New Posts  All Forums:

Posts by tpi2007

If it's needed, I'm betting it will, as the penalty of reading from the 0.5 GB segment is just too great - low bandwidth and the inability to read from the 3.5 GB segment concurrently, so the option is to move it to the 3.5 GB segment and move something less important to the 0.5 GB one.That's akin to saying that people should be happy that they got that extra 0.5 GB of 28 GB/s memory instead of nothing and game assets having to be shifted to the much slower system RAM...
You forgot to mention the small detail that this third tier can't be accessed concurrently with the other two, so the juggling of game assets around the two segments is not the same as shifting things to RAM while not losing cycles with the VRAM.
Anandtech says otherwise:http://www.anandtech.com/show/8935/geforce-gtx-970-correcting-the-specs-exploring-memory-allocation/2Are you 100% certain that there is also no per application optimization going on ?http://www.anandtech.com/show/8935/geforce-gtx-970-correcting-the-specs-exploring-memory-allocation/3
They should at least have taken a more neutral stance. We all know how businesses can be cynical, why are they taking sides when they obviously don't know what actually happened, what motivations they had and, especially, when common sense says that this is so far fetched (that nobody noticed for four months) that the best approach to appease both Nvidia and your readers is to not take sides and just report what they said. It left me uneasy with them too.They sure must...
I like Nvidia GPUs, but I just felt my intelligence was insulted two times. First by Nvidia and then by sites such as PCPer and Anandtech that said they believe them (they could at least have taken a neutral stance).I'm just having a very hard time believing that nobody from Nvidia's engineering staff read ANY of the GTX 970 reviews and didn't notice that the card's specs were being misrepresented everywhere. I also have a hard time believing that nobody from Nvidia's...
I thought my post was self-explanatory, but I'll give it another try: it matters if, depending on where on the die they cut off defective cores that affects performance from card to card, thus possibly explaining why some people are reporting problems and some aren't.It's an idea worth discussing, and more importantly, that the tech media should ask Nvidia. And that is why I quoted that Anandtech sentence that sums it up pretty well: up until now reviewers thought they...
As I also asked previously, there's another question that we may not know about: does Nvidia fuse off the other cores in exactly the same way on the GM204 die ? I wouldn't think so as that would imply all the faulty chips to be damaged in exactly the same area. So, having said that, does cutting cores here or there make any difference performance wise ? How many crossbars are available depending on where they cut off ? Nobody knows. What I know is that some people are...
As I said in the previous thread, the first part of the sentence does not necessarily imply the second (bolded) part. People who bought an Intel hexacore are getting all the memory throughput to the quad channel memory interface, they don't have to buy the octacore model to get that.Also, "higher priority access" usually means 'faster' compared to the lower priority access, which must have some correspondence to throughput, which may very well mean that you actually don't...
There's quite a few questions that still need answering: 1. If the benchmark people are saying is bugged can't properly access the last 0.5 GB on the GTX 970, but can on the GTX 980 is that because Nvidia's drivers aren't catching the benchmark and applying its damage mitigation routines to it ? I have a hard time understanding why it would be the benchmark's problem and not Nvidia's drivers. The benchmark would just be assuming the normal, it's up to the drivers to...
As I've read over at PCPer, this is because they chose to do it like this. Having less cores than the full chip doesn't necessarily imply having inferior access to the VRAM. Does an Intel hexacore only have access to a part of your RAM at the rated bandwidth and to get the whole thing you need to buy the octacore ? No, of course not. So this isn't a clear cut case of excusing Nvidia for their design choices. In any case, it's interesting to see how they apparently...
New Posts  All Forums: