Fury X now equal to 980 ti @1440p and superior @4k

[Cheddle]

its a bit unfortunate that the fury X is so highly strung... the 980ti gets a good 30% overclock on stock clocks in most instances.

 

Im running my reference card at 1,500 mhz - 24/7. I did throw an AIO watercooler on it but I got those clocks on air - It was just noisy - to be fair the Fury X has a watercooler...  

 

Most users who would spend $1,000 on a GPU would also know enough to overclock it.

[AresKrieger]

Who wants a non blower card for a multi card configuration....

It has watercooled cards too, also anyone that has more than two pcie express lanes as a little space in between is all you need

[Ekin]

Did you report the first post?

 

I reported this thread and linked to the original thread, because there was the same discussion and this post is not following the guidelines.

[Clyne]

It has watercooled cards too, also anyone that has more than two pcie express lanes as a little space in between is all you need

Say that to my friend who is planning to do 3 way GTX 570.

WC cards are usually more expensive than the aftermarket coolers anyway.

Blower cards are great for a situation with little or no air for the GPU to get or a multi card config. 

[Curufinwe_wins]

This is old meta data. Not much of an improvement, and almost all of it was actually the result of techpowerup removing a very anti-amd game from their benchmark suite (wolfenstein iirc).

Note none of this has actually been done separately from other sites so literally it could be 100% the removal of that super anti-amd game from their suite.

[X99stattrak]

This is old meta data. Not much of an improvement, and almost all of it was actually the result of techpowerup removing a very anti-amd game from their benchmark suite (wolfenstein iirc).

Note none of this has actually been done separately from other sites so literally it could be 100% the removal of that super anti-amd game from their suite.

We agree for once !

[Constantine]

Holy shit the 280X is as good as a 780!

 

That card paid for itself well.

[Lotus]

Is this really news? Nobody that owns a Maxwell card is going to leave it at stock clock speeds unless they like throwing money away. What makes Maxwell great, is the insane overclocking headroom it has. Yes people, even the reference coolers that you despise so much, can OC quite a bit. Those 83C temps you see in reviews are on stock fan curves. If you don't mind the sound of jet engines, then crank it to 100 and throw 1350-1400mhz at it for free performance. 

 

That being said, most non-reference cards are only $20 more anyways, and the 980 Ti has been on sale for $630 quite often, so it is still similarly priced to the Fury X. They both trade blows in performance, but its not some recent miracle. Comparing stock 980 Ti to Fury X, yeah, that result was already to be expected.

People keep saying maxwell has insane overclocking headroom, but what they keep forgetting is that most of it is already taken up by GPU boost 2.0. Cards do NOT run at their advertised stock clocks, or even their advertised boost clocks. They'll all run much higher than that. Hell, Luke did a video on stock Titan Xs which were all reference cards and hugely thermally limited and every one of them hit higher clocks than their advertised boost clocks without any changes. All were at "stock clocks" yet all boosted higher than their advertised boost clocks. With reference cards they go even higher.

 

There really isn't that much headroom in Maxwell cards because GPUBoost 2.0 already uses a large portion of it.

[X99stattrak]

People keep saying maxwell has insane overclocking headroom, but what they keep forgetting is that most of it is already taken up by GPU boost 2.0. Cards do NOT run at their advertised stock clocks, or even their advertised boost clocks. They'll all run much higher than that. Hell, Luke did a video on stock Titan Xs which were all reference cards and hugely thermally limited and every one of them hit higher clocks than their advertised boost clocks without any changes. All were at "stock clocks" yet all boosted higher than their advertised boost clocks. With reference cards they go even higher.

There really isn't that much headroom in Maxwell cards because GPUBoost 2.0 already uses a large portion of it.

Boost for my card is supposed to be like 1200 of I remember correctly, I'm pushing almost 1500, and when I get back home I'm going to turn voltage up and see if I can get the magical 1500

[patrickjp93]

I wouldn't call a 4% win against a reference 980 Ti at 4K, and a tie at 1440p a "leap" tbh..good nonetheless, but that is within percent error.

At this point, it's evident which GPU is the better buy and these findings are somewhat hoshposh. GCN and Maxwell are reaching their peak and drivers won't drastically increase performance anytime soon. At the price, the 980 Ti is just better in almost every way.

On reference cards nonetheless >.>

[patrickjp93]

This is what makes me happy about my Lightning purchase... speaking of Lightning when the hell is Linus gonna get one?

He'll wait for the 990TI.

[patrickjp93]

Can Nvidia just unleash the 1060 with 2GB of GDDR5X on a 512-bit bus at 8GHz effective memory clock so we can have some real news and entertainment. For that matter, let's see Nvidia do two versions of GP100, one with 16GB of HBM 2.0 on a 4096-bit bus at 1GHz, the other with a 1024-but bus of GDDR5X (same chip count as a 512-bit GDDR5 bus), just to prove HBM is a lousy failure as a memory standard (latency is more than double GDDR5). I want to see AMD and Hynix cry in a corner, Samsung too for abandoning Micron and Intel at the HMC vote.

[Humbug]

I agree that across the product range AMD seems to be beating NVIDIA performance, even without DX12 / Vulkan.

But not the 980ti. That's the one product where Nvidia still takes the win.

[patrickjp93]

I agree that across the product range AMD seems to be beating NVIDIA performance, even without DX12 / Vulkan.

But not the 980ti. That's the one product where Nvidia still takes the win.

And it's the only one that matters. Whoever has the crown at the top gets all the enthusiasts and major reviewers on their side. After that the midrange sales come flowing in regardless of actual performance at those tiers. It's a profound testament to the stupidity of mankind, just as AMD is a testament to human resilience in the face of adversity.

[Kinda Bottlenecked]

Ah... Shit... Should've gotten the lightning 290x when I had the chance...

[patrickjp93]

Ah... Shit... Should've gotten the lightning 290x when I had the chance...

Get a Lightning 390X.

[Humbug]

Wonder how much the radeon crimson driver update this month will further push up performance for AMD.

Maybe just a small performance bump but mostly focused on feature improvements.

[Kinda Bottlenecked]

Get a Lightning 390X.

Hmm? The last I checked, there wasn't one.

[patrickjp93]

Hmm? The last I checked, there wasn't one.

Give it a few more weeks.

[Albatross]

Aftermarkets? Has the Fury X caught up to them? And what about the ram difference?

[TechGod]

Aftermarkets? Has the Fury X caught up to them? And what about the ram difference?

Why not compare after market Fury Xs to after market Tis to make it fair?

[TechGod]

Wonder how much the radeon crimson driver update this month will further push up performance for AMD.

Maybe just a small performance bump but mostly focused on feature improvements.

Who knows what the actual drivers will bring. But dat software looks sexy. 

[Albatross]

Why not compare after market Fury Xs to after market Tis to make it fair?

 

Do what you want, but I'd still like to know how the gpu compares to aftermarket 980 TIs, and what kind of differences a 4 or 6gb of ram will make.

[Bouzoo]

Can Nvidia just unleash the 1060 with 2GB of GDDR5X on a 512-bit bus at 8GHz effective memory clock so we can have some real news and entertainment. For that matter, let's see Nvidia do two versions of GP100, one with 16GB of HBM 2.0 on a 4096-bit bus at 1GHz, the other with a 1024-but bus of GDDR5X (same chip count as a 512-bit GDDR5 bus), just to prove HBM is a lousy failure as a memory standard (latency is more than double GDDR5). I want to see AMD and Hynix cry in a corner, Samsung too for abandoning Micron and Intel at the HMC vote.

All articles I have found say that HBM has reduced latency when compared to GDDR5 which makes sense considering the chips are on the die, so once again, I'm pretty curious as to where did you get the info it has double the latency. 

[patrickjp93]

All articles I have found say that HBM has reduced latency when compared to GDDR5 which makes sense considering the chips are on the die, so once again, I'm pretty curious as to where did you get the info it has double the latency.

OpenCL testing on a friend's Fury.

1)Set up CLTimer,

2)write time to a public datem,

3)Access 32 sequential datems per stack and add 1 (4096/32 = 128 = 4 * 32)

4)stop timer,

5)array entry becomes the difference between timer and time,

6)time = timer,

7)rest all the GPU threads for ~300 microseconds (Fury cache toss out time),

8)restart timer,

9)if 70,000 entries recorded, send times to out buffer and quit. Else, go to 2)

10) Repeat process with a GDDR5 card (I used my dad's 970, and accessed 7 datems per burst (not 8 since 3.5GB )

11) Throw both data sets into Excel/Calc

12) divide HBM row entries by GDDR5 row counterparts, put in separate row

13) Take average of that row

14) My Result: ~1.9, or HBM has 1.9x the latency of GDDR5.