Now that NV has moved on to GDDR6 for their 20xx series, I would have to assume prices/capacity for GDDR 5x would become more favorable. As such, it would be nice to see AMD let Polaris stretch it's legs.
It's sounds ridiculous to say (because I think the speculation in this thread suggesting a straight rebrand for the prebuilt market is likely correct) but maybe AMD don't want their Polaris cards getting too close to Vega 56 performance territory otherwise they'll never move another Vega card again and so will make paltry changes.
I believe GDDR6 is only 20% more expensive than GDDR5, so unless there's shortages or something (or maybe they'd like to have those chips for Navi, especially if the rumors of Navi coming out in the 1st half of the year hold true), it'd likely be best to go with it. It'd be fast enough that they could even go with a less wide memory bus and fewer chips, which could possibly offset the increased cost of the memory itself (and would mean simpler boards, probably even less power), all while offering a good bandwidth boost (say they go from 256 bit to 192, that's a 25% decrease, but going to say 12Gbps GDDR6 would be 50% faster, which even with a 25% narrower memory bus gives you 37.5% increased bandwidth; heck they could go to a 128bit bus with 14Gbps memory and they'd still see a 37.5% bandwidth increase over 8Gbps 256bit Polaris; although I'm not sure on density - meaning I don't know what bus width they could go to and still offer 8GB VRAM).
So more memory bandwidth, less power (per chip, possibly less wide bus), coupled with improvements in the GPU, and AMD will likely talk up the perf/W more than the outright performance improvements. Say they get a 12.5% performance improvement (with 7.5% coming from the increased bandwidth, 5% from improved clocks) but get a 20% improvement in perf/W, they'd want to focus on the latter.
I doubt that. AMD has likely already sold the bulk of the Vega cards they were going to (to gamers at least, they probably would like to see more in the pro market). And since this likely will have higher margins, AMD would probably not have too much issue if they sell much more of these than Vega cards. Unless they increase CU counts by like 25%, clocks by 15%, and get 60%+ increased memory bandwidth, Vega should be able to pretty easily outperform this.
They'd make paltry changes because doing anything more wouldn't be worth the cost right now. This chip would likely be about appeasing OEMs most of all. I personally have a hunch that its also setup so that it'll be their $200 and below product, while Navi stakes out the $200-400 range (depending on how good it is, it might even be able to go $300-500, and then they could probably price it more aggressively as needed to, with it likely moving down to take over the old Polaris range at some point, maybe 2020).
GDDR5x is essentially GDDR6 from micron. Eg. it's much more similar or actually almost identical to GDDR6 and has a big difference to GDDR5. AMD would have to create a gddr5x memory controller. At that point they will need a new die and new masks so they could just s well add more CUs. But for what? To compete with Vega 56? The BOM for sure would be smaller what would that be enough to make up for development costs? I doubt it. Navi will be out in about 1 year, just not enough time for this to be worth it.
If this actually happens, it will be another rebrand and they could use 10gbps gddr5, so 20% increase in BW + some higher clocks for maybe max 20% increase on 12 nm. But even that, why invest the money into such a product?
I'm not sure the differences are that big. Nvidia supported GDDR5X on the 1080 while the 1070 had GDDR5. Plus they already were putting in the work to make a GDDR6 controller. In fact it might have benefits for Navi for them to get some experience with GDDR6 prior to it. Its to appease OEMs likely more than any other reason. Pairing these with Ryzen 2000 series should be great combo for OEMs. And since OEMs probably won't have Ryzen 2 and Navi chips til Computex or later, there's time to sell these in that market.
GDDR6 potentially lets them reduce the memory bus and use less chips, which means smaller cards would be possible, maybe even half-height cards for low profile/SFF systems. Because OEMs need cards, and pairing AMD CPUs with AMD video cards will be good for AMD. While OEMs will be lower margin, they make up for it in revenue, which brings stability to AMD's finances. They outright said this somewhat recently (about them fine with lower margin markets when it lets them sustain solid revenue).
If a Polaris refresh uses GDDR6, that could help it quite a bit compared to the 580, and depending on price, it could sell very well at mid range. Probably not on level of Vega, not quite, but it would help bump up mid range performance for OEM sales and new buyers etc. Might even make a great mining card, though time will tell what memory is used and the actual performance.
Yeah. I've seen some people claim that Polaris is bandwidth limited quite a bit, which if that's true, upping the bandwidth could bring gains that outpace the clock speed improvements. It shouldn't be above $300 and likely would be $250 tops. Definitely shouldn't be pushing near Vega. Exactly, this is to make a good pairing with Ryzen 2000 series, for DIYers, but more importantly, OEMs.
Facebook
Twitter