Whoa, lets keep it realistic. Best case game benchmark perhaps but on average it is nearer to 30% faster than a GTX980 and is dependant on resolution. At its lowest ~25% at 1080p and ~35% at 2160p.
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
**AMD Radeon 390X Graphics Card WCE**
- Thread starter Kaapstad
- Start date
Whoa, lets keep it realistic. Best case game benchmark perhaps but on average it is nearer to 30% faster than a GTX980 and is dependant on resolution. At its lowest ~25% at 1080p and ~35% at 2160p.
Soldato
There are a few months before the 390X hits, it's a bad idea to be giving out performance figures now, giving Nvidia plenty of time to sort out a competitor to combat it and spoil AMDs launch with leaked performance of the 980ti or whatever it's to be named.
The TX aint taking much market share at $1000, AMD has no need to worry about it.
The TX aint taking much market share at $1000, AMD has no need to worry about it.
If I was making a GPU that would spank my main rival's main top level product I'd be shouting it from the mountaintops. AMD's silence just tells me that they are't confident in their product for whatever reason so are keeping shtum about it until they've got it sorted.
Either way I've got a bad feeling about it.
Either way I've got a bad feeling about it.
Caporegime
Soldato
The 960, 970 and 980 have took plenty already
AMD should be really worried unless their happy with tiny market share. In which case all is well.. 
Ok trying to make some sense of it all.
The way I see it these are the important numbers.
770 294mm 230w Kepler
980 398mm 165w Maxwell
780ti 561mm 250w Kepler
titanX 601mm 250w Maxwell
280/x 352mm 250w Tahiti
285 359mm 190w Tonga
290x 438mm 290w Hawaii
390x ???mm ???w ?????
Now the 780ti was NVidia last large die, this is the chip that is being superseded by the Titan X. As we can see they have increased the die size by 40mm and the TDP has stayed the same, performance is about 35-45% depending on what your looking at.
The 980 is a much better chip than its predecessor the 770, yes the die size went up by 104mm but the performance was nearly double in some cases, all the time with a much lower TDP.
The there is AMD.
The 280 and 280x have the same TDP listed which I'm not so sure about but hey ho. The 285 is only 7mm bigger but has a good drop in TDP, performance wise in reality it isn't any faster.
The 290x has good room to manoeuvre in the die size area with only 438mm but nowhere to go in the TDP area.
Tonga shows us that AMD have been able to improve the TDP of the 285 while maintaining the same performance of the 280/x, we mustn't forget the smaller bus size there either. Can they utilise the Tonga improvements to keep the TDP of the 390 manageable while giving us an sizable improvement, they will need to as the 290X is 5-10% slower than the 780ti.
So the 390 could need to be 45-50% faster than the 290x to get anywhere near the Titan X and we have seen that NVidia finest could only squeeze 35-45% more out of the Titan X over the 780ti and that is without having to overly worry about the TDP and removing the DP compute functionality.
Of course AMD has the HBM card to play which might make loads of difference, it might not make much at all. To begin with the new memory technologies look great on paper, with massive theoretical performance, but sometimes the initial product isn't really any better than the outgoing best of the previous tech.
As always we will have to wait and see.
Disclaimer: it is all my own opinion and figures are taken from the wiki's so accuracy is not guaranteed.
The way I see it these are the important numbers.
770 294mm 230w Kepler
980 398mm 165w Maxwell
780ti 561mm 250w Kepler
titanX 601mm 250w Maxwell
280/x 352mm 250w Tahiti
285 359mm 190w Tonga
290x 438mm 290w Hawaii
390x ???mm ???w ?????
Now the 780ti was NVidia last large die, this is the chip that is being superseded by the Titan X. As we can see they have increased the die size by 40mm and the TDP has stayed the same, performance is about 35-45% depending on what your looking at.
The 980 is a much better chip than its predecessor the 770, yes the die size went up by 104mm but the performance was nearly double in some cases, all the time with a much lower TDP.
The there is AMD.
The 280 and 280x have the same TDP listed which I'm not so sure about but hey ho. The 285 is only 7mm bigger but has a good drop in TDP, performance wise in reality it isn't any faster.
The 290x has good room to manoeuvre in the die size area with only 438mm but nowhere to go in the TDP area.
Tonga shows us that AMD have been able to improve the TDP of the 285 while maintaining the same performance of the 280/x, we mustn't forget the smaller bus size there either. Can they utilise the Tonga improvements to keep the TDP of the 390 manageable while giving us an sizable improvement, they will need to as the 290X is 5-10% slower than the 780ti.
So the 390 could need to be 45-50% faster than the 290x to get anywhere near the Titan X and we have seen that NVidia finest could only squeeze 35-45% more out of the Titan X over the 780ti and that is without having to overly worry about the TDP and removing the DP compute functionality.
Of course AMD has the HBM card to play which might make loads of difference, it might not make much at all. To begin with the new memory technologies look great on paper, with massive theoretical performance, but sometimes the initial product isn't really any better than the outgoing best of the previous tech.
As always we will have to wait and see.
Disclaimer: it is all my own opinion and figures are taken from the wiki's so accuracy is not guaranteed.
Soldato
Have to agree with Boom, certainly as far as discreet graphics cards go Amds market share is diminishing and after this latest driver issue and the damp squib it turned into I'm really worried about them because if they go and Nvidia are left with the whole market to themselves then god help us all. Just typing Nvidia makes my skin crawl.
Soldato
If the 390X was a Titan-X smasher then I'd expect AMD to be making noises right about now (to stem the tide somewhat). The fact that they aren't leads me to believe the 20-25% figure over a 980 is fairly accurate and about what I'd expect
But it will be at least 50% cheaper than a Titan X and maybe the same price as a 980 or cheaper!!
Soldato
This just reeks of the wait for R600 all over again. Not good.
Caporegime
Ok trying to make some sense of it all.
The way I see it these are the important numbers.
770 294mm 230w Kepler
980 398mm 165w Maxwell
780ti 561mm 250w Kepler
titanX 601mm 250w Maxwell
280/x 352mm 250w Tahiti
285 359mm 190w Tonga
290x 438mm 290w Hawaii
390x ???mm ???w ?????
Now the 780ti was NVidia last large die, this is the chip that is being superseded by the Titan X. As we can see they have increased the die size by 40mm and the TDP has stayed the same, performance is about 35-45% depending on what your looking at.
The 980 is a much better chip than its predecessor the 770, yes the die size went up by 104mm but the performance was nearly double in some cases, all the time with a much lower TDP.
The there is AMD.
The 280 and 280x have the same TDP listed which I'm not so sure about but hey ho. The 285 is only 7mm bigger but has a good drop in TDP, performance wise in reality it isn't any faster.
The 290x has good room to manoeuvre in the die size area with only 438mm but nowhere to go in the TDP area.
Tonga shows us that AMD have been able to improve the TDP of the 285 while maintaining the same performance of the 280/x, we mustn't forget the smaller bus size there either. Can they utilise the Tonga improvements to keep the TDP of the 390 manageable while giving us an sizable improvement, they will need to as the 290X is 5-10% slower than the 780ti.
So the 390 could need to be 45-50% faster than the 290x to get anywhere near the Titan X and we have seen that NVidia finest could only squeeze 35-45% more out of the Titan X over the 780ti and that is without having to overly worry about the TDP and removing the DP compute functionality.
Of course AMD has the HBM card to play which might make loads of difference, it might not make much at all. To begin with the new memory technologies look great on paper, with massive theoretical performance, but sometimes the initial product isn't really any better than the outgoing best of the previous tech.
As always we will have to wait and see.
Disclaimer: it is all my own opinion and figures are taken from the wiki's so accuracy is not guaranteed.
285 190 Watts
280 210 Watts
280X 230 Watts
290 240 Watts
290X 270 Watts
Tonga actually has the power efficiency of Hawaii, remember the 290P (None X) is 30% faster than the 280X while using around the same power.
So Tonga is not a new benchmark in AMD efficiency.
AMD need to improve performance per watt by at least 30% (IE a 180 Watt 290X) or its not happening.
Why: because a 390X to be lets say "20% faster than a 980" it would have to be 30 to 40% faster than a 290X. without efficiency improvements it would be a 390 Watt GPU. thats not going to happen. PCB component costs too high, AIB Air cooler not possible.
Really, it can't be much more power demanding than the 290X, like 10 to 20 watts because thats the limit of what a good air cooler can handle.
So i would suggest if the 390X is happening AMD must have got the efficiency up by at least 30% as anything less than that is not enough in performance terms or would require too much power for a single GPU to run.
Last edited:
If the 390X was a Titan-X smasher then I'd expect AMD to be making noises right about now (to stem the tide somewhat). The fact that they aren't leads me to believe the 20-25% figure over a 980 is fairly accurate and about what I'd expect
Basic logic and physics will dictate it can't be. The maxwell design is very mature, very optimized, very efficient. It just isn't going to be possible to design soe thing that is substantially faster and constrained to the same power and heat levels.
Depends on CPU bottle necks at low resolutions, which is exactly why the 390X wot. Get close to its theoretical increase either. The Titan performs 38% faster at 4K, exactly down to spec.
The Titan won't but the 970 and 980s are, and the 390X won't gain AMD a significant amount either. AmpMD needs 970 and 980 competition, amongst other wholesale changes to the company.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
Basic logic, coupled with a complete lack of understanding of the architecture and you might come up with that conclusion, you are however completely wrong.
HBM offers both more bandwidth than the GM200 can achieve and lower power usage at the same time. Thus within a given power budget if you can drop memory usage from 80 to 40W, then you can up the gpu power 40W and have the same power budget.
Exactly how much power would be saved is basically impossible to quantify, AMD appear to be going with more bandwidth, but Nvidia have gone with more memory. Likely due to die size limits they also went with the less power efficient(but smaller memory controller size) 384bit bus, but with higher clock speeds. A 512bit bus would take more die space but could give similar bandwidth at say 5Ghz, lower voltage on both mem controller and memory itself. I'd think AMD saved a good 30-40W, possibly more.
Exactly, they will know more than anyone beyond AMD themselves.
Nvidia will haver chased hundred of AMD GPUs and subjected them to endless testing, made some very well informed projections on HBM based in their own internal testing, and using their own knowledge of the 28nm process.
There is only so much AMD can do while stuck on the same process.
http://www.overclock.net/t/1516264/...-390x-and-380x-in-early-2015/10#post_22933513
All I could find. I'm sure I have seen a proper source somewhere though.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
Difficult to know, more connections but at a smaller scale, wider bus but significantly lower clocks which means significantly slower running memory controller. SLower running means they can go much denser, higher speed, higher temps and higher voltage the more space you leave between transistors.
Some of the signal producing stuff within the memory controller will also be drastically reduced/simplified because it's not producing anywhere near as strong signals. I would be surprised if it was much bigger, I wouldn't be surprised if it was actually noticeably smaller for, lets say per GB/s in bandwidth.
Ultimately I don't have a clue, I've looked all over for that kind of information previously and not been able to find it talked about. It's certainly something I'll be very interested to find out.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
lol, I started answering this post ages ago, relatively similar idea to him though that is king of not trusted by anyone posting there