• 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 R9 390X Arrives In 1H 2015 – May Feature “Hydra” Liquid Cooling

Nvidia certainly know how to work their customers into a frenzy, overcharge and under spec for a good few years, then become heroes when you offer the same performance at the same price of the competitions almost year old tech.

Gregster said:
So the next iteration of AMD cards is not until 2015? In terms of tech, that is a long time away.
Click to expand...

Plenty waited for the 6 series.

Plenty waited for Titan/780 performance levels at sane pricing too.
 
tommybhoy said:
Nvidia certainly know how to work their customers into a frenzy, overcharge and under spec for a good few years, then become heroes when you offer the same performance at the same price of the competitions almost year old tech.
Click to expand...

They offer a lot more post-sale than AMD though.

AMD just want to sell you a bit of hardware and then cry about how the world is against them and how your poor experience is all NVidia's fault, whilst leaving a lot of features down to 3rd parties to develop.

NVidia might initially overcharge but you'll never feel like they've simply took your money and run.
 
Man, that's a long way off. It's also looking like a pretty big jump.

I was (and am) really hoping for a revision of the 290 that includes HDMI 2.0 and/or displayport 1.3, and crossfires nicely with the existing cards. Perhaps better cooling.

What are the chances that they're lining up something like that in the interim?
 
Boomstick777 said:
Shouldn't it be called the R9 295 then? 285X can't come in cheaper than R9 290 and beat it in performance, they would just be cannibalizing the 29X sales..

285X won't compete with the 970, AMD will release a 295 / 295X or 390 / 390X to compete with / beat Nvidia's new cards.
Click to expand...

I can see where humbugs coming from with a 285x, i dont think amd need to worry about it hurting 290 sales, im sure the 970 has already done that
 
humbug said:
The 970 is an average of about 10% faster than the 290, thats the reference one with a 947Mhz clock.

The 285X would have to be 5% faster than the 290 to compete with the GTX 970.

The 285 is 0% to 20% faster than the 280 @ 1792SP, 918Mhz, 32 ROP's, 256Bit.

With 2048SP, 1050Mhz, 48 ROP's, 384Bit it should be at least 30% faster than the 280X, it could be as much as 50%.

It might just be possible.
Click to expand...

You have some weird ass ideas about scaling.

500mm^2+ Fiji wouldn't be more than 5-10% faster than the 290X

but a Tonga with some extra bus width, few more ROPs and less than 15% more SPs is suddenly going to be 50% better than a 280X.

Full tonga even if it had a 384bit bus and 48 ROPs isn't even going to come close matching the 290 and 970. At best it would be a 960 competitor assuming the 960 gets 10SMMs. And while it would do that its power consumption would increase by quite a lot. That said I very much doubt tonga is a 384bit bus GPU because the PCB designs it has so far gotten have been for 256bit ones.
 
What do you think the yields are on TSMC's early 16nm? Not long ago I saw an article saying GF's 20nm was 10%. Is it possible to cut any kind of deal with the foundry to make such yields profitable? In the past we have heard about Nvidia having an arrangement whereby they paid only for good wafers or somesuch.
 
_Alatar_ said:
You have some weird ass ideas about scaling.

500mm^2+ Fiji wouldn't be more than 5-10% faster than the 290X

but a Tonga with some extra bus width, few more ROPs and less than 15% more SPs is suddenly going to be 50% better than a 280X.

Full tonga even if it had a 384bit bus and 48 ROPs isn't even going to come close matching the 290 and 970. At best it would be a 960 competitor assuming the 960 gets 10SMMs. And while it would do that its power consumption would increase by quite a lot. That said I very much doubt tonga is a 384bit bus GPU because the PCB designs it has so far gotten have been for 256bit ones.
Click to expand...

you're ignoring the performance more ROP's and a wider bus adds to the performance.

Pitcairn Pro: 1024 SP's, 256Bit, 32 ROP's (100%)
Pitcairn XT: 1280 SP's, 256Bit, 32 ROP's (107%)
Tahiti LE: 1536 SPs, 256Bit, 32 ROP's (115%)
Tahiti Pro: 1792 SP's, 384Bit, 32 ROP's (150%)
Tahiti XT: 2048 SP's, 384Bit, 32 ROP's (160%)
Hawaii Pro: 2560 SP's, 512Bit, 64 ROP's (210%)
Hawaii XT: 2816 SP's, 512Bit, 64 ROP's (220%)

You're also ignoring the performance texture compression adds.

Tahiti Pro: 1792 SP's, 384Bit, 32 ROP's (100%)
Tonga Pro: 1792 SP's, 256Bit, 32 ROP's (100% - 120%)
Tonga XT: 2048 SP's, 384Bit, 48 ROP's (130% - 150%) add on a clock rate of 1050Mhz vs 918Mhz (+14%) adds another +7% = (137% - 157%)

You're comparing an Apple to an Orange and scaling them, i'm taking the Orange into account.
 
Last edited:
Can I ask a semi serious Question.

What is going to be the next big arguing point, memory bus width being the currant one?

Bear in mind that the 390X or the 490x will probably have this stacked memory tech and because of that only need a much smaller bus width.
 
Kaapstad said:
I did think this idea myself and made several posts about it, unfortunately for AMD when you see Maxwells TDP and efficiency for the number of transistors used it totally blows away anything the red team have got at the moment. A 290X packs 6.2 billion and a GK110 card 7 billion, a 980 outperforms both and only needs 5.2 billion transistors to do it. This is a killer before we get into heat or power consumption.

Until AMD come up with a completely new architecture it is a case of

Anything AMD can do Maxwell can do better.

Maxwell efficiency is the killer it is a bit like engine tuning, to increase horsepower you need to increase efficiency as well (unless you want to burn loads of fuel).
Click to expand...

To put Maxwell in a little perspective. The 680/770 has 3.5B tansistors. The 980 is an evolution of that 'lite' architecture, +50% transistors for +40-50% performance. Some of those transistors (a good amount) will have been spent on finer power managment. However possibly fewer redundant structures/transistors required compared to the 770. Some perf gain coming from slightly higher clocks.

The GK110 packed 7B. It had 2x the transistors than the 680/770 for only +30% (edit: actually 40-45%) gaming performance. This is an awful and highly inferior architecture using some of the reasoning shown by ppl in this subforum. It is no surprise the 980 can outperform it using only 5.2B transistors. It should however give some indication of how much DP compute takes up. It also gives insight to the challenge facing nVidia with a full fat Maxwell core and why the lite maxwell 980 exists.
GCN maintains stong DP compute and will carry that transistor burden unless they decide to develop their own lite arch.
 
Last edited:
eddyr said:
To put Maxwell in a little perspective. The 680/770 has 3.5B tansistors. The 980 is an evolution of that 'lite' architecture, +50% transistors for +40-50% performance. Some of those transistors (a good amount) will have been spent on finer power managment. However possibly fewer redundant structures/transistors required compared to the 770. Some perf gain coming from slightly higher clocks.

The GK110 packed 7B. It had 2x the transistors than the 680/770 for only +30% gaming performance. This is an awful and highly inferior architecture using some of the reasoning shown by ppl in this subforum. It is no surprise the 980 can outperform it using only 5.2B transistors. It should however give some indication of how much DP compute takes up. It also gives insight to the challenge facing nVidia with a full fat Maxwell core and why the lite maxwell 980 exists.
GCN maintains stong DP compute and will carry that transistor burden unless they decide to develop their own lite arch.
Click to expand...


Gaming performance going from GK104 to GK110 scales almost perfectly, unfortunately a lot of people who review these cards end up CPU bottlenecking the faster GPUs resulting in poor figures.

Fortunately I have a good collection of GK104 and 110 GPUs I can test myself.:D
 
bru said:
Can I ask a semi serious Question.

What is going to be the next big arguing point, memory bus width being the currant one?

Bear in mind that the 390X or the 490x will probably have this stacked memory tech and because of that only need a much smaller bus width.
Click to expand...

The Bus width on HBM is stacked, for example...

Standard Bus:
256Bit @ 6000Mhz = 192GB/s (2GB, 4GB, 8GB Layout)
384Bit @ 6000Mhz = 288GB/s (3GB, 6GB, 12GB Layout)
512Bit @ 6000Mhz = 384GB/s (4GB, 8GB, 16GB Layout)

HBM 2+1 Stack
256Bit x2 (512bit effective) @ 6000Mhz = 384GB/s (2GB, 4GB, 8GB Layout)
384Bit x2 (768bit effective) @ 6000Mhz = 576GB/s (3GB, 6GB, 12GB Layout)
512Bit x2 (1024bit effective) @ 6000Mhz = 768GB/s (4GB, 8GB, 16GB Layout)

HBM 4+1 Stack
256Bit x4 (1024bit effective) @ 6000Mhz = 768GB/s (2GB, 4GB, 8GB Layout)
384Bit x4 (1536bit effective) @ 6000Mhz = 1152GB/s (3GB, 6GB, 12GB Layout)
512Bit x4 (2048bit effective) @ 6000Mhz = 1536GB/s (4GB, 8GB, 16GB Layout)
 
Last edited:
humbug said:
The 970 is an average of about 10% faster than the 290, thats the reference one with a 947Mhz clock.

The 285X would have to be 5% faster than the 290 to compete with the GTX 970.

The 285 is 0% to 20% faster than the 280 @ 1792SP, 918Mhz, 32 ROP's, 256Bit.

With 2048SP, 1050Mhz, 48 ROP's, 384Bit it should be at least 30% faster than the 280X, it could be as much as 50%.

It might just be possible.
Click to expand...

30% is on the really high side. Memory bandwidth isn't going to increase the performance of the 285 that much. If the launch of the gtx 970/980 taught us anything, color compression can make up for a small bus size. And the r9 285 already has that going for it.

ROPs can help performance, but I think Shaders are overall far more important. ROP's become more important once you start increasing the power and amount of work the shaders can do.

This is why you will why ROPs are more important for Nvidia's architecture compared to AMD's architecture and why you won't see such a boost from AMD when they add 50% more shaders.

Clock for Clock, Nvidia maxwell cores perform about 30-40% better than their old keplar ones. Considering hawaii and the gtx 780 ti are pretty similar in core efficiency, I could see Nvidias core still being 30% higher than Tongas. Add in Nvidias ability to achieve about 30% higher clocks than a tonga card and this is why Nvidia's needs their ROPs. They simply have way more shader power at their disposal and adding ROPs to full tonga isn't going to have a big effect.

Considering there wasn't that many big changes between hawaii and Tonga this time around and Hawaii is certainly not ROP and bandwidth starved and the drop in the number of shaders for Tonga, I could see this r9 380x or full tonga performing about 80% of what a full hawaii chip is capable of.
 
thecorrector said:
30% is on the really high side. Memory bandwidth isn't going to increase the performance of the 285 that much. If the launch of the gtx 970/980 taught us anything, color compression can make up for a small bus size. And the r9 285 already has that going for it.

ROPs can help performance, but I think Shaders are overall far more important. ROP's become more important once you start increasing the power and amount of work the shaders can do.

This is why you will why ROPs are more important for Nvidia's architecture compared to AMD's architecture and why you won't see such a boost from AMD when they add 50% more shaders.

Clock for Clock, Nvidia maxwell cores perform about 30-40% better than their old keplar ones. Considering hawaii and the gtx 780 ti are pretty similar in core efficiency, I could see Nvidias core still being 30% higher than Tongas. Add in Nvidias ability to achieve about 30% higher clocks than a tonga card and this is why Nvidia's needs their ROPs. They simply have way more shader power at their disposal and adding ROPs to full tonga isn't going to have a big effect.

Considering there wasn't that many big changes between hawaii and Tonga this time around and Hawaii is certainly not ROP and bandwidth starved and the drop in the number of shaders for Tonga, I could see this r9 380x or full tonga performing about 80% of what a full hawaii chip is capable of.
Click to expand...

Maxwell has the same shader scaling problem as Kepler and GCN 1
The 980 has 20% more shaders than the GTX 970 and yet is only 10% faster.

Having Texture compression does not suddenly remove the need for memory and raster performance, it is not suddenly all on the shaders alone.

Nvidia can only grow a 560mm^2 2816 SP Titan-X by 20% performance with shaders alone, i think you will find that they will need a wider bus and more ROP's to make up the rest.

A GPU's performance is made up of different components, simply by improving one or two does not make the remaining one scale 100%.
It just means those two aspects can perform the same or better with less, as we see in Tonga and Maxwell, you still need to scale all aspects in tandem to get 100% scaling.
 
You must log in or register to reply here.
Back
Top Bottom