Does NVIDIA really artificially butcher capabilities like FP64 of all consumer products to push their premium workstation hardware?
What is this scheme called? Are there no laws against it?
yes.
no idea.
no, why would there be?
All GPU companies do it, it's why they won't support open drivers.
If you buy a device with a powerVR gpu it won't support openCL unless the chip manufacturer paid extra for it to be enabled in the driver.
Nvidia and ATI managed to prevent much competition in GPUs using patents, so they can get away with it for the moment.
Most of their chips are also butchered versions of the full chip. But so are AMD's
>>53410383
Maxwell was gutted in order to artificially make it more efficient when running gaymes. This was mainly so Nvidia could crow about that fact and sell more gaymen cards, rather than a plan to sell more Quadros. It's also the reason that Maxwell a shit when it comes to DX12.
>>53410383
Nvidia, much like Intel is known for underhanded and shady business tactics to keep themselves market leaders. While AMD deserved to lose in the CPU market because of the fuckup that was Bulldozer, They lost ground in the GPU market through Nvidia's tactic of getting game developers to gimp AMD hardware. It even gimps Nvidias old hardware so you have to buy the current gen.
See https://www.youtube.com/watch?v=O7fA_JC_R5s
As such, it wouldn't surprise me if they're doing the same in the workstation market.
>>53410383
They don't. You fell for AMD marketing FUD.
>>53410655
got your fanboy blinders on there pretty tight huh? might be cutting off circulation to your brain.
>>53410668
>ad hominem
Is this the best you can do AMD poorfag?
>>53410645
Bulldozer was pretty good baka.
At worst i3 performance, at best faster than an i7.
Fx 6300 and 8320 are really good value.
Artificially? No, not even AMD does that.
Nvidia in specific did however cut out extra computational performance for lower power draw. That's how they did Maxwell.
>>53410701
>this is what AMDcucks actually believe
>>53410701
>faster than i7
I have a AMD CPU, but I don't think it's true. At best, it would compete with an i5.
It's called price segmentation and a perfectly valid business strategy taught in like the 3rd lecture of any marketing course at any university.
Even the average en/g/ineer should understand that consumers would have to pay higher prices if there was no price segmentation.
>>53410727
Not everything is about muh gaymes. An 8350 has similar multi-threaded performance to a 4790K in tasks that can use all eight threads efficiently.
>>53410727
I'm not saying it is because many applications don't use 8 threads, but the potential is there.
Crysis 3 benchmark is a best case scenario.
>>53410745
But that's bullshit, senpai. The Titan X cost the same as the original Titan, yet was gutted in terms of compute performance. Removing features and charging the same price has nothing to do with price segmentation.
>>53410695
>calls me out for ad homiem
>while using a strawman argument
nice try tho neighborino.
also, i'm currently running a 4770.
>>53410701
>>53410748
Only in heavily threaded benchamrks.
Let's be honest, most people who buy this price range of CPU's are buying them to play games, and that's where intel beat them. That's why they lost out. Hopefully Zen might make them competitive again but that remains to be seen.
>>53410756
It does if you want to sell Tesla cards
>>53410383
>do not run updates
>????
>PROFIT
Just install what's on the disk when you buy the fucking card and never update.
>>53410770
fx 6300 has aged significantly better than any second or third gen i3 while being slightly cheaper.
In newer well threaded titles it's as fast as the i5 2500 and even the i5 3470.
Directx 12 and vulkan will only make things better.
>>53410805
That may be true but it's already too late. AMD has already lost a huge chunk of the CPU market, and in the server market, it went from around 50% share to almost nothing because of bulldozer. It will take years for them to recover, if they ever will.
Not only that but confidence in the brand is at an all time low, so it's an uphill battle for them. I'd like them to succeed, but unless they can outperform Intels current gen for the same price and particularly power consumption for the server market, they won't succeed.
>>53410866
that's true. their share prices reflect that.
It just annoys me that bulldozer gets so much hate when it's nowhere near as bad as its made out to be. I've been really happy with my fx 8320, had it for 2 years now. 60 dollars cheaper than an i5 4460 last time I checked.
I'm not saying it's better than Intel, because it does have a lower ipc, but at the same time if you use applications that takes advantage of 8 threads then it's incredible value.
>>53410805
can confirm
the fx-6300 doing better than i3-2120 sandybridge scrubshit I bought
>>53410795
The Maxwell Quadro/Tesla cards can't do FP64 either, it's literally not in the chip architecture.
>>53411571
Adding on to that: the only nvidia cards that have FP64 capability are kepler or fermi based, with it being non-gimped in the "Big" variants of the architecture (GK110 and GF100 for example). Its why GK110 and GF100 based cards are still so expensive.
At this point if you need raw FP64 performance any AMD card based on GCN is the way to go as they all offer a fairly large chunk of FP64 (1/4th FP32 rate).
IIRC nVidia has never done FP64 that high.
>>53410645
>needs english translation
>>53410645
>2 thousand and sex
So ive got ~700 to buy a gaming graphics card to go with an i7-6700. I was thinking of getting the 980 ti? Is that a bad idea?
Stop lying about FP64
AMD GCN cards are just as crippled, 1/8th on Hawaii, 1/16th on Fiji
Fiji is SLOWER than 7970 in FP64, let that fact sink into your thick skulls
Maxwell can do FP64 just fine for programming and debugging FP64 code and getting stuff ready for Pascal FP64 performance
>>53412708
This. Hawaii and Fiji are the most crippled, since the entire ALU arrays are designed to do 1/2 rate FP64, whereas Maxwell has the strong majority of their ALUs built for FP32 only and a smaller block of them for FP64.
GCN/Haiwaii/Fiji are better GPGPUs, but you should go buy a FirePro or Tesla if that's what you're really trying to do.
I thought Nvidia was now shilling FP16, because why would you need 32 bits anyway?
>>53410610
No, they aren't butchered. Chips are likely to show problems during manufacturing.
Most of the times the faulty ones work well with some functionality disabled, so they sell it as a cheaper product. That's why there are lots of different GPUs based on the same chip.
>>53413230
half precision is so crippled that almost nobody will want to put in the effort to shoehorn it into their systems.
> 1b sign
> 10b significand
> 5b exponent
you can basically represent positive/negative numbers from 0.00006 to 65k with three digits of accuracy.
(fp32 = 7 digits precision with decimal shifted up to +/- 38 places, fp64 = 15.5 digits w/ 308 decimal leeway)
You have to REALLY build your system around shitty numerical simulation to think this is a smart tradeoff for only a 2x performance bump.
>>53413230
FP16 is only for deep learning, games want to run on FP32
>>53410383
>What is this scheme called?
Capitalism
>>53411449
probably not as fast as a third gen i5 but I'm rather confident it'll even outperform a haswell i3 4170.
Scores higher then an i3 4330 in anandtech.
>>53410383
You're wrong. Nvidia has moved from a company that makes consumer GPUs to a company that makes top grade Workstation GPUs, the consumer GPUs you can buy are just the lowest binned chipset that failure quality control.
This is the reality we face now.
>>53411809
which is why you hear so much talk from nvidia about "10x performance" for pascall PROFESSIONAL / WORKSTATION (not gaymen) cards. nvlink (PRO / WORK ONLY) + fp64 (PRO / WORK ONLY) + hbm 2 (PRO / WORK AND FULL GAYMEN BIG CHIP (1080 TI AND TITAN PASCAL) ONLY) 10x performance.
maxwell 2 was solely focused on the consumer market, and they really didn't think that fp64 was going to be in demand that much for the maxwell 2 based workstation cards but they saw a huge demand by people that they are bringing it back in pascal. thats why they still make keplar based workstation cards and they need to get rid of keplar there because of power requirements. they are getting a lot of demand for reduce TDP.
