Surprise! AMD shares internal Fury X benchmarks ahead of review embargo

[Shawn Knight]

Today marked the launch of AMD’s new 300 Series graphics cards but truth be told, most everyone is waiting to see what kind of performance the upcoming Radeon Fury X can lay down. Some reviewers already have AMD’s latest hardware in their test beds but as you may have heard, all reviews are embargoed until next week (June 24, to be exact).

As happens from time to time, there was a communications “misunderstanding” and AMD’s internal benchmarks – the ones included in the reviewer’s guide as guidelines of what to expect during testing – were prematurely posted online. With the cards (no pun intended) already on the table table, AMD has given reviewers the green light to go ahead and share these internal benchmarks with readers.

And here are the settings that AMD used in each test.

Here's a little snipped of what you can expect with a 100MHz overclock.

It’s clear to see that the Radeon Fury X outpaces Nvidia’s GTX 980 Ti in every single test. But it’s also important to remember that these benchmarks do come with an asterisk attached.

AMD obviously wants to showcase the Fury X in the best possible light. That doesn’t mean there’s anything shady going on here but it’s pretty much a guarantee that what we’re seeing is a best-case scenario. How will these results compare to true real-world performance? Unfortunately, we'll have to wait until June 24 to get those answers.

In the meantime, here's a complete specifications list of the Fury X to chew on.

Permalink to story.

https://www.techspot.com/news/61042-surprise-amd-shares-internal-fury-x-benchmarks-ahead.html

 

[VitalyT]

They are 4GB of RAM short of the promise they made, but they are 10 miles ahead in their Red vs Green B.S.

I like green colours much better though

 

[madboyv1]

Unfortunately, we'll have to wait until June 24 to get those answers.

I'll be waiting.

 

[yRaz]

With nvidias next line up coming next year I don't really know what AMD hopes to accomplish. These cards wont be out until the fall and by then nVidias new chips will be right around the corner, which I'm sure will be much faster.

Maybe they need some time to work out the drivers to get more performance, but I'm honestly not that impressed by AMD's highest end single GPU "next gen" graphics card.

I guess the price point will be the real kicker

 

[Puiu]

With nvidias next line up coming next year I don't really know what AMD hopes to accomplish. These cards wont be out until the fall and by then nVidias new chips will be right around the corner, which I'm sure will be much faster.

Maybe they need some time to work out the drivers to get more performance, but I'm honestly not that impressed by AMD's highest end single GPU "next gen" graphics card.

I guess the price point will be the real kicker

Fury X will be out this month June 24th for $649 (the same price as the 980 TI). You have this launch date written 2 times in the article and you'll also see it in the comments.
Fury will be out soon after in July for $549.
Fury Nano will be out later this summer (most likely august)
Fury dual GPU is the only card that will be released this fall.

 

[Evernessince]

With nvidias next line up coming next year I don't really know what AMD hopes to accomplish. These cards wont be out until the fall and by then nVidias new chips will be right around the corner, which I'm sure will be much faster.

Maybe they need some time to work out the drivers to get more performance, but I'm honestly not that impressed by AMD's highest end single GPU "next gen" graphics card.

I guess the price point will be the real kicker

These chips come out in a few days. With this launch, AMD is the defacto ultra high end king and high res/multi monitor king. Even the R9 200 series had a better memory bus than Nvidia's, this pushes that even further.

 

[yRaz]

Fury X will be out this month June 24th for $649 (the same price as the 980 TI). You have this launch date written 2 times in the article and you'll also see it in the comments.
Fury will be out soon after in July for $549.
Fury Nano will be out later this summer (most likely august)
Fury dual GPU is the only card that will be released this fall.

My mistake, I'm tired and I read it wrong. I thought review samples were sent out and sites weren't allowed to post benchmarks but I just thought it was going to be just their 300 series with the fury coming out in the fall.

That said, I'm still not impressed with it's performance. With Pascal coming out next year I really don't think AMD's cards are very competitive in the long haul, especially since this is considered next gen. When Pascal is released these cards wont be able to compete unless there is a very steep price drop.

 

[Guest]

Serious, "this is not impressive", first HBM card to ever be released..... fairly amazing card really and the Fury Nano will be hectic.

Saying the performance on these cards is not impressive because Nvidia will be faster when they release Pascal when it is 12 months away... seems a bit Nvidia fanboyish. AMD will have their second gen HBM out shortly after that which will be faster again.

 

[darkzelda]

Fury X will be out this month June 24th for $649 (the same price as the 980 TI). You have this launch date written 2 times in the article and you'll also see it in the comments.
Fury will be out soon after in July for $549.
Fury Nano will be out later this summer (most likely august)
Fury dual GPU is the only card that will be released this fall.

My mistake, I'm tired and I read it wrong. I thought review samples were sent out and sites weren't allowed to post benchmarks but I just thought it was going to be just their 300 series with the fury coming out in the fall.

That said, I'm still not impressed with it's performance. With Pascal coming out next year I really don't think AMD's cards are very competitive in the long haul, especially since this is considered next gen. When Pascal is released these cards wont be able to compete unless there is a very steep price drop.

You are speaking as if the GTX 980ti had been years on the market, actually it was just release and it seems the R9 Fury X is a better GPU. If you add that it's liquid cooled and has a new type of memory, it looks to me like a really good high end card

 

[Guest]

With nvidias next line up coming next year I don't really know what AMD hopes to accomplish. These cards wont be out until the fall and by then nVidias new chips will be right around the corner, which I'm sure will be much faster.

You are seriously overstimating how soon Pascal will be released. I can assure you it won't be in H1 2016. If they're lucky, they'll manage a Q3 launch, but Q4 or H1 2017 is more realistic. Nvidia not only depends on TSCM's new manufacturing process, which is already having problems compared to GlobalFoundries/Samsung (meaning AMD will probably get it first), but also on the release of HBM2, and then it will be their first time implementing it, which will also require a bit extra time to get right. This, assuming they don't screw up like they did when moving from GDDR3 to GDDR5 (they only really got that right on Kepler).
Since GlobalFoundries is currently ahead of TSMC and AMD already has HBM expeience, chances are Arctic Islands will be out before Nvidia launches Pascal.

 

[yRaz]

You are speaking as if the GTX 980ti had been years on the market, actually it was just release and it seems the R9 Fury X is a better GPU. If you add that it's liquid cooled and has a new type of memory, it looks to me like a really good high end card

The difference here is that the 980Ti is based on nVidia's current series of cards instead of a new architecture. This is over simplifying things, but it's essentially a more powerful GTX 980. This is suppose to be AMD's new architecture and it's not beating nVidias existing tech by a very large margin. I do not see Fury's performance compelling enough to justify paying $650 when something much more powerful is going to be out next year.

Since GlobalFoundries is currently ahead of TSMC and AMD already has HBM expeience, chances are Arctic Islands will be out before Nvidia launches Pascal.

Well I hope AMD comes out with something before nVidia's next gen or they are going to be seriously behind for awhile. Perhaps I am overestimating, but these performance numbers aren't really compelling enough to upgrade from videos in the last few years. The market is already flooded with similar performance videocards, it's going to take something really new before people are going to shell out the cash for a new graphics card. I was honestly expecting more out of fury than what these initial numbers are telling us. By all the hype it sounded like the Fury X would land somewhere between the 980 TI and the Titan X. Here's to hoping it has lots of overclocking headroom

 

[Guest]

This is suppose to be AMD's new architecture and it's not beating nVidias existing tech by a very large margin.

That is not correct. Arctic Islands will be AMD's new architecture (GCN 2.0, supposedly), which will only be released next year on GlobalFoundries' 14 nm process and with HBM2. Fiji, the GPU used on these Fury cards, is based on the same GCN 1.2 architecture used by Tonga, which outside of a few improvements is basically still the same GCN architecture AMD has used since the HD 7000 series. Fiji is just a larger version of Tonga (larger than Tahiti even, having 64 compute units) with HBM instead of GDDR5.
If anything, I think this should put AMD in a good light, because with only minor improvement (and HBM, not so minor) their three-year-old architecture is still quite competitive with the best Nvidia has to offer. Which is good for us customers, regardless of what brand you prefer.

 

[theBest11778]

This is suppose to be AMD's new architecture and it's not beating nVidias existing tech by a very large margin.

That is not correct. Arctic Islands will be AMD's new architecture (GCN 2.0, supposedly), which will only be released next year on GlobalFoundries' 14 nm process and with HBM2. Fiji, the GPU used on these Fury cards, is based on the same GCN 1.2 architecture used by Tonga, which outside of a few improvements is basically still the same GCN architecture AMD has used since the HD 7000 series. Fiji is just a larger version of Tonga (larger than Tahiti even, having 64 compute units) with HBM instead of GDDR5.
If anything, I think this should put AMD in a good light, because with only minor improvement (and HBM, not so minor) their three-year-old architecture is still quite competitive with the best Nvidia has to offer. Which is good for us customers, regardless of what brand you prefer.

If these numbers are accurate, Nvidia's beat this gen at the top. For the same $650 AMD gives you a liquid cooled GPU with better performance, and of course noise and thermals. There's no reason at all to consider a 980ti. As for Nvidia's next GPUs, if you play the, well next year game, you'll never buy a GPU, because next year (regardless of what year it is,) there will always be a better GPU.

 

[OortCloud]

"As for Nvidia's next GPUs, if you play the, well next year game, you'll never buy a GPU, because next year (regardless of what year it is,) there will always be a better GPU."

My thoughts exactly - you saved me some typing thanks

 

[WangDangDoodle]

I'm hoping these numbers repeat themselves in independent reviews. This is actually looking really good for AMD. For some reason, I'm surprised at the fact that they're asking so much for the Fury X. They could make a total killing by undercutting Nvidia a little bit right now. I'm not paying 650$ for a graphics card. I think 550$ is quite enough, and I would buy a Fury X at that price even though it's a small bump up from my 980.

 

[Vaibhav Sharma]

I don't even wanna imagine how much this card will cost in India, probably more than 1000$

Gtx 970 is the only card here with sane pricing of 450-500$

 

[Captain828]

Surprise! These are the exact same charts that have been circulating on the internet for a while now.

What does not impress me at all about this Fury X GPU is:
1. eats up a lot of space in the case due to the water cooling bit
- you essentially have a 2-slot card + a radiator (which can be problematic with CPU cooler towers) + tubing to handle
- not sure why they didn't go for a standard water block...

2. since both 980 Ti and Fury X are priced similarly, power consumption can become a deciding factor
- here the 980 Ti takes the cake due to its high efficiency
- note: this is based on how the 390X performs (same GCN 1.2 arch) + the very large die-size and TDP of the Fury X

3. only 4GB of VRAM can hinder performance in games that are VRAM *****s (especially on 4K res!)
- this will have to be tested in a real scenario however, but I do remember people with 980's complaining about it before where their FPS just took a nose-dive due to VRAM limitation in 4K resolutions

4. same GCN 1.2 architecture, just scaled out to a bigger chip

5. no HDMI 2.0 support

Some things I do like:
1. HBM technology
2. can fit in smaller cases due to the short length of the PCB (mostly attributed to HBM usage)

One thing I am very concerned about: operating temperature.

Yes, I know, it uses water cooling, however it is based on the same inefficient (when compared to Maxwell) GCN 1.2 arch, scaled-out to a larger die size AND the HMB memory is on the GPU.
Based on the above, I am expecting to see operating temperatures of around 80-90C, but I hope this won't be the case in reviews.

 

[dividebyzero]

You are seriously overstimating how soon Pascal will be released. I can assure you it won't be in H1 2016. If they're lucky, they'll manage a Q3 launch, but Q4 or H1 2017 is more realistic.

In which case it will be the slowest production ramp in history, or at least since...GloFo's 28SHP.
It also means AMD's Greenland GPU is royally screwed as well since it is also slated for HBM2.

Nvidia not only depends on TSCM's new manufacturing process, which is already having problems compared to GlobalFoundries/Samsung (meaning AMD will probably get it first)

Meanwhile in the real world, Samsung's 14nm yields are so bad Apple transferred production to TSMC

Since GlobalFoundries is currently ahead of TSMC.

Really. Last I heard, GloFo's yields started at abysmal, and are only just now on the way to Meh, and have only just started tape outs of FGPA's and SOC products (traditionally pipe cleaner products for a new process).
On the other hand, TSMC already has 16nmFF online, 16nmFF+ according to TSMC:

16FF+ yield is already approaching 20SoC yield. He said 16FF+ provided better maturity at risk production than any previous TSMC process...[snip]...TSMC executives cited impressive progress with the 16FF+ process node, noting that it has received over 12 tapeouts so far and that a total of 50 tapeouts are expected for 2015..

Some pretty high profile ASIC (including some small production runs), FGPA, and SoC customers are in that list, and 16FF/16FF+ is expected to account for 7-9% of TSMC's revenues by Q4 2015- which isn't bad considering the company churns out a lot of wafers on older processes (65nm, 130nm, 180nm) as well as 20, 28, and 40nm.

So, thanks for the pure play foundry update, but I'd suggest you actually check out little things like contracts, cancellations, wafer starts, and tape outs.

 

[Guest]

Think you will find with HBM the vram issue is much less, the higher bandwidth ( 512gbps) at 4 k will matter more due to throughput.
Operating temps are stated at 50 degrees c which leaves massive room for overclocking.

Power needs are also stated at 273w compared to the 250w of the 980Ti

This for me looks like a really good card for overclocking considering the capability of the cooling.

 

[baN893]

You are seriously overstimating how soon Pascal will be released. I can assure you it won't be in H1 2016. If they're lucky, they'll manage a Q3 launch, but Q4 or H1 2017 is more realistic.

In which case it will be the slowest production ramp in history, or at least since...GloFo's 28SHP.
It also means AMD's Greenland GPU is royally screwed as well since it is also slated for HBM2.

Nvidia not only depends on TSCM's new manufacturing process, which is already having problems compared to GlobalFoundries/Samsung (meaning AMD will probably get it first)

Meanwhile in the real world, Samsung's 14nm yields are so bad Apple transferred production to TSMC

Since GlobalFoundries is currently ahead of TSMC.

Really. Last I heard, GloFo's yields started at abysmal, and are only just now on the way to Meh, and have only just started tape outs of FGPA's and SOC products (traditionally pipe cleaner products for a new process).
On the other hand, TSMC already has 16nmFF online, 16nmFF+ according to TSMC:

16FF+ yield is already approaching 20SoC yield. He said 16FF+ provided better maturity at risk production than any previous TSMC process...[snip]...TSMC executives cited impressive progress with the 16FF+ process node, noting that it has received over 12 tapeouts so far and that a total of 50 tapeouts are expected for 2015..

Some pretty high profile ASIC (including some small production runs), FGPA, and SoC customers are in that list, and 16FF/16FF+ is expected to account for 7-9% of TSMC's revenues by Q4 2015- which isn't bad considering the company churns out a lot of wafers on older processes (65nm, 130nm, 180nm) as well as 20, 28, and 40nm.

So, thanks for the pure play foundry update, but I'd suggest you actually check out little things like contracts, cancellations, wafer starts, and tape outs.

I love you DBZ. I always enjoy reading your comments.

On a side note, I was hoping the switch to HBM would lead to higher performance increases than what AMD is showing here. I am not a fanboy of either camp, I merely want progress. Hope Pascale will give me the jump in performance that I am hoping for.

 

[Guest]

900% faster memory is not impressive, lol

 

[jasonSU]

But still, AMD needs 4096 bit memory interface to just overcome 980ti for like 5 % ...

 

[Cryio]

What does not impress me at all about this Fury X GPU is:
- no HDMI 2.0 support

While I understand the disappointment of no HDMI 2.0, is really no one using DisplayPort ?

 

[Guest]

You have no idea what you are talking about, this is not GCN 1.2, the tdp is 275 watt, similar to 980 ti's 250 watts but it performs better. The Nano is a 175w watt TDP , 7 teraflops beast.
I repeat, you have no idea what fiji soc is, inform yourself before posting please.

 

[EEatGDL]

Serious, "this is not impressive", first HBM card to ever be released..... fairly amazing card really and the Fury Nano will be hectic.

Saying the performance on these cards is not impressive because Nvidia will be faster when they release Pascal when it is 12 months away... seems a bit Nvidia fanboyish. AMD will have their second gen HBM out shortly after that which will be faster again.

Personally, I'm not that much impressed, simply because of the timing. A bit late and everything they got, to -in some cases- barely top NVIDIA's offering. This is all over again when AMD cards with GDDR5 memory were competing head-to-head with NVIDIA cards using GDDR3. They applied a new memory technology to minimize bottlenecks in fetching data, but without a properly designed architecture to make full use of it.

Pascal, on the other hand, is designed with HBM memory in mind from the beginning, trying to exploit its full potential. I'm not saying to wait a whole year for it; just that that's the way a new technology must be implented: designing everything else around it and not just pasting it like a cheap patch. Improvising and attaching a turbine to a car won't nearly make good use of it due to the limitations of the car; what you're looking to is a plane, with the according design to make full use of the turbine. The car with the turbine may reach 150 mph quick and safely, but not much more speed due to the design of the car, without losing control of it -loss of traction at higher speeds; while you could achieve 700 mph with the same turbine on a fighter jet.