On the XFX R9 290X Double Dissipation product page something very curious appears when you scroll all the way down to the bottom…
What’s this image over here on the right, I wonder…
Well would you look at that. The box is clearly labeled for an AMD Radeon R9 390X with 8GB of GDDR5 memory, further indicating that the upcoming GPU will in fact be a Hawaii rebrand; and that the HBM-based flagship Fiji GPU we keep hearing about (and seeing pictures of) will have a new name. Whether that ends up being R9 490X or a name like “Fury” we will soon find out. As it is, it looks like we know at least part of what to expect from AMD’s gaming event at E3 on June 16.
Hmm. What might this be about??
Of course we will have complete coverage when any official announcement is made, but for now enjoy the accidental product reveal!
Update: XFX has removed the R9 390X images from their R9 290X DD product page, but not before numerous sites took their own screenshots before posting the news as well. There has been some disagreement about what the leaked photos actually reveal, or if anything has genuinely been "confirmed", but it seems likely that the product named 390X will be a rebranded 290X with 8GB of GDDR5.
how does this image mean this
how does this image mean this –> “…will in fact be a Hawaii rebrand…”
HBM supports max 4GB (in its
HBM supports max 4GB (in its current implementation). If it is an 8GB card, it cannot be Fiji with HBM.
I didn’t say it was fiji, I
I didn’t say it was fiji, I was questioning how they came to the conclusion that it is a 290x rebrand aka hawaii?
I see what you’re saying. I
I see what you’re saying. I thought the same thing.
It could be that enhanced Tonga (although I think that’s unlikely) It could be Fiji with GDDR5 :-S It could be another new chip.
Given there history with Re-rands it may be most likely that it will be a Hawaii core or an enhanced Hawaii Core. The title of the article presents that its a Hawaii card re-brand as a fact. Unless they have information that we do not (which is possible) Something that I would expect from WCCFTech not from PCPER. I would suggest a title more along the lines of “Accidentally leaked photos suggest 390X is Hawaii re’brand”
All in all the article is just a quick update so I’m not to mad about it xD
I don’t think it will be Fiji
I don’t think it will be Fiji with GDDR5. The memory controllers take up quite a lot of die area, and they would need to be totally different for using GDDR5 vs. HBM.
It could be there are two
It could be there are two different versions of the card, one with 8GB GDDR5 and another with 4 GB of HBM to get around the HBM memory restriction.
Because Fiji is capped at 4GB
Because Fiji is capped at 4GB of HBM memory not GDDR5.
I cannot wait to buy this new
I cannot wait to buy this new 4 year old card.
As long as the DX12/Vulkan
As long as the DX12/Vulkan support is there, and there have been tweaks to the technology, and as long as the card is selling for the same price, fine and dandy. AMD can only afford so much engineering resources, and that needs to go towards ZEN, as well as Fiji/whatever with the HBM. Sure the rebrands are disappointing, but for the prices folks are paying for AMD’s product, the available R&D budget is just not there for anything but the top end SKUs to get the latest and greatest microarchitectural reworks. The top tier AMD technology will work its way down into the other product stack, but that will take time and sales will have to be good, and plentiful, for AMD to make the revenues that can support a quicker level of technological improvement.
Hopefully they have made some
Hopefully they have made some power improvements to get the TDP down.
Not much can be done with
Not much can be done with same gpu to lower its power draw without lowering clocks.
Respinning can do a lot in
Respinning can do a lot in the right case. Compare the GTX 480 to the 580. A very power hungry GPU got a lot more efficient.
I hope that they ported it to
I hope that they ported it to a more advanced process, even if it is still a 28 nm node. They may be able to get better power consumption and higher clocks if they did.
Thats a good point. If they
Thats a good point. If they actually changed the process it wouldnt be a rebrand. I know TSMC has a bunch of different 28nm processes.
hmmm so their so called fury
hmmm so their so called fury card will be their titan equivalent if it can beat or match titan for 850 that would be amazing its sucks that the r9 390 will be a rebrand i wish they would use the tonga gpu on the new cards instead since tht was a great chip.
or this could also be a fake
or this could also be a fake rumor i got tricked with powercolors devil so i don’t really know soo hard to belive anything anymore
You say it would amazing if
You say it would amazing if its 850, but then you think about a gtx980ti that is only 7-10% slower then a titan and its 650$. That amazing goes away pretty quick.
oh right i forgot about the
oh right i forgot about the 980ti but people that usualy buy this cards always pay like an extra 100 dollars for that extra 1%.oh well lets see how well it does
I assume we will get a
I assume we will get a slightly cut down Fiji at some point (like Titan X/980Ti), not necessarily right at launch though. It would be possible that a cut down version would actually still have higher specs than an Nvidia Titan X. I could imagine a cut down version of Fuji competing with the Titan X at ~$650, while the full version of Fuji sells for more. If that actually happens, then it will be interesting to see what the Nvidia fanboys say.
Is it possible that the 390X
Is it possible that the 390X is not a 290X rebrand, but an up-spec Tonga based card with more shaders? Or is that going to be the 380X?
It could be either, I guess.
It could be either, I guess. Though the fact it was listed on the 290X page leads me to think Hawaii is sticking around.
With all this talk about HBM
With all this talk about HBM on the package, could you ask the people from AMD and Nvidia why they dont just START with off package HMC RAM instead?
The SPARC XIfx already has 480GB/s using 32GB of HMC. It seems like a better way to get memory bandwidth and lower TDPs than limiting yourself to on package exotic 2.5D RAM on an interposer.
I don’t know why you, or
I don’t know why you, or someone, has been bringing up HMC in threads talking about HBM. HMC is a totally different interface. It uses TSVs and micro-bumps to stack chips as HBM does, but the bottom logic die converts the interface to 16-bit, high-speed serial links rather than a low clocked 1024-bit interface. Pushing signals at that speed through the PCB is not going to be as low power as HBM. It will be lower than DDR4 for the same bandwidth though.
The amount of die area taken for the interface may be an issue also. HBM, with the relatively low clock and wide interfaces, is very efficient on die area (important when you are trying to fit 4096 shader units). It only has to push the signal a very short distance. HMC is supposed to simplify the memory controller, but it takes a lot of die area to drive off-chip interfaces. I was just looking at the die photo for Intel’s new 5th generation 4-core chip. The eDRAM controller looks like it takes almost as much space as about 2.5 CPU cores (not sure what interface is used though). I would think that running enough links of HMC for a gpu would take a lot of power (compared to HBM), a lot of die space, and be really expensive compared to HBM. I also just found a picture of a Xeon Phi package with pads for 8 HMC chips (16 GB). This ends up being very large because of the size of the HMC packages. They are huge compared to tiny HBM die.
HMC is also not a jedec standard, so I do not know if anyone outside of the Hybrid Memory Cube Consortium has access to it. I have not seen an exhaustive list, but I don’t think AMD or Nvidia are members.
There is a good comparison article on extremetech from a while back, but I do not seem to be able to put the link directly without triggering spam filters or something.
If you only need a tiny
If you only need a tiny amount of RAM then HBM sounds great but the Knights Landing Xeon Phi youre talking about has 16GB of proprietary HMC as “near” memory and up to 384GB DDR4 per socket for 400GB total RAM and 480GB/s combined bandwidth. Knights Hill might have more HMC but i dont know of the specifics are known. It will need to compete with GV100s 1TB/s bandwidth though.
The big market for the high end GPUs like GK110 and the upcoming GV100 is supercomputers. They need a lot of bandwidth and plenty of RAM so having ALL the RAM on on a CoWoS type thing might not be possible if you need a lot of RAM.
I dont follow AMD stuff much but if its only going to have 4GB of RAM that sounds like a terrible idea for a high end GPU when GDDR5 GPUs are using 4GB in the mid range and 6-12GB for high end. You might have the compute power and even decent memory bandwidth but if this thing has 4GB of RAM youll be limited by the amount of memory.
Thats why i was suggesting that either an off package HMC with 8-32GB would be easier. I know you can get off package HMC to 480GB/s aggregate bandwidth because its been out and deployed since 2014 on the SPARC64 XIfx.
You wouldnt need a near/far setup like Knights Landing for a consumer card and off package HMC is a much smaller footprint than GDDR5 although it is larger than HBM or Intel’s proprietary Knights Landing HMC on an interposer. And im pretty sure Nvidia is going to have up to 32GB of HBM on GP100. I think AMD might be making a mistake releasing HBM so soon although it will probably work great until it starts overflowing to system RAM if a game needs more than 4GB. I am assuming here that the AMD chip is having only 4GB of RAM and that people wanting a next generation GPU will be running higher than 1080p resolutions.
The memory limit is just for
The memory limit is just for the first iteration AMD wanted to get ahead of Nvidia so the 4GB limit was good enough. Next year will be HBM 2.0 on both Nvidia and AMD and they won’t have a 4GB limit.
Most games don’t even use more then 4GB of memory anyways so I doubt that limit will matter in the first place.
Seem to Remember a TON of
Seem to Remember a TON of people sayhing 4gb wasn’t enough when that is what 980/970 launched with. Yet Some how Since AMD is doing it, it is now enough? Gotta Love people and their Double standards some times.
Considering GTA5 and other
Considering GTA5 and other new games can use all 3GB on my 780 i really doubt that 4GB is enough for a NEXT generation GPU, which this thing is.
If you need the compute power this thing is likely to have youll probably be driving a 4K display or something and youll probably need more than 4GB. That means its going to use system RAM and start hitching all over the place.
Nvidias current lineup is basically a stripped down cheap single precision only GK architecture since they got stuck on 28nm but its got plenty of RAM. It would be smart to keep the memory off package until you can put enough on package.
I have seen some test where
I have seen some test where the memory used varied significantly between different implementations. This would imply that it can be optimzed significantly in drivers. New color compression and other features may make a big difference also.
If it isn’t enough, then consumers may have to make a trade-off. If it ends up hitching in a game you play, then you probably should buy a different card. Although, most of the situations where 4GB was exceeded seem to be cases where turning on the feature isn’t really worth the performance hit. At 4k, you aren’t going to notice aliasing as much with high pixel densities, so running a high level of AA is probably a waste. This depends on your set-up though. The 40″ 4k display reviewed recently actually has a relatively low pixel density due to the massive size. I have a 30″ display at 2560×1600, and it is almost too big for how close I like to sit. I think I would rather have a smaller size at higher density.
Current systems are
Current systems are incredibly wasteful on memory amount. As a software person, I would not be surprised that they could reduce memory consumption significantly with work on the drivers. I also expect DX12 to make better use of memory. We will just have to wait and see if it is an issue with existing DX11 games though. The situations where we run into memory amount limitations is quite small.
The memory savings of emerging technology will really come into play with APUs though. HSA allows the system to just pass pointers between the CPU and gpu rather than separating a fix amount of memory for gpu use and maintaining two copies. Current implementations cause the gpu memory to essentially just be a cache of system memory, so everything needs to be copied between them all the time. HBM plus HSA will allow an APU to perform very well, even if it has a small amount of memory.
“Thats why i was suggesting
“Thats why i was suggesting that either an off package HMC with 8-32GB would be easier. I know you can get off package HMC to 480GB/s aggregate bandwidth because its been out and deployed since 2014 on the SPARC64 XIfx.”
At what cost though? Also, Fuji may be significantly higher bandwidth than 480 GB/s. I have seen up to 640 GB/s rumors.
“You wouldnt need a near/far setup like Knights Landing for a consumer card and off package HMC is a much smaller footprint than GDDR5 although it is larger than HBM or Intel’s proprietary Knights Landing HMC on an interposer.”
It may be technically correct to call the knights landing package an interposer, although it probably has active circuitry, at least capacitors and such, rather than just passive interconnect. It is also a standard style package, which is similar to a small PCB. A silicon interposer is a totally different beast Routing all of the HMC links in addition to all of the off chip interconnect for knights landing is going to make for an expensive device.
What I’m saying is that you
What I’m saying is that you can have a huge amount of memory and memory bandwidth by using off package HMC.
The difference between 640 and 480GB/s would only be relevant if the GPU itself is bandwidth constrained by 480GB/s, which I doubt it would be. At that point you’d be running into a bottleneck from PCI-e and the CPU itself probably.
And I don’t know if it would be more expensive, both in terms of money or transistors for the controller, to have the HMC off package like the SPARC64 XIfx does. You certainly wouldn’t e limited to 4GB of RAM on an ultra high end GPU though.
It seems like AMD is just doing what they did with the 7970 compared to GK110. They released their 28nm chip first, and it ended up being significantly less powerful than Nvidia’s.
The direct comparison for AMD vs Nvidia of the last generation was the GPU in the 7970 and the GK110, since they’re both the “large” 384bit bus compute focused GPUs. Nvidia did so well with their new architecture that they ended up releasing the 560Ti successor chip as the 680, which competed nicely with AMD’s big GPU.
What do you mean it’s a totally different beast? Aren’t Knight’s Landing, GP100 and Fiji all on silicon interposers of some kind? Knight’s Landing will probably be over $5,000 per chip, but I don’t think Intel cares, since they’re going into supercomputers that cost hundreds of millions and are owned by governments or huge corporations who have no problem buying them in batches of 10,000.
And only Best Buy lol
And only Best Buy lol
HDMI Ready : 1.4a
Oops.
HDMI Ready : 1.4a
Oops.
Hmmm
The headline is simply
Hmmm
The headline is simply not true. These pictures simply infer that a card similar to the R9 290 series card is in the offing, nothing more. This isn’t just irresponsible reporting it is an outright lie. That’s right I am calling the person that wrote that headline a liar. And if that person and the website are not one and the same shame on the website as well. Unless of course they are receiving some sort of compensation from NVidia. In which case; job well done.
And by the way look up the industry definition of a “rebrand” It would be nigh on impossible for this to be a “rebrand”. It would appear that a “refresh” in the pipeline, but there is no definitive proof of that either. For all you know it could be a completely redesigned chip.
Don’t worry bud 😉 ignorance
Don’t worry bud 😉 ignorance is bliss.
It has been expected for
It has been expected for months now to be a re brand, only addition will be support for like DX12 and standard 8gb ram instead of base of 4gb.
If its the same chip its a
If its the same chip its a rebrand. Just like the 4GB 770 was a rebrand of the 680 with extra RAM.
To be fair, 770, had gpu
To be fair, 770, had gpu clock bump and memory clocks were bumped from 6ghz to 7ghz. AMD did rebadge of 7970 to 280x, they dropped gpu clock 50mhz. yea its only 50mhz but still. 770 got 40mhz gpu boost and 1ghz on memory.
Oh dear…..
not looking good
Oh dear…..
not looking good for AMD
plus they admitted 10 days B4 release they dont have a release Bios and stable drivers. we may not get cards until mid Q3 ish!
I so hope these are Tonga
I so hope these are Tonga based not Hawaii.
be nice if the second top of the top card form AMD actually used latest GCN (or at least next to latest)
If not, then I hope it is at least a silicon respin for lower power. If it it will be lower power at same performance I can see that working (line Nvidia 680 to 770 jump).
If it is just a sticker on the box and bios update, then FU AMD. One can be a fan for so long before jumping the ship to the competition. The only thing that might get the fanboys to stick is perhaps the Fury (or what ever it will end up being)
so true. I used all AMD CPU’s
so true. I used all AMD CPU’s but after my FX 6300. I just couldnt get a new one. and then when I saw rebrands and only 4GB HBM. I got a titan X
Do you feel ripped off having
Do you feel ripped off having paid $1000 for a Titan X, with the 980 Ti being almost exactly the same performance and hundreds of dollars cheaper? Even when the Titan X came out, I was already saying lets wait and see what AMD comes out with. HBM seems revolutionary, so I definitely want to see what it can do before spending $500-$1000.
The entire GM architecture is
The entire GM architecture is Nvidias equivalent of a Haswell refresh. The Titan Black, Titan Z, K6000, K40 and K80 are all more powerful and sophisticated GPUs than their GM replacements. The only thing GM is good for is single precision.
That being said, AMD is likely to have a flop of a card if it has only 4GB RAM. GP100 is already being pitched by Nvidia as their next REAL GPU. It will be the next GK110 which is probably one of the best architectures ever released since its still the most powerful after three years.
Well at least we know that it
Well at least we know that it is NOT Hawaii but Grenada…the difference remains to be seen.
No one noticed the thickness
No one noticed the thickness of the card in the picture?
290x rebrand or not at least
290x rebrand or not at least it matches consoles when it comes to performance memory (VRAM) amount.
I think that AMD could not
I think that AMD could not afford new boxes and have a bunch left over from their 290X.
:p
Here’s a theory. Red Herring.
Here’s a theory. Red Herring. AMD does have a very clever marketing team, its POSSIBLE that all these leaks are pure McGuffin created by AMD so that when they big reveal finally happens we will be shocked and pleased, like they aint re-brands at all but optimized tweaked and improved parts and the flagship is a game changer.
‘Cource I know nothing, just another kid in the peanut gallery guessing and waiting.
Um looking back over last 2
Um looking back over last 2 years of AMD’s marketing, Clever is the last word that should ever be used to describe them. Complete udder idiots is more proper.
Hey just cuz the ideas were
Hey just cuz the ideas were bad doesn’t mean they weren’t clever.
Not sure why you think its
Not sure why you think its clever to claim and market 1 thing then when product comes out its not everything they claimed. Sounds more like bait and switch.
before you try to claim nvidia did that with 970, that is 1 video card vs like 2-3 years of it.
well, from my POV, they AMD
well, from my POV, they AMD is at the disadvantage, in both the hardware and driver game, so the Marketers dont have much to work with. HOWEVER they have made some clever moves to try to keep the brand relevant and popular in the minds of the GENERAL public, the Rx-XXX re-brand, the whole “It’s not an update, it’s OMEGA” thing, the media events, product giveaways, sponsoring LAN/vLAN party’s, and yes, BAIT AND SWITCH. They’ve been given a challenge “Make this as shiny as posible” and they are trying like fuck to do so.
Here’s a scenario. 2 pot dealers, both get the same shitty weed. One tells his friends how shitty it is, sells it at cost. The other sells it to people who dont know better that he will never see again, tells them its the bomb and charges extra. I know who I’d rather give my business to, I know who I wana be friends with, but the CLEVER one is the thief, no?
Theres only one sustainable
Theres only one sustainable answer. Making the best possible product and getting repeat business. If they cheat people then theyll just go out of business. Unless theyre a US automaker.
The part of my senario is the
The part of my senario is the (hope against hope) that the products that AMD is working on right now, the next CPU cores, the next GPU cores, That those are going to make or break the company. Keeping the brand alive, through any-means-nessary is the only goal. If they truly are only treading water then get ready for INTEL to be the #2 gpu, and yea, have fun wit dat
AMD Rebrandeons, so exciting
AMD Rebrandeons, so exciting to buy.
R9 290x as 390x? What is this, does AMD plans to commit suicide already? Their current lineup even barely sells and they overhyped the 300 series for months.
Why did it take them too long to release the 300 series if they just going to rebrand the most of it?
AMD was waiting on 20nm to be
AMD was waiting on 20nm to be ready to use which never happened. Nvidia was at one point as well but they threw in the towel early on waiting. AMD gambled on 20nm and lost.
Man this stuff needs to just
Man this stuff needs to just stop……. Come on out with what you are doing AMD, or just get on with throwing your business and its practices back down the toilet!
Look at that image is just a
Look at that image is just a 3D rendering.
I sure you hope you are
I sure you hope you are willing to post a public apology if the 390X ends up having substantial updates,since you “confirmed” them being a direct Hawaii rebrand by a picture on a box.Sounds almost like libel to me.
AMD R9 390X Video
AMD R9 390X Video Leaked
Check it out here
https://youtu.be/WXialS068Fk