Navi Dicussion Thread

[SteelCrysis]

https://www.tomshardware.com/reviews/amd...216-7.html

Assuming AMD’s lower prices stick, and assuming Nvidia doesn’t clap back, the Navi-based Radeon RX 5700-series cards storm onto the scene with strength that many enthusiasts weren’t expecting. Matching the competition’s fourth- and fifth-fastest gaming cards doesn’t sound particularly impressive. But remember that the Vega generation never caught up to Nvidia’s Pascal-based line-up. And although the Turing GPUs were derided for their high prices, they did bump performance up at least one tier. To AMD’s credit, Navi delivers its speed-up with more grace than past launches. Both 5700s use less power than their predecessors. They generate appreciably less noise. And the Radeon RX 5700s serve up almost universally better performance than competing GeForce RTX 2060s at the same prices.
...
Not all is perfect in Navi-land, though. The flagship Radeon RX 5700 XT runs really hot. AMD did commit to bringing down the noise levels of its reference cooler, which means the fan doesn’t spin up past a certain point. Under duress, clock rates can be pushed below AMD’s rated base specification as junction temperatures exceed 100 degrees C. Our power consumption measurements show the card’s sensors intervening to keep thermals under control.

The Radeon RX 5700 demonstrated other strange behaviors. Mainly, it consistently dialed fan speeds back under even normal gaming workloads, causing temperatures to increase. AMD won’t go into detail on what’s happening there. If we were to guess, though, it looks like the fan responds quickly to a rapid increase in temperature, up to its maximum rotational speed. As the change in temperature slows, the fan slows to allow higher thermal readings and better acoustics. This theory falls apart somewhat when we see the same behavior under FurMark, where clock rates clearly suffer as the fan does something similar. It’d be far preferable for AMD’s thermal solution to ramp up gradually like Nvidia’s Founders Edition cards. We’ll continue experimenting with the fan curve to see if it’s reacting deliberately or if this is a bug that needs to be addressed. Fortunately, none of what we saw seems to adversely affect gaming on Radeon RX 5700 or Radeon RX 5700 XT.

https://www.extremetech.com/gaming/29452...n-the-game

While I don’t have a dB meter sensitive enough to measure GPU fans in a trustworthy way, I don’t need one to tell the difference between the RX 5700 and RX 5700 XT and pretty much every other AMD reference card manufactured since 2013.

The 5700 and 5700 XT are quiet. No, they aren’t inaudible — but they’re head and shoulders better than any high-end reference card AMD has launched in this approximate price bracket for a very long time. When we met with AMD at E3, we were told that the blowers were locked to a maximum sound level of 42 dBa. That sounds (no pun intended) about right to us.
...
Overall, though, I’d say Navi is easily the most impressive AMD GPU since at least the HD 7970. That might seem surprising, given that it isn’t in an overall leadership position and it doesn’t deliver a killer knockout blow to Nvidia’s product stack — though we’d be writing a very different conclusion to that point if Nvidia hadn’t just cut its own prices.

AMD has been playing second-fiddle to Nvidia in the GPU market for years, now. It needed a GPU that could compete on power and performance. It needed an architecture that could hit its target clock rates. Ever since Fury, it has felt as though AMD and GCN were hard up against an efficiency and scaling wall, clawing for every scrap of performance they could find. With Navi, it feels like the company can breathe again.

I’m going to have more to say about features, positioning, ray tracing, and the like. But what I’ll leave you with is this: More than it needed any particular feature, AMD needed to demonstrate that it could still build competitive, well-positioned cards. The RX 5700 and 5700 XT deliver on this promise. They’re faster than the RTX 2060 and RTX 2070 and the price cuts AMD introduced will keep them quite effective against the RTX 2060 / RTX 2060S. If the company can continue to improve its designs and boost overall efficiency, the rumored “big Navi” (expected in 2020) should be something to see.

https://www.techpowerup.com/review/amd-r...xt/35.html

Abolishing CrossFire support might seem sad at first, but I think it's actually a good thing. Why invest time into a feature that is barely used by 1% of your customers? That's probably also why game developers haven't really been investing into it for quite a while, and this move will free up developer resources at AMD that are better spent elsewhere; for example, on game-ready support at launch day, something AMD has been spot on with in recent times.
...

The improved power consumption of Navi helps with thermal management because all power consumed on a graphics card is emitted as heat the cooling solution has to get rid off somehow—more heat requires faster-spinning fans, which results in added fan noise. Unfortunately, noise levels of the RX 5700 XT reference design are not good. Essentially, they are identical to what we've seen on RX Vega and Radeon VII: 43 dBA. Part of the problem is that AMD insists on sticking with blower-type coolers, which simply aren't up to the task. NVIDIA switched their cooler design for the RTX Founders Editions, and it helps them a lot. The RTX 2070 Super, for example, consumes a similar amount of power, but runs a LOT quieter with only 33 dBA, and is significantly cooler at the same time with 76°C. Temperatures of our RX 5700 XT were quite high, reaching up to 92°C during gaming; the "Hot Spot" thermal sensor even went up to 110°C. While the card does not include the idle-fan-stop feature, AMD has tuned the idle fan speeds very well. The card is whisper quiet in idle, and you won't notice it over other components in your PC.

What is shocking are the absurd number of driver bugs I encountered during the review. I listed some of them on the overclocking page, and while they can all be fixed easily with driver updates, it still makes me wonder whether anybody test these cards. Some of the issues are so obvious, I can't imagine nobody noticed them. While overclocking is possible, it is somewhat complicated and requires you to turn off most of the card's clock and power management just to gain a few percentages of performance—5.6% in our case. Especially memory overclocking was disappointing. Both companies use the exact same GDDR6 memory chips from Samsung. On NVIDIA, we can easily reach above 2000 MHz on the memory; on the RX 5700 XT, a 1840 MHz memory OC was the end of the line, probably held back more by driver/BIOS bugs than the memory ICs. Guess we'll see if they can improve that with future driver updates.

Initially announced at $449, AMD just yesterday adjusted the price of the RX 5700 XT down to $399, which helps a lot, especially when comparing to NVIDIA's recent RTX Super release. At that price point, the RX 5700 XT is $100 cheaper than the RTX 2070 Super, offering 10% better price/performance. Compared to the RTX 2070, which is $480 now, the RX 5700 XT even has a 20% advantage. The RTX 2060 and RTX 2060 Super are breathing down AMD's neck though, with price/performance being essentially identical to that of the RX 5700 XT and much better power/heat/noise, fewer bugs, and Raytracing hardware acceleration. I have no doubt AMD will fix those bugs in the near future, but if noise matters for you, you'll have to wait for AMD's board partners to release custom Navi variants with better coolers in the coming months.

https://www.techpowerup.com/review/amd-r...00/35.html

Just like the Radeon RX 5700 XT, the RX 5700 does not support CrossFire. While this might seem like a step backward at first, I think it's actually a good thing. Why invest time into a feature that is barely used by 1% of your customers? That's probably also why game developers haven't really been investing into it for quite a while, and this move will free up developer resources at AMD that are better spent elsewhere; for example, on game-ready support at launch day, something AMD has been spot on with in recent times.



AMD's new architecture brings in significant improvements over GCN we highlighted in the architecture section of this review. One important cornerstone is the long overdue reduction in power draw to make up lost ground against NVIDIA. We were shocked to see the Radeon RX 5700 beat the power efficiency of many NVIDIA Turing cards. AMD has achieved the unthinkable! Their secret sauce is undervolting—the chip runs at below 1 V all the time, whereas the XT card runs at up to 1.2 V. This move greatly reduces power consumption, but comes at the cost of maximum clock frequency, which explains why the RX 5700 is specced at considerably lower clocks than the XT. It's still not perfect, though, as AMD affixed all their voltage-frequency curve points to 0.987 V. Usually, you'd expect to see a gradual stepping pattern that drops to even lower voltages at lower clocks—a bit of efficiency is lost here. With 166 W in gaming, the RX 5700 matches the power consumption of the RTX 2060 almost exactly with more performance at the same time. The reduced power requirements will definitely fly well with people upgrading their system as there is no need for a more powerful PSU. For example, the old Radeon RX 480 used the same amount of power, too.

With such low heat output, you'd expect the RX 5700 to run quietly enough to take on NVIDIA's lineup—power draw and performance are close enough, so fan noise and temperatures should be similar, too. Unfortunately, AMD insists on sticking with blower-type coolers, which simply aren't up to the task. Noise levels reach 43 dBA, the exact same value as for RX 5700 XT, RX Vega, and Radeon VII and simply too noisy. Since the RX 5700 GPU spits out around 50 W less heat than the XT, temperatures dropped from 92°C to 79°C, but apparently, nobody seized that opportunity to adjust the fan curve to reduce noise levels—both cards are capped at around 2100 RPM. What makes things worse is the super weird fan curve the card uses. When the card starts out cool and a game is launched, temperatures will go up as expected. This increase in temperatures slowly makes the fan spin faster and faster to keep up with the heat—so far so good. Once the card reaches 72°C, fan speed will start dropping despite climbing temperatures. Over the next few minutes, fan speed will keep going down to about 1715 RPM (it should be running at 2100 RPM now), and only then, when temperatures have risen to around 78°C, does fan speed increase. I'm completely puzzled by how this could slip by everyone at AMD for them to send out review boards like that. It should be fixable with a BIOS or driver update, though. While the card does not include the idle-fan-stop feature, AMD has tuned idle fan speeds very well as the card is whisper quiet in idle and you won't notice it over other components in your PC.

While overclocking is possible, it is somewhat complicated and really doesn't yield much added performance—we gained 2.4%. Especially memory overclocking is poor. Both AMD and NVIDIA use the exact same GDDR6 memory chips from Micron. On NVIDIA, we can easily reach above 2000 MHz on the memory; on the RX 5700, the 1860 MHz memory OC was the end of the line, probably held back more by driver/BIOS bugs than the memory ICs themselves. Guess we'll see if they can improve that with future driver updates as well.

Initially announced at $379, AMD has just yesterday adjusted the price of RX 5700 down to $349, which helps a lot, especially when compared to NVIDIA's recent RTX Super release. At that price point, both the RX 5700 and RX 5700 XT have the same price/performance ratio, which is quite uncommon. Normally, the slower model has better pricing because you're expected to pay a premium for the extra performance on the higher model. Price/performance also roughly matches the Radeon RX Vega 56—with better performance, of course. Compared to NVIDIA's lineup, both the RTX 2060 and RX 5700 are identical in price/performance, while the RTX 2060 Super is 5% worse, but offers 8% more performance. What NVIDIA has going for its cards, at least when compared with the RX 5700 reference design, is better noise levels, but I'm sure custom designs from AMD's board partners will address that later this year—remember, heat output is the same as for the RTX 2060, so there is no reason these cards won't be just as quiet as RTX 2060 cards, which is a huge win for AMD, especially with its better pricing.
...
If AMD can lower prices of the RX 5700 a bit more, say to around $320 or even $299, and custom designs manage to achieve heat/noise parity with NVIDIA, they should be able to capture a significant portion of the market back from NVIDIA.

https://www.techpowerup.com/review/pci-e...xt/24.html

Looking at the results, we can see a whole lot of nothing. PCI-Express 4.0 achieves only tiny improvements over PCI-Express 3.0—in the sub-1-percent range When averaged over all our benchmarks, we barely notice a 1% difference to PCIe Gen 3. I also included data for PCI-Express Gen 2, data which can be used interchangeably to represent PCIe 3.0 x8 (or PCIe 4.0 x4). Here, the differences are a little bit more pronounced, but with 2%, not much to write home about, either. These results align with what we found in previous PCI-Express scaling articles.
...
When looking at individual game results, the effects of constrained PCIe bandwidth vary wildly. Some games, Shadow of the Tomb Raider, for example, barely show measurable differences, while titles like Rage 2 and Wolfenstein are much more dependent on PCI-Express bandwidth. I can't see a clear trend between APIs or engines as it rather looks like a dependency on how the game developer chooses to implement their game and how much data they copy from the CPU to the GPU, or even back.

These results are also good news for people who consider running their graphics card at reduced link width, like x8 or even x4, to free up precious PCI-Express lanes for other devices, like storage.