[AMD] Details on Fiji VR/395x2 surface and launch month

[CoolaxGaming]

But what makes this different from the other posts? Well this is taking into account the new details on FIJI VR, a shall fiji xt gpu.

Fudzilla and WCCFTech have reported that their "sources" have informed them on some of the details of the upcoming AMD R9 3xx series.

Advanced Micro Devices is in a very tight spot right now with it loosing money and market share, so they desperately need a new launch

There was some speculation about Fiji VR before. So all we know right now, is that Fiji VR is going to be a dual GPU, possibly FijiXTx2, and if AMD wants to continue their naming scheme, possibly the R9 395X2.

With VR improvements coming from all over the lace, "Fiji VR" coupled with Liquid VR, might just win the market for AMD, but this is unlikely to happen this year. AMD is in such a spot

Fudzilla writes -

As far as our sources are aware, the launch date is in June or the last month of the second quarter. AMD should launch the Carrizo APU at Computex at some point after June 2nd, and we heard that Fiji might come a week or two later, if not at the same event.

There will be two Fiji cards, Fiji XT a performance GPU for high end gamers and Fiji VR that will have two Fiji GPUs on a single PCB. Fiji XT is faster than Nvidia’s Geforce GTX 980, but it is unlikely that it will end up faster than the Geforce Titan.

And from WCCFTechs article -

Source -

http://www.fudzilla.com/news/graphics/37566-two-amd-fiji-cards-coming-in-june

http://wccftech.com/amd-launching-fiji-based-graphics-cards-june/

NOTE: This is NOT the same as R9 300 being launched in June, this is about Fiji VR

[werto165]

390x please be good, please be good, please be good! 

[spartaman64]

im confused the r9 395x2 has less ram per core than 390x

[ErickS89]

390x please be good, please be good, please be good! 

You're hoping for gold man... I don't see AMD delivering sadly.

[CoolaxGaming]

You're hoping for gold man... I don't see AMD delivering sadly.

Yeah, they got soo little R&D, it would be amazing if the card even beat the 980 at a cheaper price..

[CoolaxGaming]

Topic editing is now finished.

[ErickS89]

Why not? CPU-wise I can see your skepticism, but they've been delivering in the GPU sector. Plus, if HBM is everything it's meant to be, it will be faster than Nvidia's counterparts.

Because AMD knows how to hype something that's so good but in reality it's far worse or not even as good as the competition. This has been going on for years!

[The Benjamins]

im confused the r9 395x2 has less ram per core than 390x

it will launch with 4GB then later a 8GB variant will come out later

[CoolaxGaming]

it will launch with 4GB then later a 8GB variant will come out later

Thats the 390x

[MrRavens]

Cmon AMD let's see your new child!

 

Also @CoolaxGaming

 

was it suppose to say "all over the place" instead of "all over the lace"?

[CoolaxGaming]

Cmon AMD let's see your new child!

 

Also @CoolaxGaming

 

was it suppose to say "all over the place" instead of "all over the lace"?

All over the place.

God damnit, for some reason, I need to press the P key REALLY hard compared to other keys.

[MrRavens]

All over the place.

God damnit, for some reason, I need to press the P key REALLY hard compared to other keys.

 

haha, tragic

[mvitkun]

Because AMD knows how to hype something that's so good but in reality it's far worse or not even as good as the competition. This has been going on for years!

What are you talking about?

 

Nvidia has been saying the exact same thing about how performance with stacked memory will be so much better.

 

Even if you disregard HBM's impact on the performance of the R9 300 cards, which you really shouldn't, it has ~30% more cores.

So at the very least, if you expect there to be no improvement on the Fiji XT core vs the Hawaii XT core, it should perform 30% better than a 290X.

That leaves it nipping at the heels of the Titan X and once in a blue moon even beating the Titan X.

Seriously, try it out and multiply the fps rendered by the 290X by 1.3. It'll land smack dab in the middle of the 980 and Titan X.

 

That's if you blindly and stupidly ignore improvements to the core design and performance improvements from HBM.

[patrickjp93]

Why not? CPU-wise I can see your skepticism, but they've been delivering in the GPU sector. Plus, if HBM is everything it's meant to be, it will be faster than Nvidia's counterparts.

 

 

What are you talking about?

 

Nvidia has been saying the exact same thing about how performance with stacked memory will be so much better.

 

Even if you disregard HBM's impact on the performance of the R9 300 cards, which you really shouldn't, it has ~30% more cores.

So at the very least, if you expect there to be no improvement on the Fiji XT core vs the Hawaii XT core, it should perform 30% better than a 290X.

That leaves it nipping at the heels of the Titan X and once in a blue moon even beating the Titan X.

Seriously, try it out and multiply the fps rendered by the 290X by 1.3. It'll land smack dab in the middle of the 980 and Titan X.

 

That's if you blindly and stupidly ignore improvements to the core design and performance improvements from HBM.

 

How many times do I have to explain to people games have no bandwidth problem currently? Scientific computing is one thing, but games have backpressure in the ROP-TMU pipeline as it is. Giving it faster/wider memory access will only help those code segments which weren't at their core limits already, which is why core clock boosts still consistently help far more than memory overclocking. HBM really won't help gaming much for a very long time. It will make AMD's platform a bit more attractive for scientific computing though, especially given the power savings, at least until Pascal launches.

 

I hope AMD is smart enough to decrease the core count per ROP/TMU at some point, because currently the 128 rule is stupid and leads to that backpressure. 96 would be more than sufficient and give more ROPs & TMUs overall in addition to lowering the ratio, leading to far better performance overall and better scalability.

[sgloux3470]

How many times do I have to explain to people games have no bandwidth problem currently. 

 

Except there is totally a difference.  It's immediately apparent when we compare Maxwell 256 bit bus w/ compression to AMD's 512 bit bus.  

 

[ZestyJalapeno]

Here is me waiting on eBay for the richer people to dump their 295x2s...

[patrickjp93]

Except there is totally a difference.  It's immediately apparent when we compare Maxwell 256 bit bus w/ compression to AMD's 512 bit bus.  

-snip-

 

1) Learn some statistics. 3 test rounds is not reliable. 5 is the bare minimum for a 95% confidence interval

2) You're looking at a 256-bit vs. a 512, and not everything is a compressible texture. A good mass of that memory being accessed is still instructions and other data. Color compression only assists for some of the data. And it's well known color compression and decompression adds calculative overhead to the pipeline. It should come as no surprise. The analysis is simply flawed.

3) It's the 970 which we know runs into performance issues as you exceed a certain frame buffer size due to the cross bar interference to reach the last 0.5 GB.

 

4) (256/8)*7*10^9 = 224*10^9 = 224GB/s for GTX 970

     (512/8)*5*10^9 = 320*10^9 = 320GB/s for R9 290X

 

The bandwidth is in AMD's favor by a near 50% margin (and Nvidia uses a 384-bit bus for high end cards anyway, but the benefit ends around 250GB/s), but the performance is easily in Nvidia's. This should be more than enough proof bandwidth stopped being a problem a long time ago for games. Color compression did not shrink the total frame buffers by 50% for Maxwell and Tonga, so really you only proved my point for me.

 

HBM will do practically nothing for gamers. For scientific computing it will be a huge blessing.

[GrimNeo]

Here is me waiting on eBay for the richer people to dump their 295x2s...

295x2s are cheap in my country. Less than $700.

[han_han08]

 

4) (256/8)*7*10^9 = 224*10^9 = 224GB/s for GTX 970

     (512/8)*5*10^9 = 320*10^9 = 320GB/s for R9 290X

 

The bandwidth is in AMD's favor by a near 50% margin (and Nvidia uses a 384-bit bus for high end cards anyway, but the benefit ends around 250GB/s), but the performance is easily in Nvidia's. This should be more than enough proof bandwidth stopped being a problem a long time ago for games. Color compression did not shrink the total frame buffers by 50% for Maxwell and Tonga, so really you only proved my point for me.

 

HBM will do practically nothing for gamers. For scientific computing it will be a huge blessing.

 

owwkay...how did benefit end around 250GB/s again?

 

just curious, some of my friends are actually using memory interface in choosing the BETTER gpu, with the explanation going on, apparently it really doesn't

 

i can remember watching in PCper that gtx 960's specs were:

Memory frequency = 7ghz

Effective mem frequency = about 9ghz (i think because of maxwell's compression)

 

tom (from nvidia) was saying "you can't explain the effective frequency w/o the actual frequency, etc..." 

marketing? no?

 

how is "4096 WIDE IO" the memory interface?

Edited April 21, 2015 by han_han08

[sgloux3470]

1) Learn some statistics. 3 test rounds is not reliable. 5 is the bare minimum for a 95% confidence interval

2) You're looking at a 256-bit vs. a 512, and not everything is a compressible texture. A good mass of that memory being accessed is still instructions and other data. Color compression only assists for some of the data. And it's well known color compression and decompression adds calculative overhead to the pipeline. It should come as no surprise. The analysis is simply flawed.

3) It's the 970 which we know runs into performance issues as you exceed a certain frame buffer size due to the cross bar interference to reach the last 0.5 GB.

 

4) (256/8)*7*10^9 = 224*10^9 = 224GB/s for GTX 970

     (512/8)*5*10^9 = 320*10^9 = 320GB/s for R9 290X

 

The bandwidth is in AMD's favor by a near 50% margin (and Nvidia uses a 384-bit bus for high end cards anyway, but the benefit ends around 250GB/s), but the performance is easily in Nvidia's. This should be more than enough proof bandwidth stopped being a problem a long time ago for games. Color compression did not shrink the total frame buffers by 50% for Maxwell and Tonga, so really you only proved my point for me.

 

HBM will do practically nothing for gamers. For scientific computing it will be a huge blessing.

 

1. i don't have all day.  Just doing a quick comparison, but you will absolutely find the same results if you gathered benchmark data from other websites.

 

2. Yes, of course 256 bit with compression isn't as good.  That's the problem.  Maybe a 256 bit bus without the compression overhead would have less issues, but we're talking about real products not theoretical.

 

3. You see the same effect with the 980.  It's a Maxwell issue that is present on the 970 and 980 but not the TitanX (which retains 40% better performance than a 290x regardless of resolution)  This is obv because of the 384 bit bus on the TitanX removing the bottleneck that we see with the other two cards.

 

The fact that it's not present in the TitanX does support your argument that there is a point where it doesn't matter, but I think it's premature to say HBM won't make a difference going forward when we can clearly see Maxwell 256 bit cards being hamstrung by their memory bus.  

 

I'm sure someone said the same thing about GDDR5 back in the day or SATA 6.  Available bandwidth is always saturated eventually.  It might not matter that AMD is using HBM for the 390x, but you better believe it will be the standard eventually.

[ZestyJalapeno]

295x2s are cheap in my country. Less than $700.

Same price here roundabout...I forgot the word "cheap" is relative.

[GrimNeo]

Same price here roundabout...I forgot the word "cheap" is relative.

Well I'm getting two Titan X's so $2000 compared to $700 is cheap imo.

[ZestyJalapeno]

Well I'm getting two Titan X's so $2000 compared to $700 is cheap imo.

You reminded me, gotta watch out for people dumping their first gen Titans too, see if I can get two of them for a reasonable price vs the 295x2.

[ItsAFeature]

You reminded me, gotta watch out for people dumping their first gen Titans too, see if I can get two of them for a reasonable price vs the 295x2.

 

Doesn't the 295x2 trade blows with SLI Titans if not beat them? I seem to remember it putting the hurt on the Titan Z but I think that was clocked lower than the Titan.

[patrickjp93]

owwkay...how did benefit end around 250GB/s again?

 

just curious, some of my friends are actually using memory interface in choosing the BETTER gpu, with the explanation going on, apparently it really doesn't

 

i can remember watching in PCper that gtx 960's specs were:

Memory frequency = 7ghz

Effective mem frequency = about 9ghz (i think because of maxwell's compression)

 

tom (from nvidia) was saying "you can't explain the effective frequency w/o the actual frequency, etc..." 

marketing? no?

 

how is "4096 WIDE IO" the memory interface?

GPUs are intricate pipelines when it comes to graphics, even though in scientific computing only the cores themselves are used. There is a problem you don't often see discussed for GPUs, even though it's discussed extensively in CPU design: pipeline backpressure. Currently there are so many shaders per TMU and per ROP that even if the shaders are all fed, you can't make progress until the TMU or ROP finishes its task which uses data from 128 shaders, and then all of those ROPS and TMUs produce a final collaborated result before it all gets sent through the pixel shaders to the screen. If the shader count was 96 instead of 128 per ROP and the TMU count was increased in proportion, then you'd get a boost where the cores are staying fed and passing on more results. When that happens increasing bandwidth can be helpful. Until that time, there are some subsets in every scene ever drawn which don't produce back pressure. For those, increasing bandwidth helps, and this can increase fps, but not to the same degree increasing core clocks (which are also hooked to the ROPs and TMUs) will.

 

For the 960 you're well below that 250GB/s threshhold which is a good estimation even if not exact, and there are fewer cores and resources handling the same workloads, so depending on the driver's load balancing per frame, some more bandwidth may be beneficial.

 

The effective clock of 9GHz is a best-case scenario where all information is a highly compressible texture. It's some marketing, but there is truth that you can gain some extra space or more textures by using their color compression tech.

 

If you look at the HBM specs, each stack is a 1024-bit wide interface. There are 4 stacks this time. simply multiply.