Comments Locked

43 Comments

Back to Article

  • dgingeri - Thursday, August 10, 2017 - link

    Think of how much better they'd be if they hadn't crippled compute performance on the mainstream cards so badly.
  • Dr. Swag - Thursday, August 10, 2017 - link

    How did they cripple it?
  • 1mpetuous - Thursday, August 10, 2017 - link

    Mostly in artificially reducing FP16 and INT8 performance in the Geforce group, and not offering anything with remotely reasonable DP performance at all below Tesla.
  • 1mpetuous - Thursday, August 10, 2017 - link

    Sorry, I made an error. INT8 not restricted, just FP16
  • mapesdhs - Friday, August 11, 2017 - link

    I skipped the entire 600 series because of the borked CUDA performance. Nothing came along to consistently beat the 580 in the same price range until the 780 Ti. There's an AE thread on Creativecow full of people who "upgraded" to 600 series cards, and even some 700s, only to see no speedup at all or often a drop. They judged based on the number of cores, but the design was very different, lower clocks to help power delivery and other changes, especially the much lower ratio for FP64.
  • jwcalla - Thursday, August 10, 2017 - link

    I'm sure you know the business better than them. Maybe you should apply to be CEO?
  • StrangerGuy - Thursday, August 10, 2017 - link

    When a clearly luxury product is too expensive for the likes of dgingeri, regard said product as if it's his constitutional right.

    I find amazing the lengths people will go to hide their entitlement syndrome, just like how poor third worlders constantly whining about iPhone prices.
  • mapesdhs - Friday, August 11, 2017 - link

    Perhaps (I won't go into that, my MAGA hat is too much of a rant magnet...:D), but it is true that NV radically changed their support for compute on consumer cards in a way which did make it very difficut for prosumers on a budget to move up the scale in their capabilities. Presumably they did it because they thought that reasonable compute on consumer cards would hurt Quadro/Tesla sales, but that doesn't seem likely from the conversations I've had with relevant people; those who can afford Quadro/Tesla also tend to want the extra features they offer, such as ECC, better caching structures, full-speed PCIe return path, longer warranty, better MTBF, etc. Prosumers on a budget though had nowhere to go after the 580 for quite a while. NV also seemed to miss the potential for those users who, if they could, would use a GF to dev code at home while their employer would use pro cards for final deployment, eg. a guy using CUDA for financial transaction processing said final systems must have ECC, but it would have been nice to have something better than a 580 for home dev work.

    Ian.
  • WinterCharm - Sunday, August 13, 2017 - link

    Don't forget certified drivers! that's one of the main reasons to get Quadro and Tesla cards.
  • 1mpetuous - Thursday, August 10, 2017 - link

    Agreed. Most casual compute (students / academic research) customers just don't have the budget to step up to the Tesla cards at almost an order of magnitude greater cost.
  • frenchy_2001 - Thursday, August 10, 2017 - link

    1) casual compute don't usually need much fp64, but you do have a point that nv does not serve that market. Titan used to fill that niche, but after kepler, the low fp64 is a hw restriction (gm200 was not a compute chip and similarly, gp102 is visualisation, compute is gp100)

    2) Titan lately favor ai, or int8 operations
  • 1mpetuous - Thursday, August 10, 2017 - link

    Depends on your application. What I would have liked to see is segmentation by limiting the total number of DP compute cards per system via drivers, rather than pricing even a single Tesla out of reach.
  • mapesdhs - Friday, August 11, 2017 - link

    As I said above, NV did supply the FP64 market to a decent extent with the 500 series, the 580 being particularly good. But it wasn't just FP64 where CUDA performance took a dive with the next arch, check reviews of the 600 cards, the 580 hammers them for basically everything, likewise the 700s up to the 780. I agree with 1mpetuous, NV could have continued with the Titan-style driver mechanism at the very least, but that wouldn't have helped with the general drop which affects the 600s (note I'm not talking about games here, the 600s are obviously better than the 500s for that).

    Ian.
  • milli - Friday, August 11, 2017 - link

    It's not like NV could enable more compute power through drivers. NV took out huge chunks of compute power from its GF designs to lower power consumption.
    It has been one of the main reasons why AMD has higher power consumption but also has higher compute performance.
  • HighTech4US - Thursday, August 10, 2017 - link

    Why don't you ask AMD how well that is going for them.

    Nvidia (with their crippled compute performance on the mainstream cards) constantly generate positive (and growing) profits while AMD (with non-crippled compute performance on the mainstream cards) hardly makes any profit.

    Seems like Nvidia knows how to market their products better than AMD.
  • TheinsanegamerN - Friday, August 11, 2017 - link

    But he wants that extra stuff for FREE! Why should he have to pay more for a niche product, they should just give it away!

    dgingeri doesnt understand how market segments work.
  • dgingeri - Friday, August 11, 2017 - link

    Wow, step away and a little comment on their crippling compute performance becomes some major analysis on my ability to comprehend economics and market segments.

    Of course I understand that they felt (incorrectly) the compute performance would hurt their Tesla and Quadro sales, and that they felt it was good to use that to differentiate the segments better. Al lot of people don't seem to understand that with the use of PhysX, the full potential of the card can be accessed, and that this compute crippling hurt non-PhysX physics performance as well as a few other potential uses for Cuda for mainstream users, particularly cryptocurrency mining.

    I am annoyed by this, yes. It took away potential from my 980Ti, particularly that I can't even consider using my card for cryptocurrency mining in my off time, and any physics games might be able to use are locked down to Nvidia's proprietary PhysX, which costs money to license. AMD users have those other options.

    Yes, I do understand the decisions the execs at Nvidia made were done to make them more money. I consider such things, artificially reducing or eliminating features on lower priced items, underhanded tactics, as they can't sell the higher levels of items on legitimate reasons, and elitist, as it artificially relegates those who can't afford the highest end stuff to having inferior products.

    Imagine, if you would, if a car maker advertised "V-8 in every car" and yet 4 cylinders are disabled unless you buy the leather seats, 8 speaker sound system, and moonroof, and you'd have to buy the boat trailer, mobile home trailer, and detachable motorcycle to get more than half gas intake capacity. How well would you think of them for that?

    That's what I feel Nvidia is doing here.
  • CiccioB - Friday, August 11, 2017 - link

    You are saying a bunch of BS.
    Nvidia has "crippled" their GPUs in HW, not SW. They decided with Fermi that only x00 chip had to have DP units. With maxwell, GM200 didn't have them due to die size issues (GM200 was what had to be GM104 at never developed 20nm HP).
    There is not any crippling by driver, as you can use whatever Tesla or Quadro you want and you will get the same cryptocurrency numbers as you have with GeForce. So there's not artificial crippling.
    Moreover AMD has not any particular computing advantage in cryotocurrency calculations. With Ethereum, which is the algorithm used to make comparisons, the important thing are memory bandwidth and latencies, and, surprise, a RX580 with the same memory bus as the GTX1070 computes exactly the same.
    The fact that you consider AMD faster is just because they sell 580 cards at the price of the 1060, while actually mounting a GPU that is comparable to a 1070 in used resources.
    All thus implies that you are just speaking by some hate and not by reason.

    So you may understand economics and segmentation, but surely not technology. And if you do, you are just lying.
  • tipoo - Friday, August 11, 2017 - link

    How would we know?

    AMD crippled compute on consumer cards far less, but is doing worse. Nvidia tailored GPUs to games at the expense of compute performance, and more consumers bought in....
  • dgingeri - Friday, August 11, 2017 - link

    AMD as a whole is doing far worse. Their GPU division is doing quite well. Their inability to overtake Nvidia despite superior products is a direct result of their horrible driver writing. (Their drivers are buggy, and the interface on the control panel is poorly laid out and performs poorly. A LOT of people, including me, avoid AMD GPUs specifically because of that.)
  • JKflipflop98 - Friday, August 11, 2017 - link

    Please just stop. You obviously need to do more reading and less talking.
  • smilingcrow - Friday, August 11, 2017 - link

    If they'd given the consumer gaming cards more compute they'd have sold better to miners but worse to gamers.
    Mining is a lot less of a stable source of income than gaming as the whole thing could fall apart if the price falls through the floor for a long period.
    So I think Nvidia called it right and focused on a stable market and their results show that.
    If mining continues to make large profits Nvidia could design a chip just for that as they have the capital to do that.
  • dgingeri - Friday, August 11, 2017 - link

    I would rather have the option to do some minor cryptocurrency mining when my system is not being used for gaming. That's all. AMD users do have that option.
  • CiccioB - Friday, August 11, 2017 - link

    Nvidia user have too.
    A GTX1070 computes exactly as a 580 as they have both the same memory bandwidth and latency.
    The GP106 on GTX1060 that you think is crippled is actually a smaller, less energy hungry with slower and narrower memory controller which develops less TFLOPS than AMD Polaris 10.
  • smilingcrow - Saturday, August 12, 2017 - link

    Of course that is the choice, go for a better gaming card or a mining card.
    Do you prefer gaming or gambling?
    Pick your favourite sin then burn in hell, in glorious 3D! :(
  • TheJian - Sunday, August 13, 2017 - link

    ROFL. Stock up from $20-160 in very short time. To steal trumps words, you're fired :) As a business, they are doing EXACTLY what they should do. AMD should pay attention to how both Intel and NV PRICE their stuff and avoid price wars any time they can (unlike AMD, who seems to invite them).

    If they took your route, market segmentation destroyed, margins destroyed, net income destroyed. Oh and stock price canned. No, no point in that crap.

    Why would I care about compute in a gaming card? Oh you want both so you get a better deal? Not their job to be your friend. Sure I'd love that card too, but if I was them, I'd act exactly the same way because that is their job (to make as much as possible). Maximize profits from R&D.

    Cutting off crap most don't use at home is smart and produces more profit. Why sell 100% of the market die space that only 10% use? The numbers here are not exact but you should get the point. You want a PRO card? BUY one. You want mainstream pricing to go up and nvidia profit down? Ask them to act like AMD. You will see NV stock plunge and profit, R&D, etc all down with it. Why do you think AMD makes NOTHING (well, is MAKING a loss, making something?) even when launching a whole new product line? Even the new cpus look like they only got SOME pricing right. They came in with new GREAT tech, and then price war...ROFL. STUPID. They should have just barely undercut the enemy, which would have not gotten such a large response from Intel (chipset insta-death, kaby insta-death, massive price drops etc). Intel would have waited to see how much you could steal before responding if pricing was at least where they could still make good money. AMD would have maximized profits then, but instead they chose to force a price war...LOL. I'll say it again, management should be fired.

    As a customer, great pricing AMD...LOL. As a business...YOU'RE ALL FIRED MANAGEMENT! A full quarter into new cpu sales, still no INCOME...LOL. It should have been a few hundred million this Q and a billion next as server crap rolls out and 2B next year. Stupid. Hopefully they get the server stuff priced right. If intel's server chips' equivalent is $5000, yours should be $4500, not $2000 or whatever...ROFL. Heck if you chip is BEATING his in more than half the stuff you should be MORE than Intel's price! People PAY for perf, just ask Intel and NV how their high end stuff is doing...If you're better than the competition ALWAYS charge like it.

    One more point, only buy companies that get this crap. Then you don't care about pricing. My chips/upgrades are completely free this year (next year too...etc LOL). NVDA/MU (still have much MU) bought me many things this year and micron will likely be $60 in the next 2yrs and maybe even $100 as they move into stealing enterprise drives and at some point home drives massively too. WD just took their first hit, but those will keep coming to some extent for a while.
  • tuxRoller - Friday, August 11, 2017 - link

    58% gross margin! That's insane
  • BedfordTim - Friday, August 11, 2017 - link

    That is about right for a healthy business. That margin needs to cover investment in future products as well as profits.
    Other successful companies such as Intel operate on similar margins.
  • JKflipflop98 - Friday, August 11, 2017 - link

    What? Most businesses on Earth would KILL to have a 15% profit margin. KILL. Anything over a 50% profit margin is INSANE!
  • iwod - Sunday, August 13, 2017 - link

    Do some research yourself. Most business properly have more the 15% Gross Profit Margin, What you are thinking is likely Net Profit Margin.
  • tipoo - Friday, August 11, 2017 - link

    Yowza.

    They're up there with Intel and Apple. Guess that happens when you out engineer your competition by a margin.
  • tuxRoller - Friday, August 11, 2017 - link

    Intel's is around 62% and Apple is about 35%.
    It's obviously not about out engineering but simply what the market will bear.
    I'd imagine Zen will cut into those margins, but it remains to be seen if Vega is compelling enough to makeup for the massive difference in marketing budgets.
  • jabjab - Friday, August 11, 2017 - link

    Got your dates wrong on the tables, unless I've woken up today in the future and its already 2018
  • jabjab - Friday, August 11, 2017 - link

    read it fully now, my mistake
  • jabbadap - Friday, August 11, 2017 - link

    So is it safe to say nintendo switch is on their oem&ip category?
  • tipoo - Friday, August 11, 2017 - link

    Maybe, they seem to be pretty standard TX1 chips so it seems more likely in there than in Gaming.

    Either way, I don't think it would be a massive contributor here, it's probably low margin.
  • jabbadap - Friday, August 11, 2017 - link

    Well to reply to myself, it's not in oem/ip. That contains oem desktop and laptops and crypto mining crap. Switch is on tegra segment revenue.

    Source: http://files.shareholder.com/downloads/AMDA-1XAJD4...
  • Yojimbo - Friday, August 11, 2017 - link

    Switch is in Tegra on the one list and Gaming on the other. Crypto mining goes into OEM, which is why its revenue jumped so much. I don't know the product cut off for what they consider "gaming". Maybe anything below an x50 is not gaming?
  • DukeN - Friday, August 11, 2017 - link

    When you say earnings, perhaps you meant revenue.
  • Toadster - Friday, August 11, 2017 - link

    Anandtech has a time machine... it's 2017 :)
  • MadManMark - Saturday, August 12, 2017 - link

    Nvidia's fiscal year 2018 began on February 1, 2017. My company's FY18 began on July 1, 2017. The federal government's FY18 will begin on October 1, 2017.

    What Anandtech has is knowledgeable writers.
  • Ian Cutress - Saturday, August 12, 2017 - link

    First line:

    'NVIDIA announced its earnings this afternoon for the second quarter of their 2018 fiscal year (not a typo)'
  • TheJian - Sunday, August 13, 2017 - link

    Please fix the headline as EARNINGS does not mean REVENUE.

    "NVIDIA Announces REVENUE (fixed) Of $2.2 Billion For Q2 2018"

    Earnings=Bottom line. The difference between a company's earnings and its revenue is revenue is the top line amount of money the company makes by selling its product or service, and earnings are the bottom line profit the company makes after all costs of goods sold, expenses, taxes, depreciation, amortization and interest are paid.
    http://www.investopedia.com/ask/answers/070715/wha...

    I wish NV would have made this much money (amd too). That would mean some massive R&D and more great stuff in the pipeline. I really hope this was just a mistake, but if you really don't know the difference, someone else should be reporting on stocks/Q reports ;)

Log in

Don't have an account? Sign up now