GFXBench Validation Confirms Stream Processor Count of Radeon Fury

[btarunr]

Someone with access to an AMD Radeon Fury sample put it through the compute performance test of GFXBench, and submitted its score to the suite's online score database. Running on pre-launch drivers, the sample is read as simply "AMD Radeon Graphics Processor." Since a GPGPU app is aware of how many compute units (CUs) a GPU has (so it could schedule its parallel processing workloads accordingly), GFXBench was able to put out a plausible-sounding CU count of 64. Since Radeon Fury is based on Graphics CoreNext, and since each CU holds 64 stream processors, the stream processor count on the chip works out to be 4,096.



View at TechPowerUp Main Site

 

[KarymidoN]

Wow

 

[chinmi]

too bad there's no love for amd in this world. even most tpu member is nvidia user.
the world will be a better place without amd. we'll gonna be just fine with only nvidia and intel.

 

[btarunr]

too bad there's no love for amd in this world. even most tpu member is nvidia user.
the world will be a better place without amd. we'll gonna be just fine with only nvidia and intel.

The persecution-complex is high on you.

 

[Naito]

too bad there's no love for amd in this world. even most tpu member is nvidia user.
the world will be a better place without amd. we'll gonna be just fine with only nvidia and intel.

What a strange comment. I assure you there's probably just as many AMD diehards around here as there is for Nvidia!

 

[shhnedo]

the world will be a better place without amd. we'll gonna be just fine with only nvidia and intel.

I really hope you're joking. No competition is actually bad for the world.

 

[ZoneDymo]

too bad there's no love for amd in this world. even most tpu member is nvidia user.
the world will be a better place without amd. we'll gonna be just fine with only nvidia and intel.

I take it you did not have any breakfast?

On the article, could Particle simulation be so much better on Nvidia because of PhysX? or does that specifically require the particle simulator to be using that?

 

[64K]

too bad there's no love for amd in this world. even most tpu member is nvidia user.
the world will be a better place without amd. we'll gonna be just fine with only nvidia and intel.

I think you must be kidding. You've been a member on this site for 3 1/2 years. You surely must have seen that there are quite a few people that passionately support AMD here.

 

[RejZoR]

I have plenty of love for AMD, especially GPU's. Not because I'm a fanboy, I had a large share of NVIDIA GPU's as well, but because they simply make good graphic cards, they work great for me, drivers were also good most of the time, can't really complain. And R9-300 series look interesting again. Now it's just a matter of pricing...

 

[midnightoil]

On the article, could Particle simulation be so much better on Nvidia because of PhysX? or does that specifically require the particle simulator to be using that?

Some of the GFXBench tests benefit enormously from driver optimisation (as has been confirmed a few times by Kishonti), other tests don't. NVIDIA have done quite a bit of work on drivers for GFXBench in the past ... AMD supposedly little to nothing.

That probably explains why Fiji annihalates the competition in fluid simulation and has a higher absolute fill rate, but NVIDIA dominate the particle test.

 

[Petey Plane]

too bad there's no love for amd in this world. even most tpu member is nvidia user.
the world will be a better place without amd. we'll gonna be just fine with only nvidia and intel.

I cant tell if this post is pro or anti AMD

 

[GhostRyder]

too bad there's no love for amd in this world. even most tpu member is nvidia user.
the world will be a better place without amd. we'll gonna be just fine with only nvidia and intel.

Wow dude, you have no concept of what you are saying do you (Or your intentionally trolling)? So only Nvidia and Intel, guess next round we can expect double the cost for every card at each point with very small increases in performance. I am going with "intentional troll" on this comment...

4096 I think was mostly known but not fully confirmed but with this it just seals the deal. Hopefully that is a taste of what we can expect!

 

[$ReaPeR$]

i really dont consider these useful benchmarks. waiting for real world app benchmarks.

also.. why do you bother answering to nonsensical/moronic posts? do you feel the need to judge moronic opinions? (sorry for being off topic)

 

[Initialised]

I don't like how the 200 series is being re branded with a halo product at the top of the stack.

The rebrand seems especially bad seeing as the 7870 (Tahiti) already became 270X and is about to become the 370.

Nvidia did the same with the 8800 GTS 512MB/9800/GTS 250

With the Fury being a halo product akin to Titans I fear the price will be ridiculous so they wont sell and the Halo effect wont work.

 

[the54thvoid]

Some of the GFXBench tests benefit enormously from driver optimisation (as has been confirmed a few times by Kishonti), other tests don't. NVIDIA have done quite a bit of work on drivers for GFXBench in the past ... AMD supposedly little to nothing.

That probably explains why Fiji annihalates the competition in fluid simulation and has a higher absolute fill rate, but NVIDIA dominate the particle test.

That's one interpretation....

The other is that it's stronger and weaker - dependent on situation. And who gets their hands on a unreleased GPU without access to some form of software? The drivers already exist - they'll simply be getting tweaked like the nipples on a cheap hooker (male or female).

I'm logically waiting for the real thing. Either way, I win. Given that the GTX980ti is a hell of an awesome card, especially looking forward to board partner designs - IF Fury is faster (clock for achievable clock) then AMD have made the best card in GPU's in a long time. So, time will tell but I'll have me one or other.

 

[st2000]

fury is more powerful than 295x2 while only 300W - almost enough

 

[crazyeyesreaper]

I don't like how the 200 series is being re branded with a halo product at the top of the stack.

The rebrand seems especially bad seeing as the 7870 (Tahiti) already became 270X and is about to become the 370.

Nvidia did the same with the 8800 GTS 512MB/9800/GTS 250

With the Fury being a halo product akin to Titans I fear the price will be ridiculous so they wont sell and the Halo effect wont work.

They dont really have a choice. They are still stuck at 28nm which has been the mainstay for over 3 years. The process node is stagnant because the jump to 20nm never happened. Thus Even Nvidia can do nothing truly new untill 2016 probably around xmass. It is what it is. With a drop to 16nm AMD can redesign all cards to use the same GCN version and move forward but untill then we are stuck with rebrands because it does not make sense to waste R&D cash on a new GPU that performs exactly the same as a previous gen product.

 

[Tatty_Two]

I particularly like that most of us are NVidia users, 7 of us that have posted so far in this thread are AMD users..... don't feed the Trolls people! Ooops I just did

 

[bogmali]

Don't feed the Trolls people! Ooops I just did

But it was a good troll effort though....ooops I fed it too....damn it Tatty

 

[Frick]

I only use Matrox products myself.

 

[Ferrum Master]

I only use Matrox products myself.

But matrox now also uses AMD chips

 

[bogami]

Results are good. Driver not yet accomplished, and I feel that there is great potential.(30%-50%).It was always so with AMD GPU on start.
Competition must be because we can see what happens then when it did not exist , the producers prices raise regardless and beyond any justification .

 

[Frick]

But matrox now also uses AMD chips

Aww sellouts!

 

[FrustratedGarrett]

Looks like a TitanX beater. I predict that the FuryX card will beat the TitanX by a bit (10% - 15%) which is a step forward for AMD, considering it's a smaller chip, according the rumors. This kinda tells how modular the GCN micro-architecture is, compared to Maxwell's. What's even more intriguing is the fact that Next year, AMD's Zen APUs will probably have HBM and quite powerful IGP sitting right beside the CPU with full access to memory and cache, and able to swap back and forth thread with the CPU cores without going through a context switch (jump to kernel code and waste CPU cycles etc.).

 

[BiggieShady]

On the article, could Particle simulation be so much better on Nvidia because of PhysX? or does that specifically require the particle simulator to be using that?

It has nothing to do with PhysX, this is because how different GCN and Maxwell GPU architectures are ... GCN has fantastic theoretical throughput and if the type of a problem/algorithm is suitable it really shines. There are some instances where it's less so (as you noticed with particle simulation) because scheduler can't reshuffle the instructions to keep all vector units inside GCN cores fully busy all the time.
Nvidia has more versatile architecture IMO although with less theoretical throughput. They did fantastic job refining it so that instruction scheduling is much simpler and they improved cache subsystem by introducing another level of caching.