390X coming soon few weeks

Even if the consoles were break even it advances AMD's technology and gives them more leverage with the suppliers involved which likely helps all of their business areas. My company does this sometimes if it's strategically important.
 
Dayaks said:
Even if the consoles were break even it advances AMD's technology and gives them more leverage with the suppliers involved which likely helps all of their business areas. My company does this sometimes if it's strategically important.
Click to expand...

Don't forget it puts AMD into software, granted mostly just gaming, but it does get some optimization in AMDs direction. I still believe a great of performance can be had with optimizations that thus far haven't been made, although expected considering market share.
 
Dayaks said:
Even if the consoles were break even it advances AMD's technology and gives them more leverage with the suppliers involved which likely helps all of their business areas. My company does this sometimes if it's strategically important.
Click to expand...


AMD's full blown hUMA hasn't come into affect though so it's kind of minimal the effect of advancing AMD's strategy as it'll be largely useless on these current gen consoles. Next generation though will be very good for them, but that's a ways off.

It's not like AMD had to fight for all the console makers and they wont probably for the next generation either unless they royally screw up in the meantime. Nvidia really pissed in that cereal.
 
Liger88 said:
AMD's full blown hUMA hasn't come into affect though so it's kind of minimal the effect of advancing AMD's strategy as it'll be largely useless on these current gen consoles. Next generation though will be very good for them, but that's a ways off.

It's not like AMD had to fight for all the console makers and they wont probably for the next generation either unless they royally screw up in the meantime. Nvidia really pissed in that cereal.
Click to expand...

Makes me think: have we any concrete proof the consoles aren't using/nor capable of HUMA/HSA? All I ever saw was debates on the compliance of HSA in both consoles but no mention of any game utilizing it, as of yet.
 
durquavian said:
Makes me think: have we any concrete proof the consoles aren't using/nor capable of HUMA/HSA? All I ever saw was debates on the compliance of HSA in both consoles but no mention of any game utilizing it, as of yet.
Click to expand...

The PS4 supports a unified memory pool, so devs are using it there. The Xbox One doesn't, due to its DDR3/ESRAM combo.
 
You don't need a unified memory pool to support or do hUMA or HSA though.

The whole point of the hUMA/HSA memory model is to hide latency and different memory pools from the software or IOW to make them appear unified to the software when they aren't in reality. This eases development of software and can improve performance if development is done properly to take advantage of the feature.

durquavian said:
Makes me think: have we any concrete proof the consoles aren't using/nor capable of HUMA/HSA?
Click to expand...
They both have some form of hUMA/HSA since it was built into the GPU's they're using by default. Publicly I don't think anyone knows exactly how good either implementation is though but I believe its safe to assume that it won't be as good as the version that AMD released in Carrizo. Doing a unified memory model in hardware is hard and lots of effort and time are required to get it right.
 
mesyn191 said:
You don't need a unified memory pool to support or do hUMA or HSA though.

The whole point of the hUMA/HSA memory model is to hide latency and different memory pools from the software or IOW to make them appear unified to the software when they aren't in reality.
Click to expand...

So you don't need a unified memory pool to do hUMA (which stands for Heterogeneous Unified Memory Architecture) but then the whole point of hUMA/HSA is to make memory appear unified? It sounds like you do need a unified memory pool, whether it's physically or virtually unified. If I'm getting any of that wrong, please let me know.
 
Ozymandias said:
So you don't need a unified memory pool to do hUMA (which stands for Heterogeneous Unified Memory Architecture) but then the whole point of hUMA/HSA is to make memory appear unified?
Click to expand...
A unified memory pool is generally when all your memory is physically all one big 'lump'. Cache will confuse the issue some but its generally left out of the discussion for practical purposes.

hUMA/HSA is a means of using hardware to hide the differences (both in terms of address and latency) of different pools of memory (say system RAM and VRAM) from the software. It can make the VRAM and system RAM and yes even eDRAM or SRAM -----appear----- to be one big unified pool of memory to the software. Within certain limitations of course. What limitations though? Depends on the hardware associated with hUMA/HSA plus the memory pool. What that would work out to exactly for say the xb1 in terms of performance I don't know...but just because it has DDR3/ESRAM doesn't mean it can't do hUMA/HSA.

One is a physical description and one is a form of virtualization. They both get tossed around interchagably in the forums and media, even media from AMD, but they're 2 different things from a technical perspective. Both in terms of implementation and practical effect.
 
mesyn191 said:
A unified memory pool is generally when all your memory is physically all one big 'lump'. Cache will confuse the issue some but its generally left out of the discussion for practical purposes.

hUMA/HSA is a means of using hardware to hide the differences (both in terms of address and latency) of different pools of memory (say system RAM and VRAM) from the software. It can make the VRAM and system RAM and yes even eDRAM or SRAM -----appear----- to be one big unified pool of memory to the software. Within certain limitations of course. What limitations though? Depends on the hardware associated with hUMA/HSA plus the memory pool. What that would work out to exactly for say the xb1 in terms of performance I don't know...but just because it has DDR3/ESRAM doesn't mean it can't do hUMA/HSA.

One is a physical description and one is a form of virtualization. They both get tossed around interchagably in the forums and media, even media from AMD, but they're 2 different things from a technical perspective. Both in terms of implementation and practical effect.
Click to expand...

The other thing is the ability to allow both cpu and gpu to access bits of each others "playground".. And this used to be such a nono, where you now can gain the something they did on consoles "ages" ago as manipulating single memory points which normal would require a copy of the file being transferred back and forth between ram->GDDR.
 
Pieter3dnow said:
The other thing is the ability to allow both cpu and gpu to access bits of each others "playground".. And this used to be such a nono, where you now can gain the something they did on consoles "ages" ago as manipulating single memory points which normal would require a copy of the file being transferred back and forth between ram->GDDR.
Click to expand...

Actually there's nothing preventing the hardware from doing the transfer back and forth and simply hiding that from the programmer (though that would be a less ideal implementation). HSA is a programming model and the actual implementation can be abstracted away with it.
 
hUMA/HSA is hardware and software. That is why AMD says Carrizo is their 1st true HSA compliant APU even though they've had some of the hardware in their older GPU's and APU's for quite a while now.

Its true you can sort of do what it is supposed to accomplish without either the hardware or software, its been done before, but the difficulty in making the software goes up dramatically while also penalizing your software in terms of performance. The whole point of hUMA/HSA is to make things easier.
 
The Mac said:
wut?
Click to expand...

Translation: "A few weeks" means 20 weeks now which is how long it'll be when the 390x finally gets released. This thread was posted in January, 390x is coming out in June, and coincidentally, 20 months after their last flagship release.
 
I'm pretty sceptical about those benchmarks. But if by any chance they turn out to be true (especially the price part) then a couple of 390x might well replace my 780Ti Classifieds.
 
Considering the huge boost in bandwidth + extra rumored ALU's for the 390X those numbers are on the high but not unreasonable side of what was expected.

The eyebrow raising thing is that they supposedly got that level of performance without increasing power usage over the 290X while on the same process.

I hope the rumored price of $549 is correct. TitanX performance for nearly half the price would be a sure win for AMD in terms of sales IMO. They just have to make sure they have enough supply available so the price gougers don't spoil the launch.

edit: These are new charts. WCCF just re-posted them from Chiphell who just posted them late yesterday.\/\/\/\/\/\/\/
 
Last edited:
Can we stop regurgitating the same charts no one believed almost a week ago?

It's more annoying than how we have duplicate threads right next to each other.

Besides, we need Brent's OC vs. OC review since that's where Maxwell generally destroys AMD cards. :D
 
Those new benchmarks look a lot like the fake ones a few months ago.
The chances that Fiji is faster than Titan X, even at 1600p, are quite small.

And what are the odds that 1 benchmark leaker has access to 4 unreleased cards, 3 of which being unannounced?
As if the 980 Ti even exists yet.

Hang on to that Titan X spec leak, wait a few weeks for the official announcement and we'll see how accurate it is!
 
Last edited:
There have been several fakes in the last few months, which exactly do these resemble?

What information do you have that makes you believe Fiji can't be faster than TitanX? edit: I guess you retracted your claim? Fair enough but editing it out with no mention of why isn't too nice.

Charts are from ChipHell which is in China. Lots of AIB's produce their stuff there and the factory workers and insiders frequently post on ChipHell. They aren't always right but generally if its going to leak anywhere first its going to be on ChipHell.
 
Last edited:
mesyn191 said:
There have been several fakes in the last few months, which exactly do these resemble?

What information do you have that makes you believe Fiji can't be faster than TitanX?

Charts are from ChipHell which is in China. Lots of AIB's produce their stuff there and the factory workers and insiders frequently post on ChipHell. They aren't always right but generally if its going to leak anywhere first its going to be on ChipHell.
Click to expand...
We've had maybe 2 or 3 chiphell leaks on the 390X / "Titan 2" since the end of last year and they were all proven fake. For some reason those benchmarks use the same design/style/color scheme as the old fake ones.

Lots of deja vu coming from that chiphell thread.
It's pretty much guaranteed they're fake.
 
TaintedSquirrel said:
We've had maybe 2 or 3 chiphell leaks on the 390X / "Titan 2" since the end of last year and they were all proven fake. For some reason those benchmarks use the same design/style/color scheme as the old fake ones.

Lots of deja vu coming from that chiphell thread.
It's pretty much guaranteed they're fake.
Click to expand...

I think I'll create some fake graphs and leak those on chiphell and see what happens for the lulz.
 
The fakers are copying the leaks to make them seem more authentic. Happens all the time but not proof these are fakes.

You didn't answer my 2nd question. edit: I guess you retracted your claim? Fair enough but editing it out with no mention of why isn't too nice.
 
Last edited:
mesyn191 said:
The fakers are copying the leaks to make them seem more authentic. Happens all the time but not proof these are fakes.

You didn't answer my 2nd question.
Click to expand...

Just would like to mention Chiphell was the place were the 970 leaks were coming out showing it faster then a 780TI, which alot of people (including myself) kept calling fake and didn't believe it.

970 GTX was spot on what chiphell was leaking.

Have they been wrong in the past? O hell yea....
 
DASHlT said:
Just would like to mention Chiphell was the place were the 970 leaks were coming out showing it faster then a 780TI, which alot of people (including myself) kept calling fake and didn't believe it.

970 GTX was spot on what chiphell was leaking.

Have they been wrong in the past? O hell yea....
Click to expand...
It doesn't really matter, they're 28nm cards so they're going straight down the garbage chute in the first place.
 
mesyn191 said:
Hahahha you gotta be trolling. Gonna toss that 980 or 970 too right?
Click to expand...
By the time the 390X launches in June, the 970 and 980 will be nearly 1 year old...
But the Titan X is also 28nm, so Nvidia is still on the same page as AMD (not like it's their fault, though).

I'm more concerned about the future. If Nvidia is capable of releasing two, nearly three, separate GPU series in the time it takes AMD to release one, it doesn't bode well for what comes beyond GM200 / Fiji.
 
DASHlT said:
Just would like to mention Chiphell was the place were the 970 leaks were coming out showing it faster then a 780TI, which alot of people (including myself) kept calling fake and didn't believe it.

970 GTX was spot on what chiphell was leaking.

Have they been wrong in the past? O hell yea....
Click to expand...

If you throw enough darts at a board, eventually one will hit the bullseye. In fact, I'd put more credibility in an educated guess (e.g. extrapolating possible performance based on specs) from forum users here than some Chinese forum known to post anything and everything they get their hands on.
 
TaintedSquirrel said:
If Nvidia is capable of releasing two, nearly three, separate GPU series in the time it takes AMD to release one, it doesn't bode well for what comes beyond GM200 / Fiji.
Click to expand...
970, 980, and TitanX are different versions of the same GPU uarch though with most of the changes being 'throw more ALU's and bandwidth at that thing'. nV didn't do 3 different clean sheet designs. AMD and nV both release tweaked versions of the same architecture and have done so for years.

AMD used their VLIW uarch for years before switching to GCN which IIRC has a couple of years left in it.

Both AMD and nV have slowed their GPU development since the mid 2000's. The low hanging fruit is long gone and they're both stuck with TSMC.
 
mesyn191 said:
970, 980, and TitanX are different versions of the same GPU uarch though with most of the changes being 'throw more ALU's and bandwidth at that thing'. nV didn't do 3 different clean sheet designs. AMD and nV both release tweaked versions of the same architecture and have done so for years.

AMD used their VLIW uarch for years before switching to GCN which IIRC has a couple of years left in it.

Both AMD and nV have slowed their GPU development since the mid 2000's. The low hanging fruit is long gone and they're both stuck with TSMC.
Click to expand...

I thought AMD, or an article posted that AMD was moving to GloFo for even their GPUs, albeit not immediate maybe but eventually.
 
It's happening!

http://videocardz.com/55124/amd-radeon-r9-390x-wce-could-this-be-real

Next iteration of the highly successful GCN design
– Up to 4096 shader units
Full DirectX®12_Tier 3 implementation
Optimized for 4K gaming & beyond
First ever GPU designed for VR immersion
Special enthusiast water-cooled edition
Up to 8GB of ultra-high bandwidth HBM video memory
Hardware H.265 decoding and over 4x the previous H.264 encoding speed, enabling smooth live streaming
Enhanced ZeroCore™ functionallity (yes there’s a typo)
Click to expand...
 
You must log in or register to reply here.
Back
Top