Business, Graphics, Hardware

AMD Community has 29 tough GF100 Questions for nVidia


Overkill VR is a fast paced VR action game

Late last night, we received an e-mail containing 29 questions for nVidia in regards to GF100 and nVidia in general. Bear in mind that these are heavily biased questions, but we decided to run them anyways – with a few comments of our own.

All of the questions below come from AMD user forums such as Rage3D, Beyond3D and so on. We’ve forwarded the link to nVidia and expect to get an answer during this week.

 Why are there only 64 texture units in GF100? This is a major technological set back compared to GT200. Are they running out of silicon real estate for graphics features with all the HPC compute circuitry around? Have they abandoned the gamer?

BSN*: Pretty interesting question. And yet, if you don’t vectorize your data, due to lack of HPC compute circuitry a precious Cypress GPU becomes a 320 shader part.

 When will they know what clock speeds GF100 can run at? Are they running into other unanticipated problems with clock speeds due to on-chip variations with the huge die size and high power consumption? 

 Is it true that yields are currently just 2% due to poor engineering and lack of ?dual redundant vias? connecting the chip?s metal layers, and that most GF100 chips won?t even be able to hit 650MHz.

BSN*: This is quite an interesting question, given that that story was coming from satire website. In any case, Cypress has the same issue, given the mountain of Cypress wafers at TSMC – those chips can’t even hit 5850 speeds. But hey, don’t shoot the messenger.

AMD: Why is a 60 second snapshot of the Unigne benchmark used for performance comparison? Why not the entire benchmark or other real game performance comparisons? What are they trying to hide?

 Given their past troubles with packaging, how do they deal with thermal cycling reliability issues with the huge die size and high power consumption of the GF100?

AMD: Why are they artificially limiting the double precision performance of GeForce products to 1/4 of full capacity and make gamers pay for it in silicon and in power consumed?  Why aren?t gamers allowed to use all the performance they?re paying for?

AMD: NV?s distributed PolyMorph Engine structure would imply that some of those functions are performed by the shaders. Will their tessellation performance scales with shader load and degrades when the shaders are busy with other tasks?

 When will they be able to ship real DX11 products? AMD has complete stacks of desktop and mobile DX11 offerings.

 Why do 48% of INQ readers think GF100 will be too hot, power hungry and expensive?

 Why does NV keep at proprietary standards when open standards exist? Typically companies do this when they have poor products to hide.

BSN*: With AMD, all is nice and candy when it comes to talking about open source. Unfortunately, talking the talk and walking the walk are two separate things. AMD needs to seriously step up the plate as far as OpenCL is considered and stop avoiding answering the tough questions and be honest about the real state of AMD’s OpenCL implementation.

AMD: AMD now has top to bottom DX11. Why doesn’t NV have any products to compete? What happened? The Windows 7 ship date was known well in advance.

AMD: Why do NV drivers have to be 100MB in size? Users complain that it takes forever to download.

BSN*: In or
der to have the same functionality, user has to download AMD Drivers + HydraVision. 70.1MB and 28.6MB make a grand total of almost 100MB. Truth to be told, nVidia drivers are usually 30% bigger than 100MB, last one being 123MB.

 Why doesn’t NV release drivers on a regular and predictable way?


 I’ve started to see PhysX in titles, but why isn’t the API running on my NV GPU? Seems to be running on the CPU.  How many PhysX titles actually run on the GPU after 5 years of development by Ageia and Nvidia?

 In September 2009, NV claimed the new GF100 architecture would ship in November or December. What happened?

 Why are you trying to move the PC industry to a proprietary based model and limit functionality to your own products?

 When will we see OpenCL based consumer applications running on your hardware?

 Is the GF100 just NV30 all over again?

 What is NVidia?s commitment to launch quantities of GF100?

BSN*: Seriously? How about AMD coming clean about the real launch quantities of Radeon 5870 and 5850 at launch date, 5970 at launch date [yes, we know what OEMs got shafted as the board distribution went elsewhere]. To our knowledge, the only Evergreen ASIC launched with real volumes in 2009 was Radeon 5700 series, i.e. Juniper GPU.

 What are the power supply requirements for GF100?  What about GF100 SLI or 3-Way SLI?

 What is the power leakage on the GF100?

 Will GF100 be available in notebooks? Thin & light too?

 How many GF100 cards are required to power 6 screens using only 1 PC?  What are the power supply requirements for such a PC?

 Will GF100 be the single fastest graphics card in the market at launch?

 How does it compare to ATI Radeon 5870 in terms of performance per $ and per W?

 Does Nvidia keep PhysX proprietary because NV is afraid that ATI Radeon would deliver a better PhysX experience?

BSN*: A very interesting comment indeed, especially after AMD locked out a run-around that enabled PhysX-wrapper running on Radeon 3000 and 4000 series cards – back in August 2008.

AMD: How many TWIMTBP game titles [shipping or in development] are currently supporting DX11?  How many Fermi boards have you sampled to game developers?

BSN*: We would really, really like for AMD to provide those same answers as far as Evergreen boards go. According to developers we spoke with, it isn’t exactly that AMD was giving them left and right.

AMD: When do you expect DX11 support to be available across your entire GeForce product line?

AMD: Why did a poll recently conducted by website conclude that 54.0% of respondents preferred DirectX 11 over PhysX (14.2%)?

Editor’s take
One might ask why we have decided to run this story. The answer is quite simple – over the past 10 years, we just grew tired of all the bickering and petty talk, knife stabbing and all. AMD’s community has some tough questions on GF100 and it would be a good idea to get some answers beside that last minute name change decision. However, those questions are of great interest for nVidia community as well, especially given in mind that we were often asked the same questions from nVidia own partners, yet alone readers. 

Update February 11, 2010 23:01 UTC – Following the publication of this story, we got contacted by AMD and Bite PR, wanting to clarify the comments we gave out in the story. We spoke with Chris Hook, Director of PR at AMD, who told gave us the statement in regards to availability of Radeon HD 5000 series, which we all remember – was mostly unavailable after the launch.

"The 2 million DX11 GPUs shipped announcement back in January included mobile and desktop. Remember we announced the entire mobile family in early Jan, so chips would have had to ship to customers before then. So look beyond Cypress, Juniper and Hemlock when you?re looking at the numbers. We promise that they?re accurate.
BTW, no issues getting 5870 boards today ? just check online!"