3840 CUDA cores, 12GB of 11Gbps GDDR5X memory. Fully enabled GP102 like the Quadro P6000. Same price as the previous TITAN X Pascal.

TITAN X(2016) is discontinued since GTX 1080 Ti launch, TITAN Xp(2017) just replaces it outright at the same price point.

Exactly what we need, another overpriced card

As long as it says “limit 2 per customer” (indicating NVIDIA is able to sell all they can make), it cannot be “overpriced” at that price.

I would think that the availability of this GPU means that process yields have improved, or that NVIDIA has been able to stockpile enough cherry-picked parts from prior-months production to make an additional SKU worthwhile.

That’s like saying gambling/buying lotteries are rational behavior, because people are doing it. It’s not the first time Nvidia has used artificial scarcity and planned obsolescence to pump up the price and it won’t be the last.

specific types of gambling and/or lottery ticket buying can be rational behavior, at least mathematically. Occasionally, a lottery will produce a ticket-buying scenario where the expected value of the ticket is greater than the ticket price.

Likewise, with appropriate techniques, in some situations, gambling can result in odds that favor the gambler, as opposed to the “house”.

And declaring that your argument for whether something is “overpriced” is sound based on statements about “rational behavior” seems far-fetched. A very simple, commonly accepted principle about pricing is that the value of something is equal to the price that someone is willing to pay for it. Rationality of that behavior matters very little at the point the transaction is conducted.

The principle of unbacked currency (and even backed currency, depending on what it is backed by) is based (at least in part) on artificial scarcity. I’m surprised I should be mentioning this to the Fed chairman.

Overpriced simply means being charged for too high a price. By your definition a good can never be overpriced as long as some people, even one people is paying for that price. Even from the producer’s rational point of view going toward that extreme of the demand curve is limiting their own profitability and it’s damn well overpriced.

Nvidia’s price gouging is getting more predatory day by day as it solidifies its monopoly. Combine that with their well known behavior of sabotaging their old cards by driver “updates” and it should be obvious to even an outsider that we’re all doomed if AMD don’t get their sh*t together.

That just looks like a circulation definition, to me. It can’t be argued, so I won’t.

This seems to be an argument by appeal to extremes, which is also shaky IMO. The only way your objection makes sense is if, in fact, only one Titan Xp gets sold, or, if, in fact, the pricing results in some unfavorable position in the supply/demand curve. I don’t see how you could know either one of those.

This product, being introduced at exactly the same price point as the product it replaced, does not seem like it could be an example of that. It introduces more functionality/capability at the same price point. There is no day by day evolution in pricing here, from what I can tell.

Even if we survey the extreme graphics space, there have existed entries in this space at the $1000 price level at least since the original titan introduction which was approximately 4 years now (and arguably some dual-GPU products in this space are older than 4 years at the $999 pricepoint). The only evolution here has been (at most) from $1000 to $1200.


I believe we can find many examples in other product areas where prices for the high-end offering have increased 20% in 4-5 years.

For some this new Titan X is actually worth the price, particularly if most of the applications need to run on Windows (TCC driver mode).

Curious about the memory bandwidth of course.

You forgot to factor in technological progress, so replacing an inferior product of the same price doesn’t mean the pricing is necessarily getting less predatory.

The whole Titan line is premium priced and comparing Titan cards doesn’t reveal the problem. Let’s compare the relationship between Titan X (Maxwell) and Xp with their respective contemporary top-line consumer cards X80 Tis.

Titan X (Maxwell) vs 980 Ti: 29% less FLOPS/ Titan Xp vs 1080 Ti: 33% less FLOPS/

It’s very clear that the pricing of “extreme” graphics cards, which exploits more the information asymmetry between Nvidia and end users, is getting more predatory.

The fact that 980 Ti was released after Titan X (Maxwell) but 1080 Ti was released before Titan Xp only enhances this argument.

I didn’t make that claim. You used the words more predatory, and I don’t think replacing a product which is clearly better than the previous at the same price point is an obvious example of “more predatory”. Delivering technological progress at the same price point seems like a good thing. You seem to be casting it in a negative light, and I’m struggling to find any sense in that.

However, the value-based arguments seem much more sensible to me.

It’s true that Titan family products don’t deliver the same FLOPS/$ as other products.


  • This trend is not unique to NVIDIA. It’s common both inside and outside of the tech industry. Not all value metrics are perfectly linearly correlated to price
  • This doesn’t take into account that in extreme product spaces, consumers are often willing to pay disproportionately for incremental performance. Again not unique here. (** see below)
  • From a product design perspective, the cost to deliver FLOPs in the midrange of a product performance curve is not the same as the cost deliver FLOPs at one end of a product performance curve. The center is, usually, the cheapest differential cost for added flops. The first flop is usually quite expensive, so is the last. Restricting a product to perfect yielded chips is inherently more expensive than the product which can use any chip that has, at most, 1 defective SM.

** This doesn’t have to be entirely a boutique factor, either. From a programming complexity standpoint, it’s more difficult to parallelize work by breaking it into chunks and distributing, than it is to combine all the work into a single point of issuance. For example, it’s generally easier to make a matrix multiply run fast on a single GPU than it is to make it run fast on 2 GPUs, each of which delivers half the performance. As a result, even though the Flops/ is not the same, the GPU that delivers more performance in a single package is, for some cases, <b>more valuable</b> because it may eliminate some programming complexity and inefficiency, and for some problems may actually provide more delivered performance. Customers have known this for a long time, and so in many cases bought e.g. M2090 GPUs when M2075 delivered considerably higher Flops/

Yes, this is exactly the case in much of my medical imaging work. Even though two GTX 1080ti GPUs will theoretically outperform this new Titan Xp, I would prefer to use the Titan Xp to keep things simple and possibly more reliable. And again since most in the medical/dental field use Windows the fact that I can isolate a ‘compute only’ GPU from the OS makes it easier to partition the visualization functionality from the compute (1 GTX for visualization display, and one GPU compute only).

The problem is the benefit you get with Titan Xp over 1080 Ti is marginal, at best. Titan X (Maxwell) has double the memory on 980 Ti so many customers still chose it even after 980 Ti’s release. Titan Xp only has 1 more GiB memory over 1080 Ti (12 vs 11), and 15% more theoretical performance (which will become less when you factor in auto down-clocking at high temperature), and bandwidth bound algorithms are in the minority.

Or you can overclock the 1080 Ti +200MHz and call it a day :)

(1) price has nothing do with cost, but with what the market will bear (unless you live under socialism). The chairperson of the Fed should be well aware …
(2) yield curves are not linear (not sure whether they are Gaussian, but probably so)

That are two reasons that halo parts tend to sell at a premium, not just when it comes to graphics cards, but most kind of products. That doesn’t mean you personally have to buy them. I don’t buy $2000 bottles of wine, or drive a $100K car, etc. But apparently there is a market for such things.

Most buyers are well advised, in terms of bang/buck, to buy processors two speed grades below the top bin. But as CudaaduC points out, for some use cases the halo parts are “just what the doctor ordered”. If you think it is OK to use overclocked hardware for medical devices, there might be a government agency taking a heightened interest in your work/products :-)

Are you paying taxes? Do you live in a country where taxes are being spent on some form of entitlement programs? Congratulations! You live under some form of socialism.

There is also apparently a market for $500 Epipens. I hope you’re not under the impression that premium pricing not being restricted to one industry makes it an unshakable “fact of life”. Personally I rarely fall for such blatant mispricings, but that doesn’t mean I can’t point out such gross inefficiencies. If the invisible hand truly scratches everyone’s back there wouldn’t be a government whose capability is beyond that of a night-watchman state.

Most buyers are indeed well advised, either voluntarily or they just can’t afford the premium product. It’s the other buyers that Nvidia targets and exploits with their “invent problem and solve it with own product” routine.

And I am sure that paying a mark up up to 700% (P40 vs 1080 Ti) just to eliminate that once-in-a-decade-failure which may not even happen before the product is obsoleted by Nvidia themselves is a good investment.

Really funny you should being up the phrase “just what the doctor ordered”. Two-thirds of Americans see doctors who received kickbacks from pharmaceutical companies, sometimes prescribing riskier than normal or even downright unnecessary drugs. Are there also people who got paid by Nvidia to bully entrepreneurial but less-tech-inclined customers into choosing premium or “server-grade” products? We may never know :)

Interesting positions from the chair of the Fed :-) In your opinion, should companies be allowed to set prices for their products freely in a non-monopoly situation? If the answer is “No”, who should set prices instead?

Last I heard, the Epipen product has a monopoly, as the FDA has not approved any competing products. Which I find a bit odd because best I know the active ingredient is generic, while the delivery system seems very close, if not identical, to emergency injectors I know from my time in the armed forces in the early 1980s (designed to deliver an antidote to nerve gas).

By contrast, if you don’t like the price of the Titan Xp, for your compute needs you have the choice of about a dozen NVIDIA GPUs other than the Titan Xp, GPUs from competing companies, and non-GPU solutions such as CPUs or FPGAs. And nobody will die because they couldn’t afford to buy a Titan Xp.

I’m far from an advocate of socialism nor do I support government-led price policing. I also have no feud against monopolies.

Personally I yearn for an anarcho-capitalist paradise where companies set whatever prices they want for whatever services they provide. If they happened to be a natural monopoly, there’s no problem with that. It’s when they start charging distorted prices that other (profit seeking) private entities will step up in whatever ways necessary, some of which not possible nor legal under the current system.

I consider the current U.S. corporatocracy where the government serves corporate interests without oversight, the uninformed not weeded out and invisible hand “chained” a step backward.

Thanks for the clarification. The “anarcho-capitalist” angle explains it quite nicely. Seems to be the latest fad, second time in as many weeks that I have encountered this.

Just in case you really do incorporate overclocked consumer GPUs into medical devices or anything safety-critical, I’d appreciate a tip-off, so I know what to avoid.