Titan V Will Not Get NVLink OR SLI

[Drak3]

1 hour ago, SpaceGhostC2C said:

I would also like to know what kind of benefits do people expect to get from making a Volta-based gaming card

Some people are hoping Tensor Coreless Volta will be more efficient than Pascal, and that that will somehow automatically turn into better clocks.

[JediFragger]

SLi is dead, just doesn't know it yet. Powerful single card + Gsync/Freesync (when AMD can do it) = Gaming Nirvana.

[Drak3]

5 minutes ago, DeadEyePsycho said:

People are actually buying the "Titan isn't for gaming" despite the webpage having "GAMING" dead center?

The Titan V marketing material makes zero mention of gaming, and the product listing only appears under AI and Deep Learning.

 

The Xp does though.

[bcredeur97]

so whats the point of the Titan V then?

i mean it's a compute flagship right? it should have all the compute stuffs

[Bit_Guardian]

1 hour ago, SpaceGhostC2C said:

 

 

Yes, making a useless option available can be misleading, as in making a product seem fit for a purpose it's not good at (or terrible value at). Like selling tanks with baby seat mounts could be misleading.

 

I don't know what a court has to do with anything.  Things don't need to be lawsuit material to be true. It's as silly as saying "Titan XP, Xp, XpP, XPPPPP a confusing naming scheme? Good luck arguing "confusing" at a court!".

Except it's not useless in this case.

 

Actually arguing confusing in court happens a lot, as there actually is science to back argumentation based on it.

 

45 minutes ago, VegetableStu said:

oookay, why?

Forced upgrades, just as it's pure greed that you can't SLI a 980TI and a 980 despite the fact they're the same architecture, whereas AMD has always allowed this in XFire.

[Bit_Guardian]

30 minutes ago, bcredeur97 said:

so whats the point of the Titan V then?

i mean it's a compute flagship right? it should have all the compute stuffs

It does have the double precision and the memory bandwidth.

[SpaceGhostC2C]

7 minutes ago, Bit_Guardian said:

Except it's not useless in this case.

It's not useless in the tank either. But it's only useful for the people buying the wrong tool for the job.

7 minutes ago, Bit_Guardian said:

Actually arguing confusing in court happens a lot, as there actually is science to back argumentation based on it.

You have a problem with sets. Being true is more general than being upheld in court. The fact that something can appear in court discussion does not mean that it only exist in the court context

 

Things can be misleading, things can be confusing. A small fraction of the times whether they are or not can have legal consequences, and only then it becomes lawsuit material.

[Bit_Guardian]

9 minutes ago, SpaceGhostC2C said:

It's not useless in the tank either. But it's only useful for the people buying the wrong tool for the job.

You have a problem with sets. Being true is more general than being upheld in court. The fact that something can appear in court discussion does not mean that it only exist in the court context

 

Things can be misleading, things can be confusing. A small fraction of the times whether they are or not can have legal consequences, and only then it becomes lawsuit material.

No, it's definitely useless for the tank. This is a perfectly fine gaming card, for way too high a price.

 

No, as I'm a pretty solid mathematician, and in my field set theory is kinda the core of my work. It IS true, but the thing about court is there are rules for argumentation, which is why there are objection mechanisms. When you actually debunk arguments in court, you prove fallacies. You actually PROVE that you are correct, you just don't do it quite so formally as a written proof in geometry.

 

And I'm sorry but it's neither deceptive nor confusing. Either you find it useful or you don't. You can make perfectly informed, reasoned decisions about the product. If you decide to not use your head as a consumer, that's your fault.

[bcredeur97]

30 minutes ago, Bit_Guardian said:

It does have the double precision and the memory bandwidth.

True. But I mean the typical customer who buys compute stuff usually buys more than 1 no? Clustering is a big thing nowadays...

[Bit_Guardian]

2 minutes ago, bcredeur97 said:

True. But I mean the typical customer who buys compute stuff usually buys more than 1 no? Clustering is a big thing nowadays...

True, but for workstations...eh... Besides, Quadros are coming for the big kids.

[Prox1ma]

I do not think this is a card for gamers, and I am unsure how many of this is going to get sold anyway. Probably people buying this don't need SLI or nvlink

3 hours ago, Bit_Guardian said:

I sincerely hope AMD's Navi brings them to heel.

I sincerely hope this doesn't become "Wait for Navi" again

[Syntaxvgm]

I see! 

It costs 3,000$ so no one can afford two, therefore they don't waste their money on a second card that's not useful to them. 

[Subtle Corruption]

Ordering 6 with dip and fries.

[Derangel]

1 hour ago, Bit_Guardian said:

Forced upgrades, just as it's pure greed that you can't SLI a 980TI and a 980 despite the fact they're the same architecture, whereas AMD has always allowed this in XFire.

SLI and Crossfire are two different technologies. SLI was never designed with that ability in mind. Plus, it's a really shitty solution in CFX. Why would you want to intentionally hamper a stronger card and make it run at the same level as a weaker one? That is such a terrible idea. The only place that ability is even worthwhile is in laptops equipped with both onboard graphics and dedicated GPU. On desktops it's a horrible idea. You sacrifice power and scaling. AMD doesn't even talk about that ability anymore when they bother to discuss CFX, even they've abandoned it as an important feature.

[Bit_Guardian]

28 minutes ago, Derangel said:

SLI and Crossfire are two different technologies. SLI was never designed with that ability in mind. Plus, it's a really shitty solution in CFX. Why would you want to intentionally hamper a stronger card and make it run at the same level as a weaker one? That is such a terrible idea. The only place that ability is even worthwhile is in laptops equipped with both onboard graphics and dedicated GPU. On desktops it's a horrible idea. You sacrifice power and scaling. AMD doesn't even talk about that ability anymore when they bother to discuss CFX, even they've abandoned it as an important feature.

NVLink, however, was.

 

And no, SLI is actually superior since it leverages both PCIe and the secondary connector. There is actually higher bandwidth in SLI.

 

And actually AMD has not abandoned CFX. If you read their white papers on their HPC APU, CFX over IF is in fact a very huge selling point to them.

[Derangel]

2 minutes ago, Bit_Guardian said:

NVLink, however, was.

 

And no, SLI is actually superior since it leverages both PCIe and the secondary connector. There is actually higher bandwidth in SLI.

 

And actually AMD has not abandoned CFX. If you read their white papers on their HPC APU, CFX over IF is in fact a very huge selling point to them.

I was talking about AMD abandoning making a big selling point out of the ability to CF different models of cards. At least in the desktop market.

[leadeater]

16 minutes ago, Bit_Guardian said:

And no, SLI is actually superior since it leverages both PCIe and the secondary connector. There is actually higher bandwidth in SLI.

 

And actually AMD has not abandoned CFX. If you read their white papers on their HPC APU, CFX over IF is in fact a very huge selling point to them.

Actually SLI bandwidth is rather poor, even more so around the time AMD/ATI decided to drop their dedicated connector in favor of the PCIe bus.

 

Quote

Nvidia has 3 types of SLI bridges:

Standard Bridge (400 MHz Pixel Clock[2] and 1GB/s bandwidth[3])
LED Bridge (540 MHz Pixel Clock[4])
High-Bandwidth Bridge (650 MHz Pixel Clock and 2GB/s Bandwidth[5])

2GB/s isn't exactly a lot, more headroom in the PCIe bus.

[jasonvp]

4 hours ago, DeadEyePsycho said:

People are actually buying the "Titan isn't for gaming" despite the webpage having "GAMING" dead center?

This specific Titan is not for gaming.  And no "GAMING" is not dead center on its product page.  The word "GAMING" doesn't appear anywhere on its page, in fact.

[GOTSpectrum]

7 hours ago, JediFragger said:

SLi is dead, just doesn't know it yet. Powerful single card + Gsync/Freesync (when AMD can do it) = Gaming Nirvana.

Not just that, but SLI and Crossfire are both DX11 technologies. DX12 can utilize different GPUs from even different manufacturers simultaneously. The old way of linking GPUs really needs to die off and make way for the new universal standards.

[FloRolf]

6 hours ago, Bit_Guardian said:

Forced upgrades, just as it's pure greed that you can't SLI a 980TI and a 980 despite the fact they're the same architecture, whereas AMD has always allowed this in XFire.

Pretty sure you are wrong here. 

AMD allows the same GPU's to be CF'ed, not the same architecture. 

So you could CF a HD7970 with a HD 7950 but not with a HD7770 (very old example even if it doesn't feel as old) 

[Bit_Guardian]

5 hours ago, leadeater said:

Actually SLI bandwidth is rather poor, even more so around the time AMD/ATI decided to drop their dedicated connector in favor of the PCIe bus.

 

2GB/s isn't exactly a lot, more headroom in the PCIe bus.

Again, it's been proven SLI uses both the bridge and PCIe itself.

 

1 hour ago, FloRolf said:

Pretty sure you are wrong here. 

AMD allows the same GPU's to be CF'ed, not the same architecture. 

So you could CF a HD7970 with a HD 7950 but not with a HD7770 (very old example even if it doesn't feel as old) 

No, it allows the same architecture. Any GCN1/2/3/4 card can be XFired with any GPU from the same gen. The SKU (die) does not have to be a match.

[leadeater]

51 minutes ago, Bit_Guardian said:

Again, it's been proven SLI uses both the bridge and PCIe itself.

Yes but the SLI bridge gives very little gain, using it and actually having any benefit is different. All the SLI bridge does is send the final rendered image to the master GPU which is why the resolution and frame rate matters for which bridge you use.

 

It's been shown that even for the most high end GPUs in SLI and Crossfire that even in x8 link mode the bandwidth is not fully utilized and has no effect on frame rate.

 

The SLI bridge back in the day made a lot of sense when communication either went through the chipset or you were doing 3+ way SLI on PCIe 1.0 x8 which was only 2GB/s so putting everything through that was a real limitation.

 

PCIe 3.0 x16 has 15.75GB/s of bandwidth and only a very small amount of that is used for a GPU in gaming, there is no need for a bridge.

 

So I don't know how using it has any relationship to actually being useful. I think you are mistaking this with legacy, as in it's this way because it was in the past and there has been no trigger to change it since doing so requires development work both on the hardware side and drivers.

 

Far as the industry goes standardization is a very big focus so it's not going to be long until Nvidia removes the SLI bridge altogether, probably the next gaming GPUs wont have them. This is why AMD is using PCIe for the IF rather than developing their own transport layer, if there is already a high bandwidth low latency transport technology avaliable it makes little sense to develop something new unless your requirements exceed that i.e. NVLink.

 

As for the Titan V lacking NVLink that doesn't stop people from using two if they wish, there is a performance impact but it doesn't make two in a single system unusable. The better question is why you would want two of these in one systems, you'd be better off buying one and getting everything you can done with that then use the $3,000 you saved on renting time on a cluster.

https://devblogs.nvidia.com/parallelforall/dgx-1-fastest-deep-learning-system/

[leadeater]

55 minutes ago, Bit_Guardian said:

No, it allows the same architecture. Any GCN1/2/3/4 card can be XFired with any GPU from the same gen. The SKU (die) does not have to be a match.

Spoiler

 

Spoiler

http://support.amd.com/en-us/kb-articles/Pages/Crossfire-Chart.aspx

 

 

[Sniperfox47]

Why are people bitching about the price...? This is *MASSIVELY* cheaper than any of their Tesla cards, and is definitely a budget option for its intended workload... 3000 dollars vs well over 10000 dollars, for nearly the same performance? Sign me up!

 

If you're a home user and want to waste money, just do so on a Titan Xp. For gaming purposes, you'll be getting the same, or even better, performance than the Titan V counterparts.

 

@Princess Cadence I can think of a few workflows where NVLink would come in handy with these cards. Not so much for deep learning, but moreso for large scale data processing where you have to load lots of data back and forth between the GPUs.

 

Honestly, as much as cutting cost by ommiting the extra PCIe/NVLink hardware, I'd say it's probably cut out to avoid canabalizing their data center markets. This thing is really cheap, even compared to a Pascal based Tesla.

 

 

[Princess Luna]

Just now, Sniperfox47 said:

If you're a home user and want to waste money, just do so on a Titan Xp. For gaming purposes, you'll be getting the same, or even better, performance than the Titan V counterparts.

I'm not sure about this after all it is 5000+ CUDA Cores vs 3840... that is quite the difference... sure though we'll need an AiO because as much as I find the FE cooler gorgeous it will be a limiting factor in overclocking the card to extract all you can from it.

 

The TITAN V will be faster than all we have right now in the Pascal line up true enough, but anyone getting it over a 1080 Ti is doing terrible money management regardless.

 

It is true that NVLink could still havei ts usefulness but then again as you said they can't cannibalize entirely their 10000$ Tesla.

 

If my brother is really as crazy as it seems I might actually be able to toy with one of these in a couple of months, he already put his two TITAN XP for sale #lmao.