AMD Fury Series disappointment

[xthetenth]

I keep hearing how the 980Ti's hit 1500Mhz no problem yet when you look at the feedback of the aftermarket cards on places like NewEgg it paints a different picture. Many complaining about getting crappy ASIC ratings with terrible overclocking etc.

I'm guessing the truth lies somewhere in the middle. If the average overclock is around 1350Mhz how much of an improvement would that be over stock cards? 10%?

Sampling bias is in effect, but I'm not surprised that maxwell overclocking is slightly exaggerated.

 

[Madpacket]

Sampling bias is in effect, but I'm not surprised that maxwell overclocking is slightly exaggerated.

That's what I suspect as well. I've had two different 970's with dramatically different overclocking / boost clocks. One card hit 1475 with added voltage and clock speed boost while the other barely hits 1320..

I didn't see anywhere close to a 20% improvement even on the first card over stock boost clocks so I find it hard to swallow the average aftermarket 980Ti is getting near that. I know different cards and all but same Maxwell technology.

 

[guskline]

If you look here you can see the various overclocks they got with the different cards. Obviously it varies from sample to sample, but in general 1400MHz is a given with aftermarket cards. Heck, a card like the MSI Lightning boosts to 1400MHz out of the box I believe.

Performance gains over reference clocks are much more than 10%. More like 25%+ for a good overclock. Probably closer to 15-20% for a modest overclock.

Agreed. One of the best overclocking cards I've owned in a long time.:thumbsup:

 

[prtskg]

I wonder the reasons of Fury Series's failure. The card has very high theoretical bandwith and crazy compute power. (But it has only 64 ROPs.) It had to be have more real life performance but it has not.

I wouldn't say fury is a failure. 980ti is just awesome.

A new memory tech not having an efficient implementation is understandable. That would give low score in the test you showed.

This time AMD created a gpu as big as Nvidia's. Earlier they depended on xfire to match Nvidia's best. So inexperience in creating big gpus can also be reason for inefficiency in fiji.

 

[n0x1ous]

I keep hearing how the 980Ti's hit 1500Mhz no problem yet when you look at the feedback of the aftermarket cards on places like NewEgg it paints a different picture. Many complaining about getting crappy ASIC ratings with terrible overclocking etc.

I'm guessing the truth lies somewhere in the middle. If the average overclock is around 1350Mhz how much of an improvement would that be over stock cards? 10%?

I have 2 reference cards.

1 ASIC 68% can do 1450mhz
1 ASiC 87% can do 1525mhz

 

[ultron]

I wouldn't say fury is a failure. 980ti is just awesome.

A new memory tech not having an efficient implementation is understandable. That would give low score in the test you showed.

This time AMD created a gpu as big as Nvidia's. Earlier they depended on xfire to match Nvidia's best. So inexperience in creating big gpus can also be reason for inefficiency in fiji.

I wonder the HBM's inefficiency reason. Memory access patterns, driver issues or another thing?

Fiji is too big, and has too much compute units. Maybe lesser CU count could make the GPU more efficient.

 

[guskline]

Fury is probably the platform AMD is building on to stay competitive or jump ahead of Nvidia.

Let's see where Polaris takes us.

 

[itsmydamnation]

it could also just be NV and AMD took different approaches to the cancellation of 20nm, eg. nvidia backported more stuff to 28nm where AMD has done almost nothing outside of the ROP/memory/display/codec since tahiti.

Polaris will answer the questions one way or the other and then there will be no doubt.

 

[tviceman]

These are boost clocks, of course. Not stable base clocks.

My card boosts to 1510 and stablizes at ~1490mhz. Your point is taken, but it's invalid for those who have adequate cooling. The difference between 1510 and 1490 is not measurable for a benchmark run or gameplay.

 

[gamervivek]

Fiji scales well enough over Tonga, 87% at 4k is not too shabby over the 380X strix.

AMD have been focusing on Hawaii GCN and it especially shows in dx12 situations where Fury X doesn't scale well over Hawaii and doesn't do well with mantle.

Hawaii's tessellation performance received a big boost during the 390 series release, so it's not inconceivable that Fury X extends its lead over 980Ti by the time Polaris releases.

 

[parvadomus]

Fiji scales well enough over Tonga, 87% at 4k is not too shabby over the 380X strix.

I dont think so, it has exactly twice of everything, and it even has more bandwidth per shader. 87% is mediocre at best.
The problem with fiji is that AMD killed its performance to get better power consumption, or the CPU becomes a bottleneck cause of drivers, or simply the frontend cannot feed all shaders.

 

[railven]

I have 2 reference cards.

1 ASIC 68% can do 1450mhz
1 ASiC 87% can do 1525mhz

I believe my ASIC was low like 62% or 67%. With stock cooler @ 100% I hit 1505mhz on it's maiden voyage. TDR at anything higher.

With EVGA Hybrid kit, does 1514mhz. TDR at anything higher.

In both situations I think i was power limited by the stock 106% option. But I don't see a need for custom BIOS, handles everything fine.

 

[Glo.]

Fiji is a mish-mash of current technologies and architectures from AMD.

Hardware Schedulers - Fiji/Polaris
Asynchronous Compute, and cache size - Hawaii
Color Compression - Tonga
HBM - Fiji and Polaris.

The effect is... terrible. Because it is a wish-mash it performs worse than it should. It is bottlenecked everywhere, especially - on cache size. Especially it is important if you are dealing with 512GB/s. Otherwise it will never will be fully utilized.

If it would be architecture from ground up designed to work in specific environment - it would be much better, and much more efficient.

Even so, Nvidia was frightened about Fiji, and next gen architectures, so they prepared GTX 980 Ti, which overall is a better card.

 

[railven]

Even so, Nvidia was frightened about Fiji, and next gen architectures, so they prepared GTX 980 Ti, which overall is a better card.

Frightened? If one thing I've noticed about Nvidia, is they are arrogant. Extremely arrogant. Look at Fermi. They didn't even flinched they believed their own marketing so much.

$650 for reject $1,000+ chips? Mean while AMD is giving you a water cooler, full chip, and break-thru memory and still loses in the majority of benches?

Nvidia either didn't expect 980 Ti to beat Fiji or knew exactly what Fiji was and tore a whole into it's sail.

 

[2is]

Frightened? If one thing I've noticed about Nvidia, is they are arrogant. Extremely arrogant. Look at Fermi. They didn't even flinched they believed their own marketing so much.

$650 for reject $1,000+ chips? Mean while AMD is giving you a water cooler, full chip, and break-thru memory and still loses in the majority of benches?

Nvidia either didn't expect 980 Ti to beat Fiji or knew exactly what Fiji was and tore a whole into it's sail.

I'd be arrogant too if my defects are outperforming the competitions flagship.

 

[railven]

I'd be arrogant too if my defects are outperforming the competitions flagship.

Hopefully AMD learned from it's execution mistakes and doesn't let Nvidia walk all over them.

But the recent AMD youtube video crying foul at Intel doesn't inspire much confidence.

Oh well. I'm ready to upgrade, so someone is going to get my money.

 

[Techhog]

I'd be arrogant too if my defects are outperforming the competitions flagship.

Those "defects" are outperforming Nvidia's own flagship as well...

 

[MrTeal]

Those "defects" are outperforming Nvidia's own flagship as well...

To be fair, that's entirely by choice.

 

[2is]

Those "defects" are outperforming Nvidia's own flagship as well...

Not really. They trade blows in most games at resolutions up to 4k, but once you go multi monitors with really heavy vram demands Titan walks away. Not to mention what was said above. nVidia is simply not allowing Titan to go past a certain TDP.

 

[boozzer]

you guys really should be thankful to AMD, without fury x, you guys would have been stuck with titan x at 1000$ instead of at 650$.

 

[96Firebird]

you guys really should be thankful to AMD, without fury x, you guys would have been stuck with titan x at 1000$ instead of at 650$.

Nope.

 

[railven]

you guys really should be thankful to AMD, without fury x, you guys would have been stuck with titan x at 1000$ instead of at 650$.

You're 100% right. Thanks AMD. :thumbsup:

EDIT:

Nope.

Not sure how AMD losing a once loyal customer helps them, but oh well.

 

[Arachnotronic]

you guys really should be thankful to AMD, without fury x, you guys would have been stuck with titan x at 1000$ instead of at 650$.

NVIDIA released the 980 Ti first, so it's really more AMD fans should thank NVIDIA for spoiling AMD's plans to release Fury X at higher prices

 

[moonbogg]

NVIDIA released the 980 Ti first, so it's really more AMD fans should thank NVIDIA for spoiling AMD's plans to release Fury X at higher prices

Interesting twist. I see what you did RIGHT THERE. My cards still make me giddy. I'd install them inside my body if we had the technology. I want them inside me.

 

[sm625]

If Nvidia released a card with the exact same die, HBM, and everything, in place of its existing 980ti, it would sell 5x as many units and no one would call it a failure. Even if it was AMD who released the 980ti, Nvidia would still sell more.