daimonie

Hey LMG, A suggestion: Can you do a video about sustainable (gaming) pcs? What can we do the reduce the upfront CO2 emission cost of our system? How can we run the most efficient system ourselves? Are there settings that could help in reducing the power draw? Why would you do this? Well, we all know that the segways to lttstore.com are the most important. And clearly, longevity of the product is one of the best ways to reduce your overall CO2 for having a system, because you'll use less. Just like the quality products from lttstore.com (i'm joking, you don't need help to come up with segways).

I've set the PWM profile so that it doesn't do that as much. I've looked at my temperatures under different daily loads (browsing, gaming, working) and set the profile at mostly flat levels for these. So while it transitions from one load to the other it revs, but it doesn't rev during a workload.

I'd be interested in seeing a waterblock review or comparison

Numerous components (VRMs, chipset and others) don't need significant cooling - only a very small amount. In a sealed case, it can still propagate that heat through the air. In a vacuum case, it cannot so these components will heat up.

The O-ring side of this fitting can screw into your radiator. On the other side of this 90 fitting you can connect a normal compression fitting. The hard tube will go on that fitting like on any other fitting.

Wanted to get back to you on this As you said, the Tensor cores are meant to do D = AB +C matrix operations. (p15, 16 and Fig. 8 ). For RT Cores, I'll quote some stuff: "Due to its processing intensive nature, ray tracing has not been used in games for any significant rendering tasks. Instead, games that require 30 to 90+ frame/second animations have relied on fast, GPU-accelerated rasterization rendering techniques for years, at the expense of fully realistic looking scenes". p25 "While ray tracing can produce much more realistic imagery than rasterization, it is also computationally intensive. We have found that the best approach is hybrid rendering, a combination of ray tracing and rasterization. With this approach, rasterization is used where it is most effective, and ray tracing is used where it provides the most visual benefit vs rasterization, such as rendering reflections, refractions, and shadows. Figure 16 Shows the hybrid rendering pipeline." "Developers can also use material property thresholds to determine areas to perform ray tracing in a scene. One technique might be to specify that only surfaces with a certain reflectivity level, say 70%, would trigger whether ray tracing should be used on that surface to generate secondary rays." "Do not expect hundreds of rays cast per pixel in real-time. In fact, far fewer rays are needed per pixel when using Turing RT Core acceleration in combination with advanced denoising filtering techniques." "The RT Core includes two specialized units. The first unit does bounding box tests, and the second unit does ray-triangle intersection tests." After reading about BVH search, I'm actually surprised.This is a technique I'm slightly familiar with from computational physics (https://en.wikipedia.org/wiki/Octree). The 'winner' of our Computational Course was someone that used this technique to do a real-time openGL simulation of Andromeda and the milky-way colliding.. We all had done similar calculations at the start of the course, and the number of particles he had (using OcTree) was staggering. Also relevant is the difference between these methods (https://computergraphics.stackexchange.com/questions/7828/difference-between-bvh-and-octree-k-d-trees). To make the comparison: In an N particle simulation, you would make N* (N-1) / 2 calculations (pairwise) for each interaction. With the OcTree algorithm, you would be already aware of what particles are sufficiently far away to be negligible. You would still include them, but quite often you would start calculating "a galaxy" rather than "a large cloud of stars", which improved performance. So these algorithms are about reducing overhead. As I understand it from page 32, the SM launches a ray probe, after which the RT core will traverse the BVH tree and ray-triangle tests, returning hit (no hit) to the SM. At which point the SM knows whether or not it has to calculate anything. I imagine the process like: Probe: Will I hit something? RT core: Yes, you hit a cup SM calculates reflection Probe from reflection on cup: Will I hit something? RT core: Yes, you hit an inkpot SM calculates reflection Probe: Will I hit something? RT Core: Yes, you hit a light source SM: K, cool It's not calculating the path of a ray (It is straight, unless massive gravitational influence), but calculating what it will hit that is the problem. And that's what this is custom designed for. So what's the difference? Well, RT Cores are very different from Tensor cores. One thing is the size on the die. A tensor core is a small unit able to do a simple mathematical operation really, really fast. An RT core is able to massively reduce overhead in ray-tracing really, really fast.

That's how it is supposed to work, as far as I understand it. But I also understood - I think from GamersNexus video - that they are applying filters somehow. I would have to check where that idea came from (I'm at work currently :()

Seems like it. But then because BF enabled it later in their process, they had all those entities in place already. So things that shouldn't reflect - e.g. tree bark - were reflecting when they turned it all on (What I called "enable it everywhere). It's not a RTX problem, but just that they developed without RTX and enabled it later. Clearly you don't want to go refactor everything, so they tried to fix it by using some kind of filtering.

I would phrase that quite differently: It has not been used because it was too expensive computationally The extra visual quality was worth it, but not for real time gaming. Anyone who plays competitive games will keep detail settings low. NVidia reached the point where 60 FPS is doable, and wants to get some of its R&D costs back The AI features of the cards (DLSS etc) aren't used yet, but seem promising Those same features are worth it for some contexts. I've used GPUs in academic research for RT calculations for e.g. a high frequency tricolour fluorescent microscope. Tensor cores would've been awesome. All of that seems fine to me. The RTX 2070 was competitive in pricing with the 1080, so I bought it. Edit: Regarding battlefield, it seems their implementation is quite bad. From what I can gather their "enable it everywhere" gave too big a performance drop, and they filter out RTX features to control performance. That gives weird artefacts in addition to being not consistent.

I'm sorry, you seem to have misunderstood. I mentioned paraxial optics because it is a branch of optics that can formulate ray paths in pure matrix products. Apparently, it is referred to differently in English (Myoptics professor used archaic naming. Transfer matrices are the more common term. https://en.wikipedia.org/wiki/Ray_transfer_matrix_analysis) This was to argue for my question of what the differences between RT and Tensor cores are.

While RPM doesn't translate to sound equally, your point seems valid to me. It does matter what particular fans they are and such.

I'm still curious about how they are different. There's a branch of optics related here: https://en.wikipedia.org/wiki/Paraxial_approximation If you use those approximations, you can recast reflections, Snell's law and everything in terms of Matrix products (2nd order tensor products). So what's the difference between RT and Tensor cores?

The bigger it is, the slower your fans can be. It's a combination of surface area and the flow rate of air through them. Larger surface area allows a smaller flow rate. There are 420 rads on the market, e.g. https://www.ekwb.com/shop/ek-coolstream-ce-420-triple

If NVidia used Google's tech, that would make sense. But I don't know if they did. The price increase is small and seems fine for everything but the 2080Ti. I think we can all agree it went up by a larger amount (because it did), without a clear reason why. The reason might just be that the profit, if any, on the 2070 and 2080 are small (so you get more adopters) which they try to counter with the previous Titan segment of the market. As you can see in this simplified graph, the benefit of the old titan and the new rtx 2080 Ti are fairly small, even for cinematic games. https://docs.google.com/spreadsheets/d/1xGzdYeTiTEH_kJYwYAYdvyUHT8ZFLHYxhmReEjJvntU/edit?usp=sharing But aha, you say; RTX and DLSS! Well, sure, but those are true for the RTX series - making the 2080 Ti still weirdly placed. Edit: I linked the sources I used. I wanted to get a quick graph done, so I took the first single-source I could find for most cards. The MSRPs should be fine, they come from Wikipedia. Based on the graph, a logical place for the RTX 2080Ti would be around 1k.

For those, the RTX 2070 wouldn't use the new fancy tech. Rather, it would trade blows with the 1080. The 1080 Ti is stronger than that; so if you are going for results at that pricepoint, GTX 1080 Ti seems the way to go.

Oh, that's actually interesting. Did you wait for the liquid to equilibrate?

Can you dig up an example? I'd like to see Wait, can't do raytracing + HDR?

Yes, I know. The Aorus waterforce WB went from 850 to 1300, and people are trying to resell for 850 now. Point is, with a small price difference the 2080 is rather attractive. (The 2080Ti isn't as much to me. But I'm guessing NVidia plans to put something at 1k in the future..)

Just to illustrate that difference, over in the Netherlands the prices (new) are now: 1080TI's (min and max, neglecting water cooled): GV-N108TAORUS11GD at eur. 742 ROG-STRIX-GTX1080TI-11G-GAMING at eur. 1060 RTX 2080: N20802-08D6-1180633 at eur. 749 ZT-T20800B-10P at 976

I just wanted to say that the "performance on the best hardware is poor" is a bit misleading. Performance if measured in FPS is poor - but to do so, it has to do more. Arguably performance is great The best desktops start with the question: What do you want to do? RTX graphics will look great in cinematic games (e.g. Tomb Raider series). If you're mainly into playing shooters, RTX won't do much for you. So, what do you want to play? And how much money do you want to spend? What do you think of buying a used GPU?

You asked what it can do, not what games are promising to support now. Also, as far as I can tell FFXV will still have DLSS: https://www.nvidia.com/en-us/geforce/news/final-fantasy-xv-windows-edition-geforce-rtx-dlss-benchmark/ https://wccftech.com/final-fantasy-xv-windows-edition-nvidia-dlss-support/

Haha, it's several years back for me as well. https://en.wikipedia.org/wiki/Thermal_diffusivity " It measures the rate of transfer of heat of a material from the hot side to the cold side. " That's not the one we want in this story. Transport is done by pumping the fluid around, so I don't care much about it for this scenario. If you look at the image attached, you'll see a bit of what I mean. The thermal conductance will determine the rate of transfer between the IHS at T_H (Temperature Hot) and the waterblock, which should be at T_W, the equilibrium temperature of your loop. At the other end, water will transfer heat to your radiator at ambient temperature T_A. So, the difference dT1 = T_H - T_W will determine transfer at your CPU, the difference dT2 = T_W - T_A will determine transfer at your radiator. They should be balanced. The amount of heat drawn out at the CPU block will be largely determined by the thermal conductivity there, the mass flow rate and the specific heat. When you are pumping the liquid, you essentially make sure that the entire reservoir of liquid will be at T_W. However, that temperature will be determined by the specific heat: The above is the lazy equation for the heat in the liquid. The thermal resistances are effective numbers for the components+liquid. You can get them by finding the solution to a heat equation with piece-wise thermal conductivities and finding effective forms that look like Ohm's law. https://en.wikipedia.org/wiki/Thermal_resistance So, assuming that you pump the liquid fast enough to keep gradients high (T_W at equilibrium), which will be harder for the oil, you still need to have have the water at equilibrium.

Yes, I wonder what AI can do for gaming. https://www.techspot.com/article/1712-nvidia-dlss/ Oh, right - one of the two main features of the cards? (AI can do a lot of things for a lot of things. Gaming isn't going to be a use case where it can't be used)

False. NVidia has explicitly stated the Titan isn't a consumer card,and removed the Geforce Branding because of it. Regarding PowerVR, this is not exactly the same technology. https://www.reddit.com/r/hardware/comments/9akp0l/powervr_had_a_6gray_real_time_ray_tracing_long/ Comparing the 2070 to the 1080, the 2080 to the 1080ti and the 2080ti to the Titan Xp: Launch msrp for the Titan Xp was $1200, 1080Ti was $699, 1080 was $549. https://en.wikipedia.org/wiki/GeForce_10_series The RTX 2080 Ti is at $1199, the 2080 at $799 and the 2070 at $599. The 2080 and 1080 Ti are comparable in performance, as are the 2070 and 1080. The 2080Ti beats the Titan Xp. So, in conclusion. It seems that at comparable price points, you gain the generational improvement for the 2080Ti. For the 2080 as well, but it costs about $100 more than the 1080 Ti. For the 2070, you're paying about $50 for the same performance. Feel free to compare the 2080 Ti to the Titan V instead; that would net you the same performance, but the Ti is at a far better price. IMHO, these are hardly excessive premiums to pay for the extra features.

Behold, the number 5. This is a random number I thought of, and the price I want GPUs to be. Much like the way you're telling NVidia their product is too expensive. I'm not 100% happy about the situation. But the cards *are* better performers, and they a lot of extras. Those extras might not seem worth it to you - but after several years of development, that tech is worth some money as well. If the price of the tech was lower, but the games weren't out yet, you would've complained too. Releasing a brand new technology is always a losing situation, it seems.