Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

I'm very confused by Nvidia's numbering scheme. It used to be generation-model, so my 1080 is generation 10, and "80" is an arbitrary model number where higher is better. It's the logical successor to the 9-80.

The 20-x cards I suppose are OK, a big jump to signify a big change in architecture.

But now we have... 16-60? Why 16? Is this the successor to the 1060? And it's a "Ti", but there isn't a non-Ti 1660?

I'm confused.



From NVIDIA [1]

> As far as naming goes, and why 16 series instead of just using 11? Quite simply, we felt that from an overall architecture and performance perspective, TU116 is closer to the other TU10x parts than it is to prior generation GP10x. TU116 has most of the Turing architecture features, including the new dedicated cores for INT32 and FP16 operations, and it also has all of the new Turing shading features, including variable rate shading and mesh shading. And as you see, performance is closer to the GeForce RTX 2060 than it is to the GeForce GTX 1060. In fact just like you said, it performs closest to the GTX 1070, beating it in some games and losing some others. So we ultimately settled on 1660 Ti instead of 1160 Ti. ‘16’ is closer to 20, after all.

[1] https://www.gamersnexus.net/hwreviews/3445-evga-gtx-1660-ti-...


So like USB2 becoming USB3, or Java 1.4 going to 5, or Thunderbird going from 3 straight to 5.


> So like USB2 becoming USB3, or Java 1.4 going to 5, or Thunderbird going from 3 straight to 5.

Or AngularJS 1.8 going to a complete fucking mess.


Haha. Or php 5.6 to 7


I was curious about this so I looked it up. Apparently there was a PHP 6, but it was scuttled and had some of it's features backported to PHP 5.3.


USB 2->3 doesn't really help your point, since 2 comes directly before 3 in integer math, so it (3) is literally a logical successor to 2.


IIRC after USB2 came out, there was "USB Full Speed" (12mbps) and "USB High Speed" (480mbps), then they renamed "USB 2.2 High Speed" it to USB3. USB3 is literally USB2.2 with a name change, so I feel like it fits.


Are you sure you don't mean that USB 1.2 was changed to USB 2.0? I don't recall any naming confusion between USB 2.0 (480 Mbit/s) and 3.x (5 Gbit/s and up).


Ah, that's probably what I was thinking of. Thanks for the correction!


...Netscape 4 to Netscape 6. Windows 8 to Windows 10. The i7-8086k processor.


Not having Windows 9 was for a very good technical reason, not because of marketing.

An unknown number of RegEx strings are looking for Windows 95 and Windows 98 just by looking for 'Windows 9'.


[dead]


Can't do that if you want to sell an overpriced 2050 Ti.


They probably want to reserve 20xx for RTX cards only.


It used to drive me crazy as well. Couldn't understand why my old 980 Ti card ranked better than many of the 10xx cards that came after it.

I researched this recently, and this article [1] did a great job clarifying things for me.

[1] https://steamcommunity.com/discussions/forum/11/627456486990...


> x80 - For high-end gamers, high budget card, for high/ultra settings and higher resolutions. x90 - These are pretty much just two x80 glued together with a custom cooler. For the insane users with more money than sense, overpriced / overkill card.

If only marketings gurus were as clear and BS free than this guys...


So, perhaps it's me, but why do folks care about the number on the part? It's all marketing. Why does it matter? What matters is performance, right?


Software is versioned in numerical order which is pretty intuitive. At a glance it's clear which versions are minor updates, and which are major.

Cars are versioned by year/model, which again makes it pretty clear to understand minor/major updates. Sometimes significant updates are introduced in a model year, but generally the core features remain the same and it could still be considered an upgrade to that model.

Without a clear and intuitive versioning scheme it can be confusing and time consuming to make sense of a product line. And that gets frustrating if it keeps changing.


>Cars are versioned by year/model, which again makes it pretty clear to understand minor/major updates. Sometimes significant updates are introduced in a model year, but generally the core features remain the same and it could still be considered an upgrade to that model.

Tesla managed to break this trend massively, which proves a problem for things like insurance. The feature set on (say) the January 2014 Model S is very different from the December 2014 Model S, even though they technically share the same "year".


Tesla didn't break this any more than any other car manufacturer. The model year hasn't been tied to the manufacture year for a long time. See https://www.autotrader.com/car-news/why-doesnt-cars-model-ye... for instance.


I think the difference is more "model year no longer reliably indicates feature set", not "model year no longer indicates date of manufacture."

This may make finding and pricing spare parts difficult, or categorizing safety and performance metrics.


> Without a clear and intuitive versioning scheme it can be confusing and time consuming to make sense of a product line. And that gets frustrating if it keeps changing.

This is a case of Nvidia working with that sentiment rather than against it.

They were roasted by review sites previously for co-mingling architectures in the same numbering generation. So this time, they didn't.

TU106 was far too big of a chip to die-harvest low enough for a true volume x60 budget part.

Add in all the non-graphics acceleration hardware that needed to be cut to hit price and... Nvidia didn't feel this could be called an RTX 20xx part.

The 16xx is awkward, but it's the least bad choice.


> Cars are versioned by year/model

Just look at BMW or Porsche. Their numbering systems are just as obtuse as GPUs.


BMW went that way very recently. It wasn't that long ago that their model numbers were very much explicit. A 328i was a 3-series chassis with a 2.8 liter. The i and d stood for fuel-injected and diesel respectively.


Well it still makes sense to some extent. i and d still mean petrol and diesel, with e for hybrid joining the ranks lately. But in general x1x(say 116d or 114i) are entry level engines, x2x(320d) are mid-tier and x3x(430d) x4x(240i) are higher end, more powerful engines.


That was fine back in the day when there was a correlation between engine size and performance.

Since they started strapping turbochargers to everything down to 1.0l, the engine size comparison has become less important. If anything my 3.0l car is seen as a negative because of the higher fuel consumption.

The manufacturers are just trying to walk a fine line.


No one does semantic versioning for hardware products (at least consistently), but maybe they should.


It's apple and oranges. Adobe doesn't release 5 versions of photoshop every year with different performance levels.


It makes it hard to know what I'm buying if I can't put it in a mental hierarchy.


I agree that it's not totally obvious to new customers, but it's also no that hard: within a generation, the relative performance of the cards corresponds to the order of model numbers. "ti" cards are more powerful than non-"ti".

Between generations, the whole line moves up something like one level of performance, so an (X)70 should be compared against an (X-1)80 and so on.

It's not the simplest thing, but I think most people will do research once the first time they buy a GPU, and then you have your mental model from then on.


No, the mental model I have from the 90's and the 00's doesn't help me in picking up a graphic card.

No worries though, I decided to settle on integrated Intel GPU. Good enough for 2 years old games.


Because it’s nice to be able to have some idea of what the product is without having to dig into a bunch of tech specs.


By naming it 1660 instead of 2050, there is a lot more distance between the RTX/tensor core enabled products and those that are not.

So a win?


Because it’s easy to compare X110 > X109.

Or if there’s a grouping like GP describes, you can quickly ascertain which model is the one you want.


> Because it’s easy to compare X110 > X109.

That's true, but how does it help you as a customer? Wouldn't you need to know what you needed? What is the benefit of buying an X110 over an X109?

https://dilbert.com/strip/2002-06-11

I don't like "it's just a number" schemes, because they look like transparent attempts to get you to buy something for no reason. What's the difference between 3G and 4G? Well, they changed the number. Does that mean anything? How would you know?


Which may not be the model marketing wants you to want.


More than 10, less than 20. The article says as much.


Like line numbering in BASIC. Leaves room to insert.


You're not the only one. I know a couple Engineers working there. No one understands the naming scheme :-)


These numbers have always been marketing driven. Just like with cars: is there any sense to the way BMW models are named?

In this case it's probably intended to invoke the GTX 660 in the minds of customers, since that was a very successful and well-loved card when it was released.


>Just like with cars: is there any sense to the way BMW models are named?

Not sure what you mean, BMW seems like the complete opposite of what we are discussing, the way they are named is extremely functional. It used to be simpler because they had less models but it's survived the changes fairly well.

For the majority of cars the first number (1 through 8) is the segment, from lowest to highest. The next two used to be directly the engine size but because smaller engines are getting more powerful they've disconnected that and it just means the performance level. They then have prefixes for different types of car (M for performance, X for SUV, Z for convertible), and suffixes for details of the drive train (e for hybrid, i for gasoline, d for diesel). There are a few more nuances but the bulk of the naming is this.

Their range isn't just sorted well in terms of naming, the interiors and features are highly consistent between cars as well so the naming allows you to get in a car and know pretty much exactly what to expect. If Nvidia was doing the same we'd know everything about this card's positioning in the market just from the name (does it have raytracing, what peformance level, what market segment, etc).


In the past BMW model names were correlated to the chassis type & the engine inside. So 530d would mean it’s a 5 series chassis, with 3.0 liter diesel engine and so on. These days, AFAIK, that’s no longer the case and model numbers are a bit arbitrary


Nvidia have used nonsensical model numbers for a long time. Remember the GeForce 4 MX?


Having to go back 15 years for an example is usually a good predictor of the strength (or better: weakness) of an argument.

The numbering has been very logical for more than a decade: you have a product generation, and increasing numbers mean increased performance.

What is so nonsensical about that?


What matters are the specs, not the generation/age of the card.


It’s the new 1080p card since the 2060 can handle 1440p


This is the reason I made a full AMD config ten years ago: I could understand the numbering scheme of CPU and GPU. Higher is better, and for processors more expensive models have at least the instruction sets supported by cheaper ones. With Intel not some much and I had to check too many CPU for virtualization instructions.


The Nvidia numbering scheme is still higher is better?




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: