A source at a leading graphics card manufacturer claims that Nvidia isn't playing fair in its dealings with GK100 customers.
Nvidia is claimed to be exerting pressure on its board partners to stick to stock voltages in their designs, forcing them to drop hardware or software overvolting and even restricting most other hardware modifications on boards based on the GK100 GPU.
Hints as to the problem came earlier this week, when Overclockers
found a thread on the EVGA forums stating that Nvidia is enforcing 'guidelines' that have resulted in the removal of the EVBot port - used to adjust hardware settings including voltages - from the company's GeForce GTX 680 products.
'Newer 680 Classified cards will not come with the EVBot feature,
' an EVGA employee identified as 'JacobF' told a customer who had bought a board missing the EVBot port - a feature that had been present on earlier board revisions. 'Unfortunately we are not permitted to include this feature any longer. It was removed in order to 100% comply with Nvidia guidelines for selling GeForce GTX products: no voltage control is allowed, even via external device.
That seemed strange, given that EVGA had been selling the GTX 680 Classified with the EVBot port previously. If Nvidia is insisting on limiting what its board partners can do with the chips provided, surely this is something that would have been communicated at the design and evaluation stage, not after products had hit shelves?
Apparently not. A source inside one of Nvidia's largest graphics manufacturing partners, who spoke to us on the condition that they remain anonymous, explains: 'The fact is Nvidia is stopping ALL partners from allowing any form of hardware/software overvolting, or providing hardware mods beyond its very limited restrictions. They threaten to cut allocation [of GK100 parts] if hardware mods aren’t removed or avoided entirely.
While homebrew soldering-iron-and-prayer overvolting is still permitted, manufacturing partners aren't allowed to make it easy for buyers. 'We're not allowed to openly advertise the PCB markings [for overvoltage adjustment] on the GTX 680,
' our source continues.
Bright Side of News
has word from Nvidia's Bryan Del Rizzo that restrictions on overvoltage relate to Nvidia Green Light, a programme designed to provide warranty support. According to Del Rizzo, manufacturers aren't prevented from producing boards that provide voltage control outside the Green Light specifications but will receive no warranty from Nvidia if they do - potentially leaving the manufacturer on the hook if they receive a batch of bad chips.
Claims that manufacturers aren't being restricted in their designs beyond the confines of the Green Light programme are soundly denied by our source, however. We've been told that the secretive restrictions on board partners go yet further: 'They [Nvidia] also threaten allocation if you make a card faster than the [stock] GTX 690.
These restrictions are not limited to just a couple of companies, either: they appear to stretch right across the board, and are responsible for product cancellations and - as with EVGA's removal of the EVBot header from the GTX 680 Classified - hardware modifications from multiple manufacturers. They're also leaving a bad taste in board partners' mouths: where in previous generations each company has been able to push its own cards to the limit in order to beat the competition, under Nvidia's alleged new rules all GTX 680 boards will be more or less identical in performance and features.
The hardware restrictions are a loss for the consumer, too: EVGA has already stated that it won't be reducing the price of the GeForce GTX 680 Classified, despite removing the EVBot header and corresponding facility for custom voltages outside Nvidia's recommended limits - meaning buyers now get less card for their cash than before the company capitulated to Nvidia's alleged demands.
We've approached other board partners, but thus far none have been willing to comment on the record regarding our source's claims of hardware restrictions - and with our source alleging that Nvidia may even cut chip allocations for companies that talk publicly about the matter, that's no surprise.
Approached about our source's claims, a spokesperson for Nvidia provided us with the following statement:
'Some of our best and most passionate customers have told us (though forums, partners and directly) that they are frustrated with our position on GPU Overvoltaging. So we feel that it is important to explain exactly what our position is and why we feel that it is important.
'We love to see our chips run faster and we understand that our customers want to squeeze as much performance as possible out of their GPUs. However, there is a physical limit to the amount of voltage that can be applied to a GPU before the silicon begins to degrade through electromigration. Essentially, excessive voltages on transistors can over time "evaporate" the metal in a key spot, destroying or degrading the performance of the chip. Unfortunately, since the process happens over time, it's not always immediately obvious when it's happening. Overvoltaging above our max spec does exactly this. It raises the operating voltage beyond our rated max and can erode the GPU silicon over time.
'In contrast, GPU Boost always keeps the voltage below our max spec, even as it is raising and lowering the voltage dynamically. That way you get great performance and a guaranteed lifetime. So our policy is pretty simple: We encourage users to go have fun with our GPUs. They are completely guaranteed and will perform great within the predefined limits. We also recommend that our board partners don’t build in mechanisms that raise voltages beyond our max spec. We set it as high as possible within long term reliability limits.
'The reason we have a limit on max voltage is very simply to prevent damage to the GPU chips. At NVIDIA we know that our customers want to push their GPUs to the limit. We are all for it, and as a matter of fact NVIDIA has always prioritized support for hardware enthusiasts by providing tools to access hardware settings and by supporting our board partners in creating overclocked enthusiast products. Leading up to the GeForce GTX 680 release for example, we worked closely with developers of 3rd party overclocking utilities to make sure they fully supported GeForce GTX 680 and GPU Boost on the day of launch.
While Nvidia admits that it 'recommends' that its board partners don't build in overclocking and overvoltage mechanisms, it states that this is just a recommendation - a claim distinctly at odds with those of our source.