Xp with my own money and used it up until last last year when I bought a 7900XTX.
I checked GPU-Z side by side with my 2070S, titan V has 50% more ram, 50% more ROPs, 100% more TMUs, 100% more shaders, 45% more memory bandwidth for a 50% bigger die size but 2070S has 50% higher clock speeds. It's funny how even the numbers are.
It's always fun looking back at what used to be mind blowingly fast hardware. I have a 4090 and Ryzen 7950x3d....the best you can buy today...it will be fun in 5-10 years to look back.
Good - now I'd like to see if you could mount a waterblock on it and shunt mod the thing to crank those clockspeeds ;) I wonder if the cooling and the fact that it's a professional card is massively holding back the clocks on this.
best looking GPU ever made imo. I actually tried to buy a new one a few months ago for $700 for collection purposes off of ebay. I got scammed but I quickly got my money back luckily. This video is making me want to try again lol.
If I don't remember wrong, I think this card was faster than the 2080 Ti that had not been launched yet in rasterization. So owners of this card had the best of the best for quite a while until the RTX titan came out.
Worth mentioning one was designed for workstation use and not gaming. I'm sure if you have one from your workstation days or still use it for a workstation but need to do some mid-to-high end gaming. It still performs well. Buying a new GPU a 4070 or 4070 super hands down.
actually, if i had paid 3000€ .. and a game like cyberpunk ran in single digets today, i would not call it "held up pretty well". I would hope that the 4090 holds up a little better by 2028 (it was released by the end of 2022?)
@@raresmacovei8382 Yes, but did he not mention it to be on the cusp of consumer/professional hardware? That means that some people probably DID buy this hardware for gaming. I mean, lets be very,, very honest .. no gamer needs a 4090 neither .. you can perfectly enjoy 100% of the games with a much, much cheaper GPU, yet, those things are sold out.
Could do Cyberpunk at 5120x1440 with all med-high at a decent 30-50fps on my 1080Ti back then, if that's of any help. No resolution scaling. EVGA FTW3 card, at.. was it 127% power limit, and under custom water.
The same card that showed that it does not need RT cores for ray tracing. With the largest chip for a serial card (815 mm) and a beautiful PCB, with an incredible cost of $3000 and which is weaker than the 2080 Ti in games. Now it's a museum piece for an Nvidia card collector.
Could we have a follow-up with this card under water? Feels like it's should have some more performance under all that heat seeing as the memory is sharing the hotspot
I had many issues with mine for gaming, I think Fallout 4 never worked, my monitor which used DisplayPort Multi-Stream Transport never worked. For my work it was a beast however, the full double precision is still top tier. I just went back to 1080Ti. DLSS should have worked since it has tensor cores but they never enable it
Did you do the DisplayPort firmware update that series cards had? Least I'd imagine this one probably got it as well, least all the rest of the 1000-series cards got it. Just curious.
I think tensors on this are "too dumb" to make DLSS work, as they don't support 3/4 data formats/structures newers do (which would force DLSS to use standard shaders vs. tensors in such cases).
I immediately went on eBay to see if I could pick one up. Found some as parts only and was still tempted just so I could have it in the collection. Must be something about the 10 series in Gold...
The blower cooler design sadly only looks good but performs horrible. I've built a used parts PC a while ago and got a founders edition 1080 ti for it and just had to replace the blower with something else since it basically constantly hit thermal limits. For a little over 100$ in total I gave it a NZXT G12 and a 280 AiO and it's now running so much quieter and cooler.
I’m still using 5700xt cause of design, love it. Undervolted it, applied liquid metal on gpu and washers on screws. With 100watts tdp and 1750mhz core it heats up to 55c
My blower XFX 5700XT constantly goes to 200 degrees. Was hitting its thermal limit of 230 on launch (before they did a driver update to lower the clock). My ASUS 7800XT refuses to go past 140 degrees even under a full load. I can't get it to heat up past my CPU lmao. Crazy how bad blowers can be overall. These ASUS triple fans don't even turn on til 131 degrees, but the blower fan was always running.
I feel like the Cyberpunk fail is likely connected to the slow memory? I wonder how hard the memory can be overclocked and how much could that help with modern titles.
I am looking for the StarWars editions, but I idc if they are broken, I really just want to mount them on my GPU wall. They were some of my favorite cards, next to the 1080Ti
Damn those clocks are low. It'd be interesting to see this thing under water. You think most Pascal blocks would fit? The performance is closer to my OC'd 1080 than I expected.
I will never forgive Nvidia for locking Titan V users out of the RTX pipeline, yes the Titan V doesn't have RT cores, but it DOES HAVE tensor cores, quite a lot of them actually, and tensor performance was estimated close to 2080 Ti at the time of it's release. The Titan V COULD use DLSS 1/2 and DLSS 3 -> FSR 3 frame generation, but because Nvidia locked it away of the RTX pipeline just on the software side of things, then it can't.
I'd be wiry of buying these cards just as I would buying any HBM memory GPU after the mining craze and this card lived to see 2. When HBM memory dies I think that's it for the core of the GPU basically. This card may have been a beast of mining.
i modded a VEGA 56 hbm2 i fell in love with HBM memory with a super wide range of uses as well as being able to change its insides to suit the software meaning HBM2 could be emulating DDR6 DDR5 DDR4 DDR3 etc...
The reason why people use the titan V is for AI accelerated tasks. You can NVLink them, and use them co-dependently the large memory. This GPU should outperform any of the lower end 4000(Including higher end.) series cards on AI applications when stacked.
Task that in with some of the pro-grade that is locked with nvidia's software, The titan V is unlocked on the professional side granting access to numerous nvidia featuresets at a low 800$ cost.
I have one of those*
*the cat not the graphics card
i have a similar older model, silver not ginger.
my friend has a ginger though; i highly recommend both!
I have a 6 month old one. Ginger unlocked and overclocked😂😂.
@@Enozenim_LJO guessing with more brains cells .
@@marinipersonal Did you know they make autonomous "full self roaming" versions of these now? Saw one in the park today - mindblowing!
😂
Cat, are you ok?? Blink twice if you are being held at gunpoint 😨🥺
😂
Lol what you mean? That cat was living it's best life
Its an orange cat. Waiting his turn in line for brain cell access.
@wolfstorm5394 Was?! What happened, good lord I hope the cat is okay.
@@CommandLineVulpine LMAO don't panic, it's not like that
Cat videos - the real content I signed up for.
that cat gets whatever it wants
The true reason why Internet exists
talk about a scene stealer!! Probably an AMD fan.
Steve’s content is garbage there’s no cats!
@@wahidtrynaheghugh260 Not true! Steve's cat has most definitely been in his videos!
... just not so much after they got an actual studio, and later the current school-studio 😅
This is also the first card in Nvidias lineup to have tensor cores. It's a very interesting card that was quite a bit faster than those titan X small and big p's.
I thought maybe you could use DLSS on one of these, which is one of the "modern features" worth upgrading for, but a quick Google search tells me Volta has an older tensor core design that is good for AI research, but can't leverage DLSS.
True, the HBM is what always most interested me, MCD or cache dies outside the core, link with a one layer hbm2/3/4 feels like it'd be the most impressive advancement we could have for feeding P Cores moving forward, while keeping chiplet latency low. @@ChristianStout
Anyone happen to know what its eth hashrate was? Just for posterity? Gonna look up if it beat the radeon 7 just for infos sake, but I have a feeling it might be hard to find that info
@@hansolo631 it was gigantic and unmatched.
@@hansolo631no, its not hard to find that info)) i think that info was literally on Linus channel?
0:52 Cat - "Why did you stop scratching? Resume immediately!"
Still one of the coolest looking cards ever along with the silver 1080Ti and the black Titan X Pascal
The 1080ti Kingpin is gorgeous
Respect Star Wars Titan Xp - Galactic Empire and Jedi Order
no.
@@jarredgallaher2643 Copper is beautiful.
Titan X Pascal and newer generation of Titan XP never received a black badge like Titan Black. I had 4x Titan X Maxwell in SLI and world first 16K video resolution gamer and pioneer on RUclips and 2x Titan X Pascal in SLI and updated 4x Titan XP in SLI.
You maybe talking about color but every Titan were black before Titan V or Titan RTX.
I won one from Nvidia 6 years ago... absolutely love it, though changing the thermal paste (and livestreaming the process) was extremely stressful! 🤣
get a glass box to present it
Thank you for this, I'm owner of one and was thinking how it stack. Few things to note :
1) I always undervolt mine, since default power/temp limits force lower clock in long test than with it. In My case, I settled for 0.8V @ 1620MHz via Afterburner V/F curve (going too low on voltage prevents card from reaching 3D clocks).
2) Power usage gets solved with undervolt
3) Some programs will be locked to 1350-ish MHz core clock, and that's "CUDA Compute Mode" thing. To disable throttling under it, you need to go into NV profiler and manually diasable that (this setting may be reset after driver update)
3) You don't have to remove the whole heatsink to see/repaste GV100. Window part of the cooler can be removed with 4 screws to clean heatsink (the same as on Pascal/Maxwell/Kepler "FE" cards that use similar design), while heatsink removal only needs 4 more.
Those blower coolers just don't cut it. At the price of any Titan I would absolutely custom cool one.
Wait, are you talking about an orange tabby cat or GPU?
@@theduck17 I hope it's about the cat. After all, cats that are capable of running Nvidia's proprietary software, and also can be overclocked and undervolted, are something worth noting. AMD and Intel really need to take lessons from the cat, if what I said was true.
@@ffwast Yea theres a reason why they don't use then anymore it probably made sense when the card was made though
I hear they also make use of their large VRAM buffer for intensive workloads paired with their wide memory bandwidth just means users have a lot of options when feeding it FISH (Float Integer String Handouts) @@jamesbrendan5170
Volta was never meant for gaming and that's why it never saw mass release in consumer market. I remember one of my friends had procured it for Machine learning(Artificial neural network training) and they sure found the price justified and his firm recovered the money within a year or so of purchase.
ChaitanyaShukla2503
Quadro Bridge, why he called it consumer card ?
Was he not around back then ?
Only Direct sale card from Nvidia, tot in German shops.
You friend did College, you ?
Recording the money within a year (considering its price) it's quite a very long time for a company.
kitty is enjoying the tummy rubs
It depends on your use case, if you are doing FP64 work loads than this card is about 5x faster than a 4090. Also you could find SLI boards with 4 slots back then so theoretically you could have 20x the FP64 performance in one rig for 600$ more than a 4090. So for people that are doing that kind of work at 600$ it is actually a really great price.
Very much so. If my problems didn't need more VRAM and ECC memory wasn't a must it would have been a steal even at full price. The Quadro GV100 I use is still available new today at even higher than original MRSP - if you need active cooling and display outputs there's no better GPU than the Titan V/GV100 for FP64 CUDA, otherwise you're onto the A100
@@pamelawhitfield4570If ECC is a must, and you want moar VRAM, you could get a bunch of P6000 cards.
radugrigoras
You need the Quadro bridge to stack the mem, SLi support ?
4x Sli on X99 i did, waste of money ! Maya only i guess.
this is assuming SLI actually works the way the marketing purported it to. which it patently did not. which is why it failed and was forgotten. this comment is laughable.
@@lucasrem SLi doesn't originally stack Ram nor is it meant to, it is to increase the output resolution of the viewport, not the scene complexity or texture ram. The texture has to be replicated across to every GPU for it's TMU to access and map correctly, the compositer of the output comes last in the chain and has nothing to do with the textures other than compositing the stream as it comes out from the TMU. That's what used to irritate me to no end when I read magazines in the "SLi era" because they used to add the TRam if the cards up as if that number had a meaning - it never did. If you SLi connect 4x4Gb cards you still have a 4Gb card, not a 16Gb card. You're adding together the rasterizing and resolution "power" of the GPU's but not their texture ram.
There were multiple modes used but actual SLi does not "stack" anything other than output resolution, that's it. I ran a quad SLi set for a while, two 7950Gx2's bridged, and ended up having to try to explain what it is/does and isn't/doesn't do many times to people mostly when I upgraded to a single 8800Ultra I think it was, it was so long ago I forget now. My 4x 7950's with 512mb each did not result in a 2Gb GPU bridged, it resulted in a 3.5 times faster 512mb GPU, even that is kinda 'wrong' - it resulted in a 7950 with 3.5 times the output resolution capability of a single card - mostly. It's not 100% apples to apples that way either, but lightyears closer to ram stacking them pretending it's a 2Gb card in some form.
Yay! Another cat video on the weekend for the win🎉! On a related note, the cost of the Titan V is worth a lot of supplies of tuna🤭.
I’m glad you’re looking into this card. I was wondering for quite a while now how well it holds up.
the main selling point of this gpu back then (and depending of use case, still is) is it's FP64 precision, since anything that can do these type of calculations are in the stupidly high mark price range, at least so far, not even the ADA RTX 6000 is even close to it in these type of calculations, since it's going directly for a tesla gpu, and amd tends to have better rations on FP64 (XTX has 1.9 FP64 Tflops, while the 4090 has 1.2 tlops, titan V 7.6 Tflops), but their support is limited since a lot of apps are cuda exclusive, while hip and zluda are still on the works.
It's a very specific use case, but for those that need it, so far only the A100 tier of gpu's can satisfy that need, or getting and old titan V does wonders.
It's FP16 is also shockingly good for its time. Basically the first real AI enabled card.
@@Wobbothe3rdexactly, this thing was purpose-bred to be a machine learning monster, its ability to play games was secondary
Nowadays there are GTX Titans 6gb that can be found for 80-100$. Its msrp was 999$ in 2013. I've been thinking about getting one of these just to put it on a shelf. Great stuff for collectirs considering its current price
I remeber that 999$ was really outrageous price for GPU 😂 Good old times
Wow… it’s 7 years ago…
I still remember myself watching videos religiously abt it
why wow, should you forget it ?
In my opinion the older founders edition cards looked really nice. Those old blower cards, even though they're loud and run hot, look really cool.
I have a Titan Z in my display case, I love the design of those cards.
I got both a Z (retired now) and a V (still in daily use). For FP64 CUDA compute without the need for ECC, even at full retail price they were really good value compared with the alternative quadro / tesla cars.The marketing for them was a bit more problematic, but if you actually needed their unique features the price actually made a whole lot of sense imo.
sad HBM did not stay I loved how small the R9 nano was used it for years, they are getting stupidly big
Oh, it's still around. Just not for us mortals. 😅
HBM still exists, but it has been proven to not be sturdy enough for consumer graphics cards and would make maintenance and repairs very expensive, since any vram issues would require replacing the GPU instead of just the RAM chips like on regular cards.
Legendary card...came on the market and everyone forgot about it as it was an oddball and insanely expensive :D I think the Volta core itself found it's way into datacenters instead.
That’s what Volta was originally for if I’m not wrong. It’s why there wasn’t any GeForce Volta Cards which is kinda of a shame
yep. the titan v was nvidia's way to dump the dies with only 3/4 the memory bus functional, the datacenter versions had the full 4096bit bus and 16 or 32GB of HBM
just saw one for sale near me for a good price. already have a founders 3080 ti so definitely not a necessity but would love one for the collection. perfect timing with this being uploaded 3 minutes ago haha
00:52 cuteness overload!
Kitty is so cute.
It would be pretty cool for you to upgrade all the thermals and see how well it can OC. See if there's more in the tank. A good show case for Kryosheet too :p
channels all have Cats, you need that here ?
Battlefield 2042 has an in game FPS monitor, complete with CPU & GPU frame time & sim time. As well as graphs. You can turn them on in game console by pressing "~". You can stumble your way through by type "fps" & seeing all the options. You can find the proper commands online as well.
Ogh boy, been eager to see revisit of this old beast for ages!
You completely missed the point of that gpu, try with fp64 workloads and you will see the difference. It also is competitive in fp16
I have Titan V with waterblock and shuntmodded with 2x5mohms resistors for both power plugs. It results with about 525W max power limit instead of stock 300W (75*2 + 150*2 + 75W). At 2100@1.093v card can draw 450W in gaming. Highest power draw I saw was in GPUPI in CUDA mode (PL2 state unlocked in nvinspector) and it was 500-525W with slight power throotle.
I did direct comparison with 3090 in metro exodus DX11 at 5k. 3090@1900mhz was about 40% faster than Volta@2100Mhz and 1000mhz HBM2 (58 vs 41FPS).
This guy contributes more to Global Warming than entire nations.
Why would you ever consider it??
niiiice !
I've got the exact same setup, bought it with shunt mods already installed. I'm running a custom fitted 280mm AIO to keep temps under control (40-50C under full load, hotspot around 70-75C)
What sort of temps do you get? I wish I could find a waterblock available somewhere.. love this card, apparently the only one in my country. Super rare!
@@user-kx6om4xm5v Either for FP64 workloads or purely for fun. For me it was the latter.
Honestly the increase in performance on the 4070 is amazing in some games like cyberpunk, but relatively minimal in games that were around the same time this was released. It goes to show how well developers optimize games for the current gen of hardware.
WOW! I still have 3 of them. I wad crazy mining back then. They paid for themselves back then. 😅
Oh, and its performance gets way better when you use liquid metal on it and replace the junk thermal pads with Thermalright pads. 👍🏾
Would be cool to see if you can find a waterblock (or try fit an AIO if the HBM height isn't horrific) and shunt mod it, it was a financial mistake, but maybe you can try limit the "damage" by getting some more fun from it :) Clock was in the 1500 range a lot of the time, if it's like Pascal/Turing then should be able to get around 2GHz maybe with good cooling/power limit.
ayyyy. got one for cheap recently off a server farm. Loaned it to my friend to play Baldur's Gate before he got his new card. Pretty solid.
Would love to see for science how the Titan RTX holds up today!
Did the temperatures improve once the card when new TIM was applied? 🙂🤷♂️
Also where is the channel with additional cat content? 😁
This is really interesting. It’s amazing how capable it is so long after release.
Accounting for inflation the $3k price in 2017 would be $3.8k today.
I wonder how current $2k cards will compare to $600 cards in 6 years from now.
Still have one, Pretty sure the cooler is the exact same heatsink (inside) as 1070/1080/1080Ti.
They become temperature limited pretty fast, you really have to crank the fans up to 70% or so to get the most out of it without going totally deaf.... That said if you think that's loud go and compare to a Tesla V100 in a datacenter rack.
I briefly had it watercooled which makes all the difference, but went back to air cooler for more compatability (so if anyone needs a spare waterblock..)
In gaming they're about the same as a 2080Ti which isn't surprising since Turing was based on Volta (they took away the compute and added a token amount of ray tracing).
rtx2080Ti: 4352sp x2 x2Ghz = *17.5Tflops* FP32
TitanV: 5120sp x2 x1.7Ghz = *17.4Tflops* FP32
(in single precision and lightweight tasks like games you'll run at higher clocks, unless you're temp/power limited like in the video, I have seen 2Ghz on mine under water and at the start of air cooled tests)
However the FP64 double precision is miles different.
rtx4090: 16384 x2 x2.6Ghz /64 = *1.3* Tflops
TitanV: 5120 x2 x1.41Ghz /2 = *7.2* Tflops
I've sold my rtx3090 but keeping the TitanV, with current prices I nearly bought another for NVlink (back when nvidia would let you buy multiple cards rather than force you to buy the new rtx5090 Ti super extreme for £4000).
Edit: also with a 815mmSq die (plus HBM2) it cost Nvidia a fortune to make these cards.
Now Nvidia is back to the far cheaper ~600mmSq dies using GDDR memory but still charge a fortune.
I won't lie... I really like seeing Der Cat. He's so fluffy and loveable, and he looks at GPU like he wants to do some serious gaming.
Anyway, speaking of GPUs... Using a 2080 right now and it's surprisingly good even today. Sure, it struggles a little with shadows sometimes, and ray tracing... But still a solid card for 1080p.
New games and new cards usually benefit from driver optimizations. Whereas new games don't get optimized drivers with old cards so much. Games that have been out for awhile will most likely still show optimizations, even if they haven't been updated for these old cards in awhile. Comparing unoptimized, but compatible games, does give insight into how much the raw performance is.
Marvelous looking card
Deserve to run inside high airflow case and cold room
would have been nice to see some memory speed and bandwidth comparisons :(
I was surprised to hear this described as "aging well" with performance like that. My 5700xt is only 1 year newer than this thing (based on how old you said it was), and it seems to have aged much much better; especially for the $400 I paid for it on launch. It's still kicking and they go for like, $100 used now. I'd say my $400 1070 aged better too, but it died after 2 years and don't seem to exist anymore. My 290X is still kicking somehow, too. And don't get me started with the amount of 580s my friends have that are still running the newest games somehow. I still remember when the older Titans were shown to me as a reason to regret buying the 290X lol. I can't believe anyone ever bought these things. Even the 4090 somehow manages to be a better value proposition..
I saw one in a shop last year and was curious, but I'm not into collecting so the price was too much for me.
I don't recall exactly how it was marketed at the time but I don't think it was ever targeted at gaming. It would be more interesting to see it in workloads it was intended for vs gaming GPUs of the time and today.
Shunt Mod!! Would be interesting to see how it would do with more power and cooling.
I still have my 1080 TI water cooled GTX card and it still kick ass but it is starting to show its age which for 6 years is normal and actually good because it's still kicking butt!
for 200$ would be a nice vintage setup
can you do some benchmarks on the cat too? maybe a comparison video next to a non orange cat?
my titan Xp is still holding it down in 2024! its finally showing its age though, im looking to upgrade finally.
I'm curious to see more about the Titan RTX nowadays, being the only Titan released since RTX became a thing. From memory it was basically just a 24gb version of the 2080Ti with a little bit more performance, but I don't think many people bought it.
It also had the workstation-specific optimizations that the Titan series was known for. Sadly Nvidia then discontinued the Titans and replaced it with the RTX 90 series, which completely lack the workstation features that made the Titans such a good prosumer option.
Yea, a friend got one for cheap(at the time) at his job and it still runs most new fine at 1080p if you lower the settings a bit. It was truly a beast of its time
best buy i make so far. here on my machine still strong and fun to play.
for viewers: if you're into older card, may I suggest getting the 1080ti 11gb which still valid today (and perhaps another few more years to come) use the almost latest graphics drivers and install the software to adjust the card to always run at 50% fan speed on idle, and boost up on higher clock speeds.
Excellent review. We also enjoyed the video card. :P We love Sheik!
I love the new format of your channel - the cat videos
1:14 One of the largest consumer tech retailers in my country had it in stock so I'm sure at least a few people bought them with the intention to just game.
This card is HIGHLY desired for distributed computing. Smokes just about everything released so far at much lower wattages. My teammates are snapping up every one they can find on Ebay. The FP64 performance and Tensor cores get put to great use and out performs every current and older consumer Nvidia gpu.
Now, for around $200 to $250, you get the GTX 1080 Ti which is almost identical in terms of power to the Titan V and I think it's a good deal for an X99 motherboard.
I think a lot of people still have the Nvidia 10 series for its ability to run modern games without too much problem, even if you can't set all the settings to full.
Every 1000 series I have encountered broke =\ I got a 1070 and it died in 2 years. My coworker got a 1070 and it died in 2 years (exact same model from MSI. He bought his a month later, and it broke a month after mine). My sort-of-coworker/friend got a 1080ti and it broke after crashing in Gears 5. I am surprised any 1000 series are left. In my house we have AMD cards going back to the ATI days, but for some reason all these Nvidia cards keep breaking. Not sure what it is, but I always have to buy a new AMD after an Nvidia breaks. It's been a constant plague for 12+ years of PC building.
Isn't that also the first card to feature Tensor cores?
I enjoyed that retrospective. If you want to run it more quietly, mount it in a vertical orientation in a case like the Silverstone FT05. I'd be interested to see your perspective on the Titan X (being a glorified 980ti).
Your content is really interesting. Keep it up!
I bought one of these a few years ago for $700 on ebay because I never had $3k to buy a new one.
Coil whine was INSANE but performance was on par with the results you got.
I’m a sucker for LEDs and since the Titan V logo didn’t light up, I swapped back to my Titan Xp which was only slightly worse in performance.
A head to head with some of the Radeons would be interesting. Instinct mi25, mi50, mi60 and Radeon Pro VII. Radeon's from the vega 20 generation are cheaper than the Volta card so that's why I'm suggesting them.
I got a small collection of cards! This is the 1 like to add but only at a way cheaper price so it'll probably be years before i can add! My favorite out of all the cards i own is probably evga 980ti kingpin! Its such a beautiful card
one thing is for sure- it looks incredible
Awesome video, love it. As others in comments mentioned this card might not great for gaming but for compute especially FP64 its still a king, as such if it comes to it would be awesome to see some tests for compute work that benefits from it. Especially as it is the last relatively consumer oriented card that has FP64 performance, the only other cards with double precision unlocked are older Nvidia teslas or the the new A100 which is just incomparable for the price reasons. I do wish nvidia started doing again the FP64 unlocked cards, with them putting it on top of what used to be a quadro lineup, that way we get great cards for compute, without the data center oriented shrouds.
What really gets me with that GPU is the die size. 815mm^2. This thing is a similar size to the H100. I used to own one. It was in a system that also had a Titan Xp installed. Work provided the Volta and I sadly had to give it back, but I bought the Xp with my own money and used it up until last last year when I bought a 7900XTX.
I checked GPU-Z side by side with my 2070S, titan V has 50% more ram, 50% more ROPs, 100% more TMUs, 100% more shaders, 45% more memory bandwidth for a 50% bigger die size but 2070S has 50% higher clock speeds. It's funny how even the numbers are.
It's always fun looking back at what used to be mind blowingly fast hardware. I have a 4090 and Ryzen 7950x3d....the best you can buy today...it will be fun in 5-10 years to look back.
Good - now I'd like to see if you could mount a waterblock on it and shunt mod the thing to crank those clockspeeds ;) I wonder if the cooling and the fact that it's a professional card is massively holding back the clocks on this.
Did you see the graphics card hitting 90 degrees quickly and didn't even remember to change the processor's ceramic paste???
best looking GPU ever made imo. I actually tried to buy a new one a few months ago for $700 for collection purposes off of ebay. I got scammed but I quickly got my money back luckily. This video is making me want to try again lol.
A Friend, an OSRS youtuber, had one back in the day.
I think he got it from a sponsor, though.
I can’t believe how old this card is now, I remember when this came out it blew my mind how powerful it was
I though everyone forgot about this card. Good idea to bench it to see where it lines up these days.
If I don't remember wrong, I think this card was faster than the 2080 Ti that had not been launched yet in rasterization. So owners of this card had the best of the best for quite a while until the RTX titan came out.
Worth mentioning one was designed for workstation use and not gaming. I'm sure if you have one from your workstation days or still use it for a workstation but need to do some mid-to-high end gaming. It still performs well. Buying a new GPU a 4070 or 4070 super hands down.
actually, if i had paid 3000€ .. and a game like cyberpunk ran in single digets today, i would not call it "held up pretty well". I would hope that the 4090 holds up a little better by 2028 (it was released by the end of 2022?)
This wasn't a gaming/consumer GPU.
Also, if CP2077 was ran in RT mode, it makes this somewhat unfair.
@@raresmacovei8382 Yes, but did he not mention it to be on the cusp of consumer/professional hardware? That means that some people probably DID buy this hardware for gaming.
I mean, lets be very,, very honest .. no gamer needs a 4090 neither .. you can perfectly enjoy 100% of the games with a much, much cheaper GPU, yet, those things are sold out.
Could do Cyberpunk at 5120x1440 with all med-high at a decent 30-50fps on my 1080Ti back then, if that's of any help. No resolution scaling.
EVGA FTW3 card, at.. was it 127% power limit, and under custom water.
@@raresmacovei8382 it wasn't, but people still bought it for that purpose lol
The same card that showed that it does not need RT cores for ray tracing. With the largest chip for a serial card (815 mm) and a beautiful PCB, with an incredible cost of $3000 and which is weaker than the 2080 Ti in games. Now it's a museum piece for an Nvidia card collector.
Could we have a follow-up with this card under water? Feels like it's should have some more performance under all that heat seeing as the memory is sharing the hotspot
This was an excellent video not only for the information, but the amount of cat getting attention was perfect. More cat!
Ah yes, Ngreedia, once again, proving that their products aren't worth even a small fraction of what they charge for them.
I had many issues with mine for gaming, I think Fallout 4 never worked, my monitor which used DisplayPort Multi-Stream Transport never worked. For my work it was a beast however, the full double precision is still top tier. I just went back to 1080Ti. DLSS should have worked since it has tensor cores but they never enable it
Did you do the DisplayPort firmware update that series cards had? Least I'd imagine this one probably got it as well, least all the rest of the 1000-series cards got it. Just curious.
@@Nightykk I didn’t do it, as I swapped back to my 1080Ti which didn’t have issues and left the Titan in computation workstation.
I think tensors on this are "too dumb" to make DLSS work, as they don't support 3/4 data formats/structures newers do (which would force DLSS to use standard shaders vs. tensors in such cases).
I don't sell anything because I rarely don't have the money to buy what I want. I love it.
I'd love to see a video with this card with liquid metal, water cooling and a power mod
Best & last GPU to support native 3D Vision.
I immediately went on eBay to see if I could pick one up. Found some as parts only and was still tempted just so I could have it in the collection. Must be something about the 10 series in Gold...
We have a cat right from the beginning, its all fluffy and gpus are all good.
The blower cooler design sadly only looks good but performs horrible. I've built a used parts PC a while ago and got a founders edition 1080 ti for it and just had to replace the blower with something else since it basically constantly hit thermal limits. For a little over 100$ in total I gave it a NZXT G12 and a 280 AiO and it's now running so much quieter and cooler.
I’m still using 5700xt cause of design, love it. Undervolted it, applied liquid metal on gpu and washers on screws. With 100watts tdp and 1750mhz core it heats up to 55c
My blower XFX 5700XT constantly goes to 200 degrees. Was hitting its thermal limit of 230 on launch (before they did a driver update to lower the clock). My ASUS 7800XT refuses to go past 140 degrees even under a full load. I can't get it to heat up past my CPU lmao. Crazy how bad blowers can be overall. These ASUS triple fans don't even turn on til 131 degrees, but the blower fan was always running.
I feel like the Cyberpunk fail is likely connected to the slow memory? I wonder how hard the memory can be overclocked and how much could that help with modern titles.
The cat is so cute oh my god 😍
I am looking for the StarWars editions, but I idc if they are broken, I really just want to mount them on my GPU wall. They were some of my favorite cards, next to the 1080Ti
Damn those clocks are low. It'd be interesting to see this thing under water. You think most Pascal blocks would fit? The performance is closer to my OC'd 1080 than I expected.
I will never forgive Nvidia for locking Titan V users out of the RTX pipeline, yes the Titan V doesn't have RT cores, but it DOES HAVE tensor cores, quite a lot of them actually, and tensor performance was estimated close to 2080 Ti at the time of it's release. The Titan V COULD use DLSS 1/2 and DLSS 3 -> FSR 3 frame generation, but because Nvidia locked it away of the RTX pipeline just on the software side of things, then it can't.
I'd be wiry of buying these cards just as I would buying any HBM memory GPU after the mining craze and this card lived to see 2. When HBM memory dies I think that's it for the core of the GPU basically. This card may have been a beast of mining.
i modded a VEGA 56 hbm2 i fell in love with HBM memory with a super wide range of uses as well as being able to change its insides to suit the software meaning HBM2 could be emulating DDR6 DDR5 DDR4 DDR3 etc...
ive been using emulation to discover more uses
Got my VEGA 56 to expand to 20Gb of HBM2
thanks for showing us this cards performance and value today
"Hi and welcome back to another cat video." Best start to a graphics card review, ever. Big respect on your love of cats. 🙂
The cat is the real content of this video.
nvidia's old cooler design was so good, man. i like the new ones too but those old 700, 900, etc series cards just looked cool as hell
The reason why people use the titan V is for AI accelerated tasks. You can NVLink them, and use them co-dependently the large memory. This GPU should outperform any of the lower end 4000(Including higher end.) series cards on AI applications when stacked.
Task that in with some of the pro-grade that is locked with nvidia's software, The titan V is unlocked on the professional side granting access to numerous nvidia featuresets at a low 800$ cost.
How well does it holdup against non-NVIDIAs today like AMD and Intel. The CEO was a very limited edition.
I would like to see this beast with a custom cooler or with a waterloop overclocked balls to the wall with a 400W power limit!