I wonder how much of an improvement the AIB cards will be, since the Navi cards have power limit restrictions. My plan is to add a Navi card to my custom loop, mainly for silence. But I would try some overclocking if it doesn't produce too much heat for the loop to stay silent. I'm sure we'll see a reference vs AIB design comparison on air, but I'd really appreciate it if someone included a water cooling performance comparison, too. Very hard to get one, I know.

Grumpymangrumbling2019:

Releasing a card that isn't any faster than their previous is kind of bonkers especially when you look at the power consumption. This silicon should of had more and be competing with the 2080 not the 2070, There is no good reason to purchase one of these it wont change your gaming experience. Disappointing for the consumer but interesting to the tech.

You would be right to complain if we were talking 5900. But since this is two performance tiers down from AMD's usual flagship range, your complaint is somewhat moot. I say "somewhat" because the estimated power consumption is disappointing.

We will have a great product line, but as some have said here at Guru3D, AMD often doesn't handle the launch very well. Navi will get more interesting as more time goes by and I expect a significant price drop only after one month from release. It seems they want to try the Nvidia way to sell the default model at a higher price for early adopters, so the upcoming custom cards are cheaper, the real deal. What a stupid move if this happens, but I guess the launch price is set in stone now, so it will. Wait till August and you will have a great deal and if you wish to wait till Black Friday you will get an awesome deal! Navi is a small chip and it can price out Nvidia any day, until they refresh Turing with 7 nm, then Radeon is screwed once again, which is sad for Navi 20, most likely making it a rather unpopular product. If I were AMD, I would sell Navi for this short period of time with a lower profit margin just to gain marketshare as much as possible to avoid consumers upgrading to 7 nm Turing instead of Navi. If one has bought a new GPU in autumn, he will most likely not buy a new one in a long time after that, no matter what comes up.

Grumpymangrumbling2019:

I was hoping for a upgrade and as usual AMD fails to deliver.

So.... despite your obvious pessimism, why exactly were you hoping a mid-range GPU, which we don't have any real benchmarks of, was going to be something you'd upgrade to? That's like ordering a steak, knowing full well what one tastes like, and despite this, you complain that it doesn't taste like a chocolate cake before you even got it. This is no different: you already had in your mindset that this GPU wasn't going to be what you wanted, it was never built to be what you wanted, and even though there's no real performance data on it, you're already disappointed. Your "hopes" are delusional.

Mpampis:

I wonder how much of an improvement the AIB cards will be, since the Navi cards have power limit restrictions. My plan is to add a Navi card to my custom loop, mainly for silence. But I would try some overclocking if it doesn't produce too much heat for the loop to stay silent. I'm sure we'll see a reference vs AIB design comparison on air, but I'd really appreciate it if someone included a water cooling performance comparison, too. Very hard to get one, I know.

I can tell you from my experience with V64 Nitro. The card out of the box was 12% faster than the V64 reference. By undervolting was able to keep it stable at 1700Mhz core clock (for benchmarks went to 1750), and 1100 HBM while was quiet. Performance to put in perspective, it beat royally my watercooled GTX1080 @ 2190. As for in relation to current cards, a reference V64 at 1576/945, keeps up with a RTX2070 @ 1905 been neck and neck of within up to 5 fps difference depending the games. So add an extra 13% on top for the V64 Nitro when fine tuned. Similarly here, given that the reference model is faster than the 2070, tweaking an AIB or watercooling, it will be within single digit % of the 2080 and RVII. Cards costing >50% more. Also from the AMD benchmarks, we saw some games like AC, the RDNA gets big perf boost over GCN where GCN was lacking.

Fediuld:

I can tell you from my experience with V64 Nitro. The card out of the box was 12% faster than the V64 reference. By undervolting was able to keep it stable at 1700Mhz core clock (for benchmarks went to 1750), and 1100 HBM while was quiet. Performance to put in perspective, it beat royally my watercooled GTX1080 @ 2190. As for in relation to current cards, a reference V64 at 1576/945, keeps up with a RTX2070 @ 1905 been neck and neck of within up to 5 fps difference depending the games. So add an extra 13% on top for the V64 Nitro when fine tuned. Similarly here, given that the reference model is faster than the 2070, tweaking an AIB or watercooling, it will be within single digit % of the 2080 and RVII. Cards costing >50% more. Also from the AMD benchmarks, we saw some games like AC, the RDNA gets big perf boost over GCN where GCN was lacking.

There's more performance on the table too with memory timing tweaks, 3%+ It affects Samsung more since Hynix uses tighter timings. Also Vulkan uses GCN well, it can use the 4096 shaders from the V64. I have tested both in World War Z, V56/V64 and the V64 is 7-8% faster at same clocks These Navi cards will be trading blows in the future, guaranteed. Also had a Nitro, beautiful card.

Fediuld:

I can tell you from my experience with V64 Nitro. The card out of the box was 12% faster than the V64 reference. By undervolting was able to keep it stable at 1700Mhz core clock (for benchmarks went to 1750), and 1100 HBM while was quiet. Performance to put in perspective, it beat royally my watercooled GTX1080 @ 2190. As for in relation to current cards, a reference V64 at 1576/945, keeps up with a RTX2070 @ 1905 been neck and neck of within up to 5 fps difference depending the games. So add an extra 13% on top for the V64 Nitro when fine tuned. Similarly here, given that the reference model is faster than the 2070, tweaking an AIB or watercooling, it will be within single digit % of the 2080 and RVII. Cards costing >50% more. Also from the AMD benchmarks, we saw some games like AC, the RDNA gets big perf boost over GCN where GCN was lacking.

If I'm not mistaken, Vega cards where power hungry and the blower fan sucks (pun intended), so it does make sense that the better cooling options offered by AIBs would make a big difference on air. But what I'm really interested in finding out, is how much of a performance difference should we expect between a watercooled reference design Navi GPU, and a watercooled aftermarket Navi GPU. From what I've heard, Navi is power restricted, so if the VRMs on the reference cards are good enough, the performance differences will mainly come from better cooling.

Yeah, I don't see how these cards have bad performance. It's a great 1440p card that's probably going to last some time. I'm more disappointed in the power draw and the prices, as I think it should have been 50$ cheaper which would make the custom AIB prices what is the current MSRP. Anyway, I'd wait for the refresh from NVIDIA and buy whatever is better, there's bound to be better drivers and probably a discount by that time.

Yeah it will be interesting to see how these stack up both against Vega (Since VII is still AMD's top performer.) and also NVIDIA's 2000 series with the new "Super" versions landing later this year. https://wccftech.com/nvidia-rtx-super-graphics-cards-msrp-leaked/ So possibly 800$ for the 2080 Super using a 2080 Ti chip, 600$ for the 2070 Super using a 2080 chip and 430$ for the 2060 Super using a 2070 Chip. (Non-overclock models for the chips though.) Will be interesting to see how things go and also how the cards will perform a few months down the line though that seems to be more of a thing for AMD and some larger changes in the driver altering things around though overall performance might be fairly similar. Makes for some competition though AMD seems to be aiming for 2020 for a more high-end version which should also see NVIDIA's move to 7nm ~thereabouts~ via Samsung I think it was from recent info? EDIT: Though depending on pricing both for reference and AIB price/performance could see some competition between the two as well with a higher base price from AMD possibly being a issue if NVIDIA can respin the existing chips for higher-end re-releases to counter. Well price/performance for one but also driver and software and features plus there's also Linux and Mac and not just Windows and how the situation looks on these platforms. But first things first, will be fun to see reviews of Navi here even if it's Little Navi and not the Big Daddy GPU chip just yet. 😛 (Possibly, suppose there's still some uncertainty around how this will all be handled and how Papa Navi of 2020 will differ from Teen Navi of 2019 here as the mid option ha ha.)

Grumpymangrumbling2019:

well you can buy a vega 64 for less and slide the power down to 180 watts and only loose single digit performance look. Im sorry but i dont agree with your opinion because it has no basis in reality. I can lower power usage of my vega 64 that costs less than these cards and still be more efficient and faster. These cards have no business existing. [youtube=CgyytnVDUqs] https://i.imgur.com/3S01xVF.jpg

Just to note, while AMD is known to ship their cards with high voltage, not all cards are the same with how they handle under volting. One card might be able to run close to 230w, but that does not mean every Vega 64 can. But let's also think of a couple different things with what this brings up. If we are talking about the value of undervolting or overclocking, this is already less tdp than Vega, so it may be possible to undervolt it more than what you could any Vega. If we're talking about overclocking, even with any voltage increase per clock you're still using less power. But now let's talk about people who run these stock that will not need to worry about undervolting, as not everyone does. Let's also note that this is AMD finally getting off of just a pure GCN base. This is a start, and as such it's going to take time to see a difference. Look at how little difference we saw from Pascal to Turing.

GCN or well the GPU's AMD has running with the GCN system felt like they hit a limit pretty quickly where it takes a large amount of voltage to sustain the higher clock speeds and then the die shrinks and improvements to the GPU architecture helped but the stock values are still pretty high and a normal user would either stick with stock or do very limited overclocking which the new Wattman changes have made it more in-depth and less locked for Vega but changing the power draw slider or changing seven separate voltage sliders yeah at most lowering down the power draw and then maybe trying the actual voltages but that's a smaller percentage of the total user base doing things like this with the card set to around 1.2v depending on model but a few tweaks and yeah 3 - 5% performance drop or less and of course binning and variances from one to the other allowing for 1.050 or even lower but I can't imagine most users would try shaving off 100 - 150mv just like that. It sounds like Navi is going for higher clock speeds too trying to boost near the 2.0 Ghz range even if no factors are limiting what it can do so I expect a similar higher stock setting for voltage even if 7nm here might allow for lower values than 14nm Vega though it could also allow for some nice undervolt results again but stock is usually how it gets benchmarked and compared and then we still have the larger chip for next year. Will be interesting to see how that goes, Wattman and improvements here will hopefully work better than Vega and VII on launch although third party models and later GPU's could see testing on newer drivers with further improvements to any software side issues. (Vega in particular got a number of changes but that also applies to the VII though for that I think it was primarily the launch drivers not doing quite as well with Wattman for overclocking results including voltage changes.) Competition wise though AMD running the GPU's for what they are worth almost hitting the limit at stock settings seems understandable and then power users can pop a water block on the card and start min-maxing the voltage and power draw values to find a spot where it boosts without capping the thermal or power limits and sustains it too. 🙂 (Which again I would think is a fairly small part of the user base overall though it's a interesting thing to be able to set these up like that although every single card is going to have at least some variance to what it can do and where it hits peak values.) Just how it feels like after the problematic stock 290 cards and then up to Vega which I guess even applies to the 7000 series and then revisions and releases with various improvements like 7000 to 7000 Ghz for a smaller one or the 290 to the 390 getting the thermals into a better stock state and so on, my view on it is a bit limited and it will be interesting to hear more on Navi here also in regards to just how much it would drop from lowering down the clock speed depending on where in the process it hits a limit or bottleneck although AMD has improved that too over time although Vega and the VII are more about sustaining the boost and trying to find a good balance for power draw and voltage since gains are otherwise fairly low outside benchmarking though not entirely non-existent either. 🙂 Assuming it can boost to 1.9 Ghz then 2.0 Ghz will probably be attempted pretty quickly although 1.7 - 1.8 seems like it's going to be where it lands without some tuning or getting the stock fan off and on a water block first but we'll just see in about two months then. Going to be interesting alright and from there to whatever is next whether it's a larger chip or a revision of the chip and architecture too. EDIT: Well hitting a limit or not I suppose is also due to AMD having the GPU's perform pretty close to what they can without fine tuning them via third party tools or Wattman now that it's also a option and has seen various changes over time. (Both clock speed wise and for GPU core voltage.)

5700xt Timespy leak. [spoiler]https://cdn.wccftech.com/wp-content/uploads/2019/06/5700xt-Time-Spy-Table.png [/spoiler]

Fox2232:

Navi is not Mix of GCN. You have no clue what you are writing about.

Didn't Lisa Su say this iteration of Navi is GCN with RDNA together?

Undervolting seems to help as well : https://www.hardwareluxx.de/index.php/artikel/hardware/grafikkarten/50149-zweimal-rdna-als-navi-die-radeon-rx-5700-und-radeon-rx-5700-xt-im-test.html?start=25 just need the drivers to allow for fanspeed alteration and both temperture and noise will be great again 😉