Once again, AMD whiffs on offering a CLC option. Vega64 had it at launch (heck VegaFE had it), but Radeon VII and Navi don't . . . ?
It is because 5700XT hits temperature problems(I thinknthisbis mentioned in the TPU review) and downclocks a lot.
Have EK design an AIO solution, call it the future edition, charge an extra $100 over msrp, solve all future review issues... This isn't hard AMD, someone already did all the work on HOW to do it, you get the opportunity to improve on it and still sell the crappy blowers as reference.Once again, AMD whiffs on offering a CLC option. Vega64 had it at launch (heck VegaFE had it), but Radeon VII and Navi don't . . . ?
I'm not talking about "losing ALU performance at higher clocks"? Huh?
I'm not saying Navi loses performance as it overclocks, just that perhaps it doesn't get quite as much extra gain per clock. We know different GPUs gain more from overclocks than others (irrefutable), and perhaps Navi is less than Turing.
At 1.5GHz Navi => Turing
At ~1.85Ghz Turing => Navi
An analysis based on the 3 sources we have both presented anyway. We need more to know, and aftermarket coolers to test higher clocks.
Once again, AMD whiffs on offering a CLC option. Vega64 had it at launch (heck VegaFE had it), but Radeon VII and Navi don't . . . ?
Or simply AIB partners GPUs will show better performance, and potentially - better efficiency, than AMD's own reference design.I'm not sure why I'm jumping into this, but assuming if what you've posted is true it would imply that Navi does have more performant shaders, but that the card is designed in such a way that a bottleneck prevents from operating at maximal effectiveness at higher frequencies.
It's really the same argument as using low resolutions (720p) to benchmark CPUs for gaming because higher resolutions result in a GPU bottleneck. I'm sure that as people play around with the card more they'll find where the limiting factors are, but it's always a question of whether or not it's something that AMD can easily fix. If like Polaris, Navi tends to benefit more from a memory OC then it's just a matter of pairing it with faster memory to solve the performance degradation. If it's some other hardware limitation it will require design tweaks from AMD that are unlikely to occur until the next generation of RDNA cards is released.
https://techteamgb.co.uk/2019/07/08...er&utm_campaign=Feed:+techteamgb+(TechteamGB)Has anyone got any info for the new features? Particularly the reduction of input latency? I've been waiting for what I assumed was a more robust review by AT, but I get the feeling what's up is the final version.
This is going to be depressing.Since this is Navi GPUs thread, its worth starting the talk about small Navi GPU.
Im sure that I am the only weirdo who was actually much more excited than for midrange chips. Considering that Navi has performance per ALU parity with Turing its worth looking at the possible core configuration for small Navi GPU. 20 CU's with 128 Bit GDDR6 memory bus. 224 GB/s memory bandwidth with 1280 ALUs. It will be equal to 1280 CUDA core Turing GPU, at certain clock speeds.
The closest thing to 1280 ALUs with Turing architecture is GTX 1660. Most of GTX 1660 GPUs clock to 1980 MHz, like here:
There are other versions of GTX 1660 which are clocking up to 1995 MHz:
AMD will clock small Navi to similar core clock like medium Navi has: 1980 MHz, which is the same level for GTX 1660.
In theory small Navi because of ALU count difference should be behind GTX 1660, however, there comes the matter of GDDR6.
Small Navi might have higher bandwidth available than 192 bit GTX 1660, which might affect performance and mitigate the difference between them.
Overall, previously I thought 5600/Small Navi will be between GTX 1660 and GTX 1660 Ti in performance. Right now I believe that Best case scenario for this GPU is being on par with GTX 1660.
The other factor of this is power draw/efficiency. I don't think we should expect anything better than 105-110W for whole board power draw, for highest-end SKU based on Small Navi GPU. I believe small Navi also will come with 8 GB of VRAM version, thanks to 2 GB GDDR6 chips, at a price premium compared to 4 GB version.
P.S. I would be actually very pleasantly surprised if somehow AMD would break the CU count and release for example 24 CU part, directly competing with GTX 1660 Ti. I would buy that straight away, for my build.
You forget that if Small Navi is slower than GTX 1660, AMD will have to price it lower. That GPU still will be on par, or slightly faster than RX 590. So RX 590 replacement, with much better efficiency, but possibly 80$ cheaper.This is going to be depressing.
A Polaris 20 performance replacement for the same or more money. Less, but more efficient shaders clocked higher for similar performance with less power used.
Who is spending more than a Polaris 20 to save a few 10's of watts?
You are using previous GCN GPUs logic for brand new AMD's architecture. You might be correct, but the reality is that its much more likely that AMD's GPUs will simply add GPU blocks in scalable way: 20, 40, 60, 80 CUs, etc...Anything is possible for the smaller Navi. AMD does not have any established pattern at all.
Polaris: 36 / 16 / 10 CUs
Fiji / Tonga (both GCN 3): 64 / 32 CUs (though Hawaii stuck around between)
Hawaii / Bonaire (both GCN 2 kinda): 44 / 14 CUs (though Pitcairn the immortal was between so that explains the larger segmentation, not to mention Tahiti sticking around until Tonga)
Tahiti / Pitcairn / Cape Verde (GCN 1): 32 / 20 / 10 CUs
If they do go with only 20 CUs (1280 ALUs) for it, then I hope they release a more cut down Navi 10 to fight the GTX 1660 Ti. Obviously margins will be smaller, but better than abandoning the ~$280 market imo. I want a 1920-2048 ALUs, 224-bit, 7GB please (they would never do that configuration, just would be fun to see a 7GB card). It would destroy the 1660 Ti and potentially the 2060 in one fell swoop.
RX 580 is pretty much 50% of a 5700XT in performance. I'm fairly certain that a 20CU 128bit memory bus Navi will also be 50%. Also RX 590 pricing was an aberration that confused everyone or maybe it was a good preparation for the Navi prices so that we can use it as a comparison to justify the new norm.You forget that if Small Navi is slower than GTX 1660, AMD will have to price it lower. That GPU still will be on par, or slightly faster than RX 590. So RX 590 replacement, with much better efficiency, but possibly 80$ cheaper.
Nvidia still did not released GTX 1650 Ti, for good reason.
P.S. You would not save few tens of W's, but almost 150W if it small Navi will be as I predicted(110W) .
Apologies if I missed this while skimming through the thread, but is there any ETA for the 3rd party non-blower versions of the 5700/5700XTs?
I think AMD had price drops pre-approved given the most likely course of action by Nvidia.
If I could tell the AMD cards were $30 overpriced, then AMD could.
AMD jacked up the price again, pissing off it's user base, and citing NV's greed as the reason (I also fully understand they need to make money, why I wasn't so bothered during HD 7970 and why I haven't been bothered with RX 5700). While I don't believe SUPER is going to blow Navi away, it's definitely has the potential to leave them holding the bag. If NV wanted to they can easily shift their prices down more so, and then use their brand recognition to cause damage to AMD while not having to sacrifice as much on margins. AMD probably is expecting this why they gave themselves SO MUCH wiggle room.
The sad thing here is, most of the NV people accepted the price hike because NV was giving them at the time the better product. Now people are accepting the price hike just because AMD is "competitive" with NV again.
~mid August
Prices are increasing because of this:The sad thing here is, most of the NV people accepted the price hike because NV was giving them at the time the better product. Now people are accepting the price hike just because AMD is "competitive" with NV again.
Of course we can, the problem is the increased power from the data movements.Prices are increasing because of this:
https://pbs.twimg.com/media/D_DVhRDVUAAk6L1.jpg:large
It has two consequences. Yielding large dies may become on smaller nodes completely unfeasible from financial point of view. Secondly: Prices will go up.
The only way we can have "normal" prices in future, and it appears today, also, is if we will have chiplet based dies. But how to make GAMING GPUs with this tech? at this point of tech, we cannot make gaming GPUs chiplet based, because the trick is to make games see all of the chiplets as one GPU.
And how will the game see it as one GPU?Of course we can, the problem is the increased power from the data movements.