WCCF: AMD’s Flagship ‘Fiji XT GPU’ Debuts Radeon’s Titan Equivalent Branding

[maddie]

The specs on that look to have 64 ROP. Isn't that unlikely? max clock 1000mhz also seems low.

Also, can someone show me anywhere that says this is 'official'?

As RS pointed out some time ago. Tonga tech rops have more than twice the fill rate of Hawaii rops. Effectively this could mean 64 rops here is double+ previous 64 in Hawaii.

Max clocks could be low if the shader count is massive [I speculated a while ago more than 4096]

Compubench is a professional site with official partnerships to most big technology companies [check the about us: partners page https://compubench.com/partners.jsp]

This not a normal review/test site.

 

[Kenmitch]

That is one gorgeous card.

Nice looking but the photos do it no justice.

The quick YouTube tease is much better
.
Probably already linked. If not https://youtube.com/watch?v=Ra4V_SUDrGw&feature=em-subs_digest

Very quick tease but looks impressive.

 

[LTC8K6]

Is this Fiji?

https://compubench.com/result.jsp?benchmark=compu20&test=572


https://compubench.com/device.jsp?b...=AMD+Radeon+Graphics+Processor&testgroup=info

What happened to the first link?

 

[maddie]

The new question is which Fiji model is this.

Is this Fiji max or one of the harvested die?

Don't misunderstand me here, but a 30% increase while great is not that great when seen in the light of a massive die increase + HBM memory and a better density/lower power new foundry process.

 

[exar333]

As RS pointed out some time ago. Tonga tech rops have more than twice the fill rate of Hawaii rops. Effectively this could mean 64 rops here is double+ previous 64 in Hawaii.

Max clocks could be low if the shader count is massive [I speculated a while ago more than 4096]

Compubench is a professional site with official partnerships to most big technology companies [check the about us: partners page https://compubench.com/partners.jsp]

This not a normal review/test site.

Significant Tonga gains still remain to be seen (sort of a mixed-bag). The low clocks are somewhat disappointing, if true.

Your post reads wort of like an advertisement for the site? Am I reading that right?

Hoping the stock performance is killer, but tough to gauge with these types of benchmarks. Need some real world-world performance.

 

[exar333]

The new question is which Fiji model is this.

Is this Fiji max or one of the harvested die?

Don't misunderstand me here, but a 30% increase while great is not that great when seen in the light of a massive die increase + HBM memory and a better density/lower power new foundry process.

This is the real wild-card to me, along with HBM. Will the new process lower stock tdp or prove to really allow for some extreme frequencies? I don't think we know the true gains from HBM yet either.

Regardless of what we see at launch, I am interested in what sort of driver tweaks might be in-store for Fiji down the road, as HBM is more and more exploited. If max OC performance is on-par with 980Ti/TX, I am more inclined to try Fiji knowing it likely has some extra performance to be squeezed out later. Maxwell is a lot more mature and I don't expect to see a y huge surprises there in 2015/2016...

 

[maddie]

Significant Tonga gains still remain to be seen (sort of a mixed-bag). The low clocks are somewhat disappointing, if true.

Your post reads wort of like an advertisement for the site? Am I reading that right?

Hoping the stock performance is killer, but tough to gauge with these types of benchmarks. Need some real world-world performance.

I wish, then at least they might pay me.

Was trying to show that the results are likely legit.

 

[LTC8K6]

The results appear to have been removed and re-posted.

Why is a 280X much faster than a 290X?

 

[exar333]

I wish, then at least they might pay me.

Was trying to show that the results are likely legit.

LOL, point taken.

I didn't question the site, just a challenge sifting through all the rumors these days...

 

[maddie]

The results appear to have been removed and re-posted.

Why is a 280X much faster than a 290X?

Double precision performance of Hawaii is 1/8 of single precision performance Double precision performance of Tahiti is 1/4 of single precision performance

 

[Enigmoid]

Double precision performance of Hawaii is 1/8 of single precision performance Double precision performance of Tahiti is 1/4 of single precision performance

Has nothing to do with it as the titan x dominates tests.

 

[maddie]

Has nothing to do with it as the titan x dominates tests.

I assumed he was talking about the 'Ocean Surface Simulation' test which had Fiji far ahead of all other cards.

In that test the 280X was faster than the 290X. DP compute is the reason.

This is not a battle for supremacy here, just trying to understand/explain results.

 

[maddie]

The results appear to have been removed and re-posted.

Why is a 280X much faster than a 290X?

Noticed that also. Refreshed a page and Fiji was gone replaced by 'AMD Radeon Graphics Processor', then it changed back to Fiji.

They probably tried to and then realized it was too late to stop the data leak.

 

[LTC8K6]

It shows a benchmark for R9-270 (X4) in crossfire I presume.

But R9-270 can only do 2X in crossfire?

 

[LTC8K6]

I assumed he was talking about the 'Ocean Surface Simulation' test which had Fiji far ahead of all other cards.

In that test the 280X was faster than the 290X. DP compute is the reason.

This is not a battle for supremacy here, just trying to understand/explain results.

It had the AMD card about 20% faster than Titan X.

Overall, the benches seem to be all over the place, though.

 

[Enigmoid]

I assumed he was talking about the 'Ocean Surface Simulation' test which had Fiji far ahead of all other cards.

In that test the 280X was faster than the 290X. DP compute is the reason.

This is not a battle for supremacy here, just trying to understand/explain results.

Nothing to do with DP as the titan x with 200 GFLOPS does very well in comparison to other chips with far less DP.

 

[maddie]

Nothing to do with DP as the titan x with 200 GFLOPS does very well in comparison to other chips with far less DP.

I'm willing to learn something.

Why is the 280X faster than the 290X in the 'Ocean Surface Simulation' test?

 

[LTC8K6]

I'm willing to learn something.

Why is the 280X faster than the 290X in the 'Ocean Surface Simulation' test?

There's a 280X that's slower than the 290X cards in the same test. It's on the next page.

 

[maddie]

There's a 280X that's slower than the 290X cards in the same test. It's on the next page.

Strange results indeed, to me at least. I'm sure an expert can understand it.

 

[sm625]

There's a 280X that's slower than the 290X cards in the same test. It's on the next page.

That 2nd one is a prototype and it is using older drivers.

 

[monstercameron]

It shows a benchmark for R9-270 (X4) in crossfire I presume.

But R9-270 can only do 2X in crossfire?

For compute, I don't think there is a limitation to how many cards you can have installed, but there are limitations for latency sensitive applications like graphics.