Monday, March 18th 2019
![NVIDIA](https://tpucdn.com/images/news/nvidia-v1721205152158.png)
NVIDIA GTC 2019 Kicks Off Later Today, New GPU Architecture Tease Expected
NVIDIA will kick off the 2019 GPU Technology Conference later today, at 2 PM Pacific time. The company is expected to either tease or unveil a new graphics architecture succeeding "Volta" and "Turing." Not much is known about this architecture, but it's highly likely to be NVIDIA's first to be designed for the 7 nm silicon fabrication process. This unveiling could be the earliest stage of the architecture's launch cycle, would could see market availability only by late-2019 or mid-2020, if not later, given that the company's RTX 20-series and GTX 16-series have only been unveiled recently. NVIDIA could leverage 7 nm to increase transistor densities, and bring its RTX technology to even more affordable price-points.
99 Comments on NVIDIA GTC 2019 Kicks Off Later Today, New GPU Architecture Tease Expected
seen this generic canned statement once too often. :shadedshu:
In actual fact my comment really was cute, having a fun little poke to show another side of the coin. Also amusing is you picking the only part of what I said that you could 'disprove', and in doing so you managed to perhaps inadvertently achieve even more than you intended.
www.techpowerup.com/249077/nvidia-confirms-issues-cropping-up-with-turing-based-cards-its-not-a-broad-issue
Would you like to rescind your vitriolic response?
Ray tracing support is implemented through Vulkan, an open standard, and DXR in DirectX, a proprietary Microsoft standard, and neither of these are locked to Nvidia hardware.
Are you freaking kidding me? Tensor cores aren't (I mean it's a generic idea of hardware accelerating a trivial operation).
However RT cores are (figuring what and how they do to do it so quickly is not trivial)
Most cases of reported problems were bugs in the launch driver, totally unrelated to the few defective cards. What they do on an API level is clearly defined.
What they do on a transistor level is proprietary, and shouldn't matter to end-users or programmers. This is no different from any other units inside GPUs and CPUs.
Failure rates of ~2% for DOA/almost immediate failures and upwards to 5% during the warranty is completely normal for consumer electronics. There is a reason why manufacturers provide warranty after all. Failure rates of 1 per million (0.0001%) is unheard of for consumer electronics, ask any retailer or manufacturer. Personally I've RMAed a lot of stuff over the years, including many hard drives, several graphics cards, laptops and even two identical server boards which were DOA.
Consumer electronics are tested during production, and serious makers do have a basic QA procedure on every sample, usually a test run of a few minutes and a visual inspection. They usually also do a lengthy test of a small random selection of each batch. But they can't run every sample on a 2 week stress test before shipping, not only would it cost too much, but it will also cut like 3 months of the product's life expectancy, doing too much testing might actually hurt RMA rates. Most DOA products are products that have failed between the factory and the end user. Transporting products around the world is not risk free, rapid changes in temperature and pressure will cause weak parts to fail, and you really can't completely eliminate this from the factory side.
So far, there is no evidence supporting that Turing have higher failure rates than normal, In fact, according to Steve at Gamers Nexus, several board partners are claiming that Turing so far have lower than normal failure rates. So until you have evidence proving he is lying, stop spreading your FUD.
I'm not talking about turing. The chips are fine so the quality and the production and failure rates make sense and it is below 1%. What you are referring to is the memory failure. Which causes the failure. You are focusing on the GPU as a product which has components from dozen of vendors. I don't and that's the difference. You consider graphics card as a product and that's ok. Maybe you should focus on, for instance chip as a product which will be used in variety of cards form different vendors. Then your 2-5% is ridiculous. That's what I'm talking about.
For instance it has been said that the Turing cards it is mostly due to memory not the chip. So it means that memory modules aren't being examined or tested properly.
So if you have 5% of broken cards and the percentage of these broken cards is for example 95% with mem issue, no it isn't normal.
The "memory problem" you refer to with Turing is not a specific memory problem, it's in fact all kinds of random problems which happens with any series of graphics cards. The problem with this matter is that all kinds of normal problems are mixed into one giant bag and called one problem. Most of the reported cases had no hardware defects at all, and were experiencing one of the various distinct driver issues. The remaining cards have various hardware defects, which are covered by warranty. And that "space invader" pattern doesn't necessary mean it's a memory problem, just that something causes memory corruption. It could just as well be power delivery, memory controller or even weak traces on the PCB. You need proper debugging tools to find out what's causing it.
If it fails it is defective for any reason. In terms of turing and memory related issues. What is it exactly if you please enlighten me? If it is not memory specific then what is it? Don't tell me random components. I just don't buy it. Large chunk of the problems was memory related. Maybe it's because it is new technology but doesn't change anything.
People reported various symptoms, after a few weeks at least two distinct driver issues were identified and fixed. Most of the reported cases in forums etc. fit one of those buckets. Those that remain are probably some kind of hardware defect. I'm telling you a symptom is a symptom, and doesn't necessary reveal the underlying problem.
For the end user it doesn't matter, since a reproducible hardware defect is a defective product, and they will get a new one.
When it comes to the so-called memory diagnosis from various forum users, who claimed it was one bad batch from one supplier, but it was later debunked as the "problem" was reproduced across two memory vendors and multiple batches of chips. Some of these cards may certainly have failing memory chips, but you can't claim that just based on the crash.