Apparently, some people think that AMD will be releasing an RX 490 based on Polaris 10 with an extra four compute units, bringing the total number of stream processors to 2560. I'm guessing that people expected it to be a nice, round number or something, but that's not the case. According to Evan Groenke, Senior Product Manager at AMD, the die has 36 compute units, and there is “nothing else hidden on the product that end users might be looking forward to unlocking”.
Really, this kind-of makes sense. AMD seems to have designed this chip around the performance target of VR, which the RX 480 hits. I don't think that it would really make sense to push about 11% more compute processors into the design, decreasing their yield per wafer for such a relatively small gain.
We are expecting an RX 490 card to land at some point though, thanks to a mistake in publishing on AMD's part. It won't be Polaris 10 or 11.
I suspect 490 might be based
I suspect 490 might be based on Polaris XT rather than Polaris. Although I am waiting for after market 480s and better drivers and stocks in retail to straighten out.
Though I’ve seen rumors of
Though I’ve seen rumors of 4gb cards actually being ignored cards with 4gig disabled
AMD Retail Radeon RX 480 4GB to 8GB Memory Unlock Mod Works, We Benchmarked
https://www.techpowerup.com/223913/amd-retail-radeon-rx-480-4gb-to-8gb-memory-unlock-mod-works-we-benchmarked
No idea why, damn
No idea why, damn autocorrect
Meant: being 8gb cards with 4 gig disabled
Since there is no reason
Since there is no reason whatsoever to mount 8G on a 4G product this means that someone on the production line made a big mistake and got the 8g cards in the 4g pipeline somehow.
If you got one lucky you.
Actually there is a good
Actually there is a good reason to do so. The chips needed for the 4G product aren’t available in quantity today so to launch the 4G product on time AMD shipped them with 8GB instead of 4GB. That’s possible because both products use 8 chips. Don’t expect it to remain that way for long as there have been rumors that full production of the VRAM needed for the 4GB cards is due within the next few weeks, but for now people buying a 4GB RX 480 are likely getting 8GB with 4GB disabled in the BIOS. Just like has happened with R9 390s that had shaders disabled in the BIOS.
So at some point down the
So at some point down the road AMD is going to pull a Kingston and do a chip swap without a model number change?
That’s the thing to watch out for, are they going to send those cards out for review again? Heck they could end up being faster. Still it’s a shady practice to set the precedent of shipping a product for release/review with the intention of changing the components and possibly the performance at a later date.
AMD and NVIDIA and their
AMD and NVIDIA and their AIB’s have been doing this forever. You put in what works – whatever can hit the speed and efficiency target is a candidate to be installed on the board. Overclocking then may vary, but the product still ships able to do whatever it says it will. Hynix and Elpida were popular swaps on past cards from both companies. I’ve seen Micron GDDR5 as well. Sometimes a model of card (say, HD5870) had both kinds in the wild, extreme overclockers would seek out cards with the more overclockable chips because maybe one did 1.5 GHz (6 GHz) but the other was limited to 1.4 GHz or so overclocked. That made a difference in performance for people pushing the bleeding edge in overclocking competition.
Right, a brand new GPU
Right, a brand new GPU micro-architecture on a brand new 14nm process node and AMD is not over provisioning with more compute units/other resources so that RX 480 yields can be maintained at high enough levels to meet the demand for the RX 480. And then there is the question of some unused SKU slots in the naming scheme for Polaris. Maybe AMD does not want to tip their hand at this point in the game, but you know that there has to be an answer waiting in the wings to Nvidia’s GTX 1060 should it be priced aggressively, and Vega may not arrive in time for AMD to divert attention from AMD’s need to answer to Nvidia’s 1060. Vega is for the flagship GPUs anyways, while Polaris is for the mainstream. Go figure on what AMD will have in the hold to boost Polaris 10’s top end.
The answer is very simple. RX
The answer is very simple. RX 470. And if 1050 comes out, AMD will push 460.
With RX 480 AMD says to the people “No reason to spend more than $300 if you just want a VR card”. You can just spend $199-$249. Especially with the custom cards, many will see that there is no reason for them to go out and buy a GTX 980, a GTX 980Ti or a GTX 1070, or a R9 390/X, or Fury.
Nvidia probably didn’t wanted to bring GTX 1060 so fast in the market, but now that they know that AMD can mass produce Polaris GPUs easily,and flood the market with RX 480 cards, they have no other chance but to release the GTX 1060.
And when they do it, AMD will come out with RX 470 and will say the same thing to all those out there thinking to upgrade their older mid range cards. No reason to spend $250 for a GTX 1060 or an 8GB RX 480. If you just want 1080p gaming, here is a nice RX 470 at $149 – $179, with enough performance and memory to cover you, enough performance to be an ideal upgrade for someone with a R7 360 or a GTX 750Ti.
And Nvidia will have to respond, first by dropping prices of GTX 960 and then rushing out GTX 1050 at $150, only to see AMD coming once more with RX 460 at $100 and saying to the people. “You can do your job at $100. No reason to spend $150”.
AMD does have an advantage in low mid range market. It’s one step ahead of Nvidia. It wouldn’t last much, as Nvidia’s advantage in the hi end market – Nvidia will stay unchallenged for at least 6 months, but it will be enough to help them regain market share.
“over provisioning with more
“over provisioning with more compute units/other resources so that RX 480 yields can be maintained at high enough levels to meet the demand for the RX 480.”
No you have not read the post that you are replying to, the post is about having a sufficient supply of RX 480s, and not the lower binned 470 parts. It’s about what all the GPU makers do to assure that the top binned part has good enough die/wafer yields to assure sufficient supplies of the RX 480/other parts at time of first release on a new 14nm/other new process node. There is always some over provisioning of compute resources to assure that the highest end SKU(in this case the RX 480 mid range best part) at the time of its release has sufficient yields. No one makes a GPU part without some over provisioning, and this is especially true where a new 14nm fabrication process node is concerned.
The main reason that AMD/others choose to not speak about this practice is due to the nature of their market end users! So there is a good reason for not upsetting the gaming primitives as they are not very bright as a whole and they bring the average IQ of the knowledgeable gamers/games market down to the level of lowest common denominator(LCD). This is an internal engineering practice that all processor makers use to keep their die/wafer yields from causing too much loss, and is especially useful for the GPUs that have many more execution units than any CPU.
All GPU SKUs are binned with respect to over provisioning, no matter what the top market range part(RX 480/other GPUs) is at first introduction of that part/SKU. There is always some extra hardware compute provisioning to make up for defective units. When the process node becomes more mature the yields will become better, and that’s when you will see a higher than RX 480 part from AMD when the bins fill up with parts that have more working execution resources than the RX480.
Both Nvidia and AMD/others do this so what’s the big deal, but LCD FanBoys are that part of the market with the single digit IQs and they can not be rationally dealt with.
Yes I did. You didn’t
Yes I did. You didn’t realized what I said, or just don’t like my thoughts about AMD’s planning. I see AMD trying to take more and more market share moving to lower models, one step ahead of Nvidia in every turn. On the other hand, if I understood you correctly, you are expecting a GPU with more stream processors hidden in a drawer to be revealed when 1060 comes out. It’s not going to happen. AMD doesn’t have the money for that. Next GPU from AMD will be Vega. Until then we will have the custom GTX 480s and probably latter, after GlobalFoundries have improved their process, an RX 485 with higher frequencies, lower or the same power consumption based on the same Polaris 10 GPU with the same number of stream processors.
VR is 90fps at 1080×1200 time
VR is 90fps at 1080×1200 time two… But here is the kicker, because the pixel resolution is low, you need massive AA to compensate.
The RX 480 is bad choice for VR, for multiple reasons.
One example is project car… The RX 480 is not enough.
Same goes with all future AAA titles.
A faster card is ALWAYS nicer for VR as you can better sustain 90fps, lower latency, and turn on better Anti Aliasing.
The RX 480 might be fine for luckys tale or some other dinky VR game.. but for AAA titles I think the GTX 1070 is the minimum.
A 1.6ghz polaris with 2560 SP would actually be a decent card for VR. But clearly AMD think its overkill.
And BTW, who buys a $200 GPU for a $600 to $800 VR headset display ?
AMD pricing/marketing makes no sense.
You certainly don’t need a
You certainly don’t need a titan x class card like the 1070 for VR. I use a 290x in project cars and it’s flawless. The current standard is that all vive and rift games must operate fully at minimum specs. 90fps and all. I get that on my 290x.
“And BTW, who buys a $200 GPU
“And BTW, who buys a $200 GPU for a $600 to $800 VR headset display ?”
Somebody waiting for the price of headsets to go down after more people buy $200-300 GPUs.
It isn’t that uncommon to run
It isn’t that uncommon to run a display more expensive than the video card. I consider it smart to invest in a more expensive, higer quality display. You could easily keep a display for many years. A gaming video card will be obsolete quickly. A VR headset is a display.
The current target is the 970
The current target is the 970 really, due to the large installed base at that level and above. A 290/390 and several other older cards are close enough to work perfectly fine also. The 390 compares quite well to a 970 with DX12 titles. A 1070 is expensive overkill for VR right now. Also, it isn’t exactly uncommon for people to have an expensive display with a cheaper video card. No matter what Oculus wants us to believe, a VR headset is a display. How much is a good g-sync display? There are probably a lot of people running a display that is around twice as much as there video card. I consider it smart to spend more on the display and less on the video card. A video card can be obsolete in a year or two. A good display can last for a very long time, much longer than the video card. Also, a good video card connected to a cheap display will still look bad.
Even Sapphire got a 490 8GB
Even Sapphire got a 490 8GB GDDR5 listed:
https://www.reddit.com/r/Amd/comments/4rgoz4/look_what_i_found_while_doing_a_sapphire_ticket/
AMD design not manufacture,
AMD design not manufacture, their designs are sold to third parties that build them. So from their perspective they might not know, but this raises an interesting question. At the manufacturing level who is calling the shots? Both Nvidia and and AMD share a number of sources that release the reference designs. Example XFX, MSI etc…