No mention in the TPU or AT articles about this having ECC. Given the worry about corruption there has been with consumer TU102, I wonder how successful this will be in the semi-pro market where memory corruption could be a bit more serious than a quick break for some retro gaming.
coercitiv
Diamond Member
- Jan 24, 2014
- 6,795
- 14,852
- 136
Stuka87
Diamond Member
- Dec 10, 2010
- 6,240
- 2,559
- 136
Cards have failed with both brands of memory, and if it was simple corruption, rebooting would fix the issue as the buffers would all be cleared.
IntelUser2000
Elite Member
- Oct 14, 2003
- 8,686
- 3,786
- 136
I wanted to buy a used R9 290 from Craigs, so I went to this guy's house. He had GTA5 loaded on a 1440p monitor to prove the card works.
It looked beautiful. I'm used to 1080p monitors, and I didn't think I'd notice it beyond that. But some games look absolutely amazing at beyond 1080p resolution.
All the power to you if you can settle on a low resolution monitor. Then you can play all the games smoothly with a $250 GPU. My current monitor is a 19-inch with a max resolution of 1280x1024. Decent refresh rates at 75Hz though. It felt really different at first, but after a week or two I don't notice it anymore.
Just for refresh rates I do want a 144Hz monitor.
That Titan is a workstation card only (in a remotely sane world!). Slightly cheaper than Titan V & a bit faster except for double precision.
So confusing. Some experts on this forum told me that there was no price hike, and there was just a shift. That the Titan was just cancelled, and the real Titan is the 2080ti, and there was just miscommunication from Nvidia regarding the "rebrand". So now the price hike is real as the 2080ti is no longer the Titan?
Yep looks that way, at least in terms of matching products like for like starting at the top end of both product range and working your way down from the top. It's become more complex to compare these cards now, I can't easily do it in my head anymore from a spec point of view really, I dunno if there can meaningfully be equivalent cards from both ranges anymore.
sze5003
Lifer
- Aug 18, 2012
- 14,278
- 663
- 126
I don't think the 2080ti was ever meant to be a Titan. They knew they could charge that much for an actual rtx Titan and that's what they are doing now.
Yes, this hasn't changed the 'new' structure of the line up.
The Titan was due a few generations basically an overpriced gaming card - the 2080ti is still in that spot.
They'd changed a bunch of work station /deep learning etc specific things fur this Titan.
The Titan was due a few generations basically an overpriced gaming card - the 2080ti is still in that spot.
They'd changed a bunch of work station /deep learning etc specific things fur this Titan.
LTC8K6
Lifer
- Mar 10, 2004
- 28,520
- 1,575
- 126
Titan isn't marketed as a gaming card, though.
NV's page does say this about NVLINK with Titan:
SMOGZINN
Lifer
- Jun 17, 2005
- 14,225
- 4,461
- 136
It is a gaming card. It is the new Titan. Their marketing department is not stupid, they know that by naming it that they are announcing it as the new top end gaming card. They can talk out of both sides of their mouth, but they are intentionally positioning it as a gaming card with the name. If they truly did not want to have it viewed as the new top end gaming card they would have given it a different name. To those doing deep learning and research it would not matter what the name is.
It's their (half) generational replacement for Titan V. Did they push that as a gaming card?
A tiny bit but not really.
A tiny bit but not really.
LTC8K6
Lifer
- Mar 10, 2004
- 28,520
- 1,575
- 126
Calling it a gaming card, and then having NVLink behave differently seems odd. It also doesn't seem like it would have much of a gaming lead over the Ti?
My bad, reading the wrong Titan X Pascal line. Still, with both a shader and clockspeed advantage over the 2080Ti Titan RTX should at least match the performance advantage we had with Ti vs Titan for Pascal.
I don't believe that NVLink will work any differently on this one than it has on any other Turing or Volta GPU. You only get to use both memory pools separately in compute or specially written DX12, not off-the-shelf SLI.
LTC8K6
Lifer
- Mar 10, 2004
- 28,520
- 1,575
- 126
The quadro cards appear to pool the memory?
Multi-GPU Technology Support
NVIDIA NVLink High Speed Interface
Connect a pair of Quadro RTX 5000 boards with NVLink to essentially double the amount of GPU memory available, double CUDA, RT and Tensor Core counts, and scale application performance by enabling GPU-to-GPU data transfers at rates up to 50 GB/Sec.
Available Accessories
Quadro RTX 8000 48GB 96GB 10 GigaRays/sec 4,608 72 576
Quadro RTX 6000 24GB 48GB 10 GigaRays/sec 4,608 72 576
Quadro RTX 5000 16GB 32GB 8 GigaRays/sec 3,072 48 384
Quadro RTX 4000 8GB N/A 6 Giga Rays/Sec 2,304 36 288
https://www.pny.com/nvidia-quadro-rtx-5000
Multi-GPU Technology Support
NVIDIA NVLink High Speed Interface
Connect a pair of Quadro RTX 5000 boards with NVLink to essentially double the amount of GPU memory available, double CUDA, RT and Tensor Core counts, and scale application performance by enabling GPU-to-GPU data transfers at rates up to 50 GB/Sec.
Available Accessories
- RTX5KNVLINKX8S2RKIT provides an NVLink connector for the RTX 5000 suitable for standard PCIe slot spacing motherboards, effectively fusing two physical boards into one logical entity with 6144 CUDA Cores, 768 Tensor Cores, 96 RT Cores, and 32 GB of GDDR6 memory, with a bandwidth of 50 GB/Sec. Order PN RTX5KNVLINKX8S2RKIT when ultimate performance and capabilities are required with two RTX 5000 boards (application support required).
Quadro RTX 8000 48GB 96GB 10 GigaRays/sec 4,608 72 576
Quadro RTX 6000 24GB 48GB 10 GigaRays/sec 4,608 72 576
Quadro RTX 5000 16GB 32GB 8 GigaRays/sec 3,072 48 384
Quadro RTX 4000 8GB N/A 6 Giga Rays/Sec 2,304 36 288
https://www.pny.com/nvidia-quadro-rtx-5000
Last edited:
Right - but there is a caveat. It won't be pooled under normal DX11 style SLI. Your app has to support that to have it work like that, and unfortuntely almost no games have done anything other than DX11 SLI or DX12 in zero effort DX11 style implicit multi-adapter. Neither of those are going to add memory pools. Not that you would really need it anyways. Realistically you only need that large of a memory pool for compute and professional workloads, which should be supported.
LTC8K6
Lifer
- Mar 10, 2004
- 28,520
- 1,575
- 126
Gigabyte GeForce RTX 2060 OC pictured, features 1920 CUDA cores
Grain of salt but VC were reliable during turing leaks, 1920/2304 against the xx70 this time round to place it round 350-400 bucks (1070/ti/Vega56) and warrant the RTX buy-in tax (?)
Grain of salt but VC were reliable during turing leaks, 1920/2304 against the xx70 this time round to place it round 350-400 bucks (1070/ti/Vega56) and warrant the RTX buy-in tax (?)
- Jun 8, 2003
- 14,387
- 480
- 126
1920 Touring cores with fast memory should put in above the Vega 56 but slower than the gtx 1070ti on this chart. I'm guessing $350, thats not too bad.
Nice to finally see something. A cut down TU106 is pretty surprising. That's a large die for a a xx60 card, nearly the size to GP102/1080ti. Probably looking at performance a hair ahead of a 1070 (+5-10%) for $100 more (probably $399 MSRP). I'd be buying a 1070 at $300 or less if I was looking to upgrade.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 23K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads
- Started by Tigerick
- Replies: 18K
-
-
Discussion Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000, etc.)- Started by Vattila
- Replies: 13K
-
Facebook
Twitter