I don't know if you're posting these statements to provoke arguments, but just in case you're ignorant of the actual world market for GPUs, I'll state the following.And AMD targeted those segments and that pricing because they had to, not because they wanted to. When Pitcarin launched >$300 launching a new card on a new node at a price that was essentially unlowerable was not a smart thing to do. The $200 480 was most definitely a bait and switch as those cards are nowhere to be found for instance -> $200 480 = marketing speak.
People will not pay whatever, they will pay up to what they think the card is worth. The problem is that Nvidia has been working hard to cause the consumer to overvalue their GPUs.
This is a major problem that AMD has to resolve. The problem is that Nvidia has all the brand recognition. If two products are more or less identical in terms of features and price the average consumer will choose the Nvidia option. Many of the people on this forum will do the same.
Nah, this trend has been going on since at least the unified shaders days. The amount of polygons, viewing distance and number of objects really didn't scale all that much compared to the amount and precision of shader effects that is applied today.Theoretical FLOPS generally mean very little between architectures as design tradeoffs greatly determine effective throughput. AMD has always been shader heavy, even to the point what that kind of throughput is unutilized (Tahiti, Fiji) due to bottlenecks elsewhere.
I agree that as time has gone on games have become more and more shader heavy (probably due to consoles).
I don't know if you're posting these statements to provoke arguments, but just in case you're ignorant of the actual world market for GPUs, I'll state the following.
The under $300 segment is about 80% of total sales with the under $200 segment a major part of that 80%.
If you want to increase market share and revenue, don't you think it wise to target those segments?
As some have been saying. The first priority of AMD is to stabilize itself financially by stopping losses and begin to generate some profits, however small that might be. Then you can look at competing and winning across all market segments as this takes more investments.For the short term, yes. But those largest segments of the dGPU market are the segments that have been steadily declining, probably due to quality iGPU options that are starting to overlap those lowest-end tiers and should be creeping into the midrange tiers over the next 2 years with Zen's APUs.
The only segment that is actually growing within the dGPU market (A market that, overall, is steadily shrinking) is the enthusiast/high end market. This is pretty much owned by nVidia right now because I think the Firepro class cards while great, don't represent a significant enough amount of units in that pro designer segment to make a real dent in the diversity that nVidia offers there.
AMD still has very far to climb. THe numbers a few weeks ago about them gaining ~10% in dGPUs is promising, but that is in the largest, yet shrinking segments of the market, on top of those numbers probably being attributed to nVidia simply shipping half the units they shipped compared to the previous quarter. I think to difference is something like AMD shipping an extra 100k units vs nVidia shipping 2-3mill less units than before.
That's a sobering reality.
This is why Vega is so important. I think there is still time for them to be competitive in that enthusiast class and gain some real market & revenue share and establish a nice chunk out of the only segment that will matter 4-5 years from now--but Vega has to deliver. I don't think they have any room to maneuver if they can't gain any performance wins or real performance value, but as the Nvidia partisans have said over and over again, Vega has to gain margin wins for AMD. Don't expect this to be some $300-350 minimum tier pricing if it is a great card that AMD can establish market share with. They will have to price it higher, and it has to be worth that price, imo. The good thing is that nVIdia has currently given them a ton of wiggle room to undercut and still grab some nice profits....but nVidia will have been there so long by the time Vega approaches that they just need to cut 1070/1080 price by ~$100 each and then toss the 1080ti into the field.
As some have been saying. The first priority of AMD is to stabilize itself financially by stopping losses and begin to generate some profits, however small that might be. Then you can look at competing and winning across all market segments as this takes more investments.
The $100-$150 card performance is much higher than iGPU levels. Sub $100 cards for sure have problems and these are not being designed anymore.
Not you here, but I see some thinking as if this is a sports team where you change the coach and some players and go on to dominate the next season from a previous disastrous one.
Folks, it doesn't work like this, so all the naive cries of small margins relative to Nvidia just exposes your ignorance about how these businesses operate. This is a long multi-year process that has just begun.
I don't see AMD being in a position to dominate the performance table before Navi. They will have to settle for better perf/$, which might not be so terrible, with Nvidia's high pricing allowing reasonable margins at lower prices. For this to work however, They MUST have competitors at ROUGHLY all performance levels to prevent Nvidia from selectively reducing prices on the only parts being made by AMD so as to stifle margins in that segment. By competing across all segments they prevent Nvidia from targeting them selectively without also doing some self damage.
In the end I predict the following.
Vega holding the line and offering better perf/$ than the Nvidia equivalent, but not the performance crown.
Navi is where the real gains in absolute performance will be made and AMD will try to achieve higher margins together with higher market share.
I dont see this happen.If 232mm2 256bit polaris cant beat 200mm2 192bit GTX1060 then vega will not beat GTX1080.I think the biggest issue for AMD has been the new 14nm process. I think because its generally designed to power their new Zen CPU it might not have been ready for GPU, thus why the Polaris cards all run at 1200Mhz up till 1350Mhz overclock, yet Nvidia seems to run at 1500, and many can overclock up till 1900.
I think with a more mature 14nm process and what AMD have learned from the Polaris architecture, they can win over Nvidia. Also consider that Nvidia's next GPU's are refreshes, these do not perform significantly better than the gpu's they replace, its usually smaller bumps in performance for the same price. Trying to court undecided buyers.
Going by die size alone is wrong because the new leaks put the clockspeeds for Vega much higher than Polaris. Even die size wise, Vega will be around 400mm2 and hold a bigger advantage compared to GP104 than Polaris to GP106.
12TFLOPs for professional cards would mean they are clocked at 1450Mhz+ and even higher for desktop versions, which would put them easily over 1080 but very likely still slower than Titan XP. Getting such high clocks in a few months time does seem pretty unlikely though.
Two stacks of HBM2 at 2 Gbps would draw 16W total, compared to 8 GB GDDR5 256 Bit with 37W power consumption from RX 480.Given current die size and perf/w characteristics, small Vega will need to be ~375mm2 and ~240 watts in game to keep up with the 1080. If small Vega is using HBM, then power requirements would drop by 15 watts or so.
AMD's claims of P10's efficiency improvements leading up to release was nowhere close to reality. AMD demonstrated only minimal perf/w gains throughout the entire 28nm process with each GCN iteration. Any improvements Vega will have in perf/w outside of HBM use will be (more than likely) minimal.
And last bit. People still believe that Vega will behave just like Polaris is. It will not, for simple reason. It is completely new architecture.
Two stacks of HBM2 at 2 Gbps would draw 16W total, compared to 8 GB GDDR5 256 Bit with 37W power consumption from RX 480.
This is completely ignored by some posters.Two stacks of HBM2 at 2 Gbps would draw 16W total, compared to 8 GB GDDR5 256 Bit with 37W power consumption from RX 480.
Polaris GPU alone at stock clocks draws around 110W of power. Now much more power would draw GPU with 2 memory controllers instead of 8 32 bit ones? Every unnecessary part, that is taken out of the die saves power.
And last bit. People still believe that Vega will behave just like Polaris is. It will not, for simple reason. It is completely new architecture.
Nowhere near this number - and 4GB vs 8GB RX 480 reviews show this:
I don't see how you can draw a conclusion based on anything in that chart. You do know the 4GB card doesn't actually have half the GDDR5 chips?
The PCB you're looking at is from the 4GB card. It uses Hynix memory, and only 4GB rather than a flashable 8GB, rather than Samsung currently housed on the two 8GB-equipped Nitros.
The Nitro's trick in keeping an enclosed heatsink is to locate the 8-pin power connector at the front, providing up to 225W of juice to the card. Remember we mentioned two models? The 4GB iteration is clocked in at 1,306MHz boost core and 7,000MHz memory while the 8GB OC raises the boost core to 1,342MHz and uses 8,000MHz-rated memory as standard.
Ok, and what does this have anything to do with 8GB HBM2 made from 2 stacks of 4 GB, and power consumption of it?Oh, I guess you missed the fact that this isn't the reference design:
Another website:
Less than 10W difference, and that's mostly because of slightly higher core/memory clocks on the 8GB model:
http://hexus.net/tech/reviews/graphics/94969-sapphire-radeon-rx-480-nitro-4gb-8gb-oc/
There is no hype train. Its only your assumption. Vega is Graphics IPv9, Tonga/Fiji was Graphics IPv8, Polaris was Graphics IPv8.1. Vega is the same graphics family with Greenland GPU, Raven Ridge APU GPUs. It is brand new architecture. What this means is that we know NOTHING at all.The hype train with AMD future releases never ceases to amaze me. There is simply no reality check that can slow it down.
You're still off. All your doing is comparing 8gbit GDDR5 to 4gbit GDDR5. This has nothing to do with the overall power consumption of GDDR5. It's the same number of chips....
Weird. I never said anything about the power consumption of GDDR5. But I guess you just need to deflect away from your original poorly thought out argument.Actually you're still off. My last post on this subject before this turns OT.
http://images.anandtech.com/doci/10675/P4Slide.png
Tesla P4 board. GP104 drawing 36W and only 20W for the rest of the board - including 8GB GDDR5.
So no, 8GB GDDR5 doesn't use 37W.
http://www.tomshardware.co.uk/amd-radeon-rx-480-power-measurements,review-33595.htmlAnother hint: We have 8GB GDDR5 Tesla P4 parts with GP104 at 50W. So no, the memory chips don't draw anywhere near 37W.
Sapphire Nitro 4 and 8 GB versions of GPUs have to use the same amount of memory chips, because both have to have 256bit memory buses. The difference in power consumption can come from... memory clocks.If the memory is supplied with power via the voltage converter block right next to the PCIe connector (which is fairly obvious), then the total power consumption should amount to approximately 37W.
If you would know anything about the Memory chips, GDDR5X consumes half as much power as GDDR5 does, while having 50% higher bandwidth.Actually you're still off. My last post on this subject before this turns OT.
http://images.anandtech.com/doci/10675/P4Slide.png
Tesla P4. GP104 drawing 36W and only 20W for the rest of the board - including 8GB GDDR5.
So no, 8GB GDDR5 doesn't use 37W.