Possible Radeon 390X / 390 and 380X Spec / Benchmark (do not hotlink images!!!!!!)

[PayneUK]

I have little doubt HBM will be a big thing, the issue is will the 390x have enough raw ‘grunt’ to actually make use of it?

Probably not, and that’s where people need to manage their expectations, it’s possible in cross/tri-fire we will really see an improvement at big resolutions but single card? Don’t expect a titan killer quite yet.

 

[nashathedog]

Time for me to join the gaming enthusiasts club.

I've been going back and forth deciding how much to spend on a GPU and which one to get. But after some convincing I'm increasing my GPU budget to around £500. So now this card could be a possible option.

You reckon it'll be much better than a 295x2? Because that's £500 and if this 390x is around £500 then it would be between those two cards as my Asus P8P67 only supports crossfire.

Come on AMD, give some news..

Your m'b may support crossfire but the thing is AMD barely support crossfire so you'll be better of with a 390/390x regardless.

 

[nutsnut]

Time for me to join the gaming enthusiasts club.

I've been going back and forth deciding how much to spend on a GPU and which one to get. But after some convincing I'm increasing my GPU budget to around £500. So now this card could be a possible option.

You reckon it'll be much better than a 295x2? Because that's £500 and if this 390x is around £500 then it would be between those two cards as my Asus P8P67 only supports crossfire.

Come on AMD, give some news..

I've got an Asus P8P67 pro and have run Crossfire and SLI, what MB is yours exactly?

 

[humbug]

I have little doubt HBM will be a big thing, the issue is will the 390x have enough raw ‘grunt’ to actually make use of it?

Probably not, and that’s where people need to manage their expectations, it’s possible in cross/tri-fire we will really see an improvement at big resolutions but single card? Don’t expect a titan killer quite yet.

No one serious is talking about a "Titan-X killer"

AMD said themselves they are bottlenecked by GDDR5, of course that could just be bluster...

Yet if its not going to make any difference then why spend a mountain of cash to develop it?

Essentially there are two sides to this argument, 'Cite rumours' AMD would get about 60 - 70% out of those hardware improvements.

There isn't much needed to get it to Titan-X level, actually nothing at all, with that its already as fast.

So these are the choices:
HBM will add a bit.
HBM will add nothing at all.

Or another option: A bit more..

 

[D.P.]

At best case scaling the 295x2 is 190% the performance of a 290X.

If rumours are true:

First add at least +10% for Texture Compression and Tiling 'Tonga'

(= real performance)

Hawaii: - Tonga +10% (= 110%)
Hawaii: 64 ROP's - Fiji: 128 ROP's + 100% (+20%) (= 130%)
Hawaii: 2816 Shaders - Fiji: 4096 Shaders +45% (+35%) (= 165%)
Hawaii: 1Ghz - Fiji: 1.05Ghz +5% (+3%) (=168%)

HBM
Memory Bandwidth
Hawaii: 320 GB/s - Fiji: 1024 GB/s +230% (=???%) (needs +20%)

Of course its all just my opinion but i do think i'm being a little bit cautious on the performance gains the Hardware gives without HBM.

I think its entirely possible.

1+1 =5?


What an utterly ridiculous methodology - performance is not a linear sum of parts.

Things like increased memory bandiwdth are only useless if you are bandwidth limited, and even if you are and alleviate that all what happens is you hit the next bottle neck.

Much more effective would be look at the number of transistors, die size and make an estimate of efficiency increases. As much as you love AMD they can't break the laws of physics.

 

[Orangey]

What an utterly ridiculous methodology - performance is not a linear sum of parts.

Wasn't it you saying you could predict Titan X by the cores + clocks?

 

[D.P.]

Wasn't it you saying you could predict Titan X by the cores + clocks?

On an identical architecture as MAXIMUM theoretical limit, not an in game average.

That was in response to people saying they were disappointed with the TX performance because they thought it would be twice as fast as the 980. I merely pointed out the MAX theoretical difference would be 40% over the 980 due to cores + clock, with real world difference less than that at lower settings. Which is exactly what we see.




290 is a different unknown architecture and things like 230% increase in bandwidth are meaningless

 

[Mauller]

People forgetting that HBM has far lower latency than GDDR5, it also has Parallel read and write to the same Die. Both of these improve performance alone, regardless of the bandwidth increase.

HBM is a completely different architecture to GDDR5, you cant directly compare them by bandwidth alone. unlike GDDR4 - 5 etc.

 

[Kaapstad]

Wasn't it you saying you could predict Titan X by the cores + clocks?

You can.

 

[Kaapstad]

People forgetting that HBM has far lower latency than GDDR5, it also has Parallel read and write to the same Die. Both of these improve performance alone, regardless of the bandwidth increase.

HBM is a completely different architecture to GDDR5, you cant directly compare them by bandwidth alone. unlike GDDR4 - 5 etc.

If a GDDR5 setup is not bottlenecked switching to HBM will give zero performance improvement. Graphics cards get their performance from the core not the memory.

 

[Orangey]

I'm wondering what the wattage saved will be used for though.

 

[Mauller]

If a GDDR5 setup is not bottlenecked switching to HBM will give zero performance improvement. Graphics cards get their performance from the core not the memory.

Processors are always bottlenecked by memory, especially memory that is off die, there are two things that determine the bottleneck but they are both independent of eachother; Latency and Bandwidth.

It is the reason why CPU's have a large amount of die dedicated to prediction and caching.

When a processor is bottlenecked by latency, many of its clock cycles will go to waste, regardless of the bandwidth there is between the Processor and the memory cells, its a matter of speed and capacity. A processor can't perform work, when it has no data to crunch, even if it can be sent gigs of data at a time.

so with lower latency memory, the processor can do more work in the same amount of time, etc.

 

[Kaapstad]

I'm wondering what the wattage saved will be used for though.

I don't think it does save as much wattage as has been said.

A 12gb Titan X does not use massive amounts of power compared to other cards. A 4gb 290X has a lower transistor count on the core and a lot less memory yet uses more power than the Titan X.

 

[Kaapstad]

Processors are always bottlenecked by memory, especially memory that is off die, there are two things that determine the bottleneck but they are both independent of eachother; Latency and Bandwidth.

It is the reason why CPU's have a large amount of die dedicated to prediction and caching.

When a processor is bottlenecked by latency, many of its clock cycles will go to waste, regardless of the bandwidth there is between the Processor and the memory cells, its a matter of speed and capacity. A processor can't perform work, when it has no data to crunch, even if it can be sent gigs of data.

so with lower latency memory, the processor can do more work in the same amount of time, etc.

Total rubbish

 

[Ayahuasca]

I think HBM will have a nice advantage at the higher resolutions but for those expecting miracles at 1440p and 1080p I wouldn't build your hopes up.

By the time it's really needed for the majority of users both sides will be using new memory technology anyway.

 

[Mauller]

Total rubbish

It is not rubbish, it is a basic principle with computer architecture.

a processor may not be bandwidth starved, but it can still be starved due to latency.

its a matter of transferring something to the processor quicker, compared to giving it more with each transfer.

 

[Kaapstad]

It is not rubbish, it is a basic principle with computer architecture.

a processor may not be bandwidth starved, but it can still be starved due to latency.

its a matter of transferring something to the processor quicker, compared to giving it more with each transfer.

My system RAM runs at 3200mhz

If I bench most games I get the same performance if I down clock the memory to 3000mhz, 2666mhz, 2400mhz, 2133mhz, 1866mhz or 1600mhz.

Just because something can go faster does not mean that it is needed.

 

[Mauller]

My system RAM runs at 3200mhz

If I bench most games I get the same performance if I down clock the memory to 3000mhz, 2666mhz, 2400mhz, 2133mhz, 1866mhz or 1600mhz.

Just because something can go faster does not mean that it is needed.

you have just stumbled onto one of the most misunderstood things with Ram Frequency and timings, even though the frequency is increasing, the timing increases with the frequency as the timings are a measure of how many clock cycles the memory controller has to wait before it receives a signal, the overall latency from the memory cell to the memory controller barely changes as the frequency changes, just the overall bandwidth increases. But between different DDR generations, latency has slowly decreased.

when it comes to the performance, 1600mhz DDR3 has plenty of bandwidth for games.

this does give a good explanation, if you would like to read about it.
http://www.chipestimate.com/techtalk.php?d=2011-11-22

 

[africanos23]

If a GDDR5 setup is not bottlenecked switching to HBM will give zero performance improvement. Graphics cards get their performance from the core not the memory.

+1

I think GDDR5 up to 8K will not have memory bandwidth issues with a big enough bit-bus (384+). HBM might be better down the line @ ~16K or more.

 

[Squink]

+1

I think GDDR5 up to 8K will not have memory bandwidth issues with a big enough bit-bus (384+). HBM might be better down the line @ ~16K or more.

I am parroting DM, but I thought one of the chief benefits of HBM at this stage is the much lower power consumption, which allows AMD to squeeze more out of the core without heat issues.