Thats not new, they've been recycling that for a long time. They also had no idea of the potential Titan X perf/power prior to launch, when other leakers have been spot on.
Thats not new, they've been recycling that for a long time. They also had no idea of the potential Titan X perf/power prior to launch, when other leakers have been spot on.
Well, if this helps some sleep better at night enjoy it. There has been no indication that this is reality though.
5150Joker
Diamond Member
Edit: nevermind, misread what you wrote. When did they mention 20% over 980 before? This is the first time I've seen that.
There's no indication that ANY of the speculation is reality. That's the beauty of speculative reporting.
Last edited:
AtenRa
Lifer
- Feb 2, 2009
- 14,003
- 3,362
- 136
Despoiler
Golden Member
- Nov 10, 2007
- 1,968
- 773
- 136
Here is my response to that article.
Fudzilla's reading comprehension fails. "The slide below clearly implies that you need Affinity multi-GPU rendering in order to reduce latency and increase content quality."
No the slide does not imply that your *need* multi-GPU rendering. That slide shows they have multi-GPU support in the SDK and their multi-GPU implementation reduces latency and increases quality. ie using SFR when Crossfire is in use. Not only that, where would they come up with the fact that FIJI was dual GPU on the same card? If they are claiming the slide proves something wouldn't having multiple single chip FIJI GPUs in the demo boxes also be a possibility? I'd be embarassed to put my name on this article as the author.
Fiji XT (3072 sp) - 15 - 20% faster than GTX 980
Bermuda Pro (3840 sp) - 35 - 40% faster than GTX 980
Bermuda XT (4096 sp) - 45 - 50% faster than GTX 980 (60 - 65% faster than R9 290X)
i am guessing this is how these chips end up.
The frames per watts used is the basis for performance measurements now. Who cares how many frames card X gets if card Y uses less watts!
Just kidding....Or am I....Hmm.
The optimist(for AMD) view is that the captain jack rumor about being 20% faster than 980 is the 380X and the 390X is >60% faster than 290X and beats Titan X by at least 10%.
The pessimist view is that the captain jack is 390X and is only 20% faster than 980, with the other rumors being just hot air or at the most there is another dual-gpu card which barely beats Titan X. And the rebrands with no freesync or updated GCN.
The pessimist view is that the captain jack is 390X and is only 20% faster than 980, with the other rumors being just hot air or at the most there is another dual-gpu card which barely beats Titan X. And the rebrands with no freesync or updated GCN.
RussianSensation
Elite Member
- Sep 5, 2003
- 19,458
- 765
- 126
As I have already stated, this is total nonsense. If you take 4096SP/256TMU part clocked at 1.05Ghz and add Tonga's 40% improvement in colour fill-rate/memory compression + HBM, you cannot end up only 20% faster than a 980 at 4K. It's physically impossible because of how well GCN scales. That's not even taking into account that AMD should double the geometry performance from R9 290X based on the fact that Tonga has MORE tessellation performance (and even pixel fill-rate) than the mighty R9 290X. VR Zone is smoking something or cherry picking some 1080P/1200P scores with AMD CPU driver overhead.
Right now a 980 is only 8-15% faster than a 290X. An R9 390X with 53% higher theoretical performance increase in terms of shaders, pixel throughput and texture units will blow the doors off a 980 at high resolutions. Let's use some common sense for once.
European sites show even less advantage for a 980 at 4K vs. 290X than US based sites:
6% for Computerbase
6% for Sweclockers
The fact that the editor of VR-Zone even posted such non-sense as an R9 390X is only 20% faster than a 980 shows he is completely clueless as to the scaling of GCN architecture, architectural improvements in Tonga, and the rumoured specs of R9 390X.
Elfear
Diamond Member
- May 30, 2004
- 7,163
- 819
- 126
I agree. The only way the 980 + 20% rumor is correct is if 390X isn't a 4096SP/256TMU part. If AMD wasn't able to squeeze that many shaders onto a 28nm chip, we would see less performance than we're hoping for.
ShintaiDK
Lifer
- Apr 22, 2012
- 20,378
- 145
- 106
How fast would a 300W 2x2048SP crossfired part be in average? Just for the theoretical part. The 295X2 is 500W and 2x2816SP.
How long is a piece of string?
We don't know how Fiji behaves so extrapolating from Tahiti or other GCN is pointless.
ShintaiDK
Lifer
- Apr 22, 2012
- 20,378
- 145
- 106
Point is you can have 4096Sps and only be 20% faster in average. Its not something impossible.
AnandThenMan
Diamond Member
- Nov 11, 2004
- 3,991
- 626
- 126
ShintaiDK
Lifer
- Apr 22, 2012
- 20,378
- 145
- 106
Who knows. We dont even know if its a new chip, a dual GPU card or a rebrand and there isnt even any release date. Only the 310 to the 370 series is known from the drivers.
That would blindside every tech news site. It would imply AMD has full control & able to prevent AIB leaks.
That would be my guess also. And further improvements for HBM parts starting from release.
I don't consider fudzilla or wccftech as tech news sites. they are click bait trash. videocardz is slightly better than these 2. As for the leaks I am confident chiphell was spot on with their leaks.
http://www.overclock.net/t/1530716/...-fiji-380x-and-bermuda-390x-benchmarks-leaked
that chiphell leak between Bermuda XT and GM200 shows just 5 games. GM200 Titan-X's lead over R9 290X in games like Farcry 4, Middle Earth Shadow of Mordor, Civilization Beyond Earth is 25 - 35%. CRYENGINE games like Ryse and Evolve also show a similar 30 - 35% lead for Titan-X over R9 290X. I am also guessing that the lower texture unit count (192 in GM200) is going to show up in R9 390X (with 256 TMU) being a huge 20 - 25% faster than GM200 Titan-X in games where performance is limited by texturing performance. On average across a wide range of games and across a range of reviews we will see the performance play out just as I predict. the R9 390 will match or edge out the Titan-X while the R9 390X will take the GPU crown by a good margin (8 - 10%).
flopper
Senior member
- Dec 16, 2005
- 739
- 19
- 76
Main reason nvidia released the titanx now instead of waiting as they knew what is coming and then they wouldnt sell many cards.
guskline
Diamond Member
- Apr 17, 2006
- 5,338
- 476
- 126
flopper, Nvidia would still sell Titan Xs even if AMDs new card outperformed it by 10% and undercut it by $200.
I'm not be critical. Nvidia, much to its credit, has built an incredibly loyal fan base.
But for the GTX970 snafu, I've been impressed with their cards.
Perhaps the new CEO of AMD is aware of this and is holding back the 390/390X until she is sure it will release as a polished product. Due to the wide Nvidia fan base, they can afford to release product and suffer some negative reviews without as much overall affect.
On the high end AMD is still relying on the R9-290/290X. The chip was solid but the reference cooler was poor. The R9-295 is a great but niche product that was way too expensive on release (great price now BUT the new chip is almost here).
Since then Nvidia has released the Maxwell architecture and taken the gamer base by storm.
Simply put, from what I read Titan X is King.
I'm not be critical. Nvidia, much to its credit, has built an incredibly loyal fan base.
But for the GTX970 snafu, I've been impressed with their cards.
Perhaps the new CEO of AMD is aware of this and is holding back the 390/390X until she is sure it will release as a polished product. Due to the wide Nvidia fan base, they can afford to release product and suffer some negative reviews without as much overall affect.
On the high end AMD is still relying on the R9-290/290X. The chip was solid but the reference cooler was poor. The R9-295 is a great but niche product that was way too expensive on release (great price now BUT the new chip is almost here).
Since then Nvidia has released the Maxwell architecture and taken the gamer base by storm.
Simply put, from what I read Titan X is King.
Last edited:
RussianSensation
Elite Member
- Sep 5, 2003
- 19,458
- 765
- 126
R9 295X2 came out April 8, 2014 for $1499 and dropped to $650-660 10 months from launch. Dual Titan Xs today cost $2000 but in 4-6 months from now that level of performance could be had for $1300. Next fall, a $500 14nm/16nm will beat the Titan X (recall $330 GTX970 ~ $700 GTX780Ti) as 14nm cards with HBM1/2 should bring a massive increase in perf/watt and overall performance. People have short memories but when R9 290 launched at $399, it made a FAR bigger impact on the overall market in terms of price/performance than a $330 970 did. Before 290 came out, that level of performance was 62.5% higher ($650 780). Usually AMD's 2nd tier cards (i.e., 5850/6950/7950/290) bring an incredible price/performance at the high end. I have little doubt that those who skip dual Titan Xs will be immensely rewarded by the summer time with a GM200 6GB/390 non-X.
The Titan X is a win-win for NV and its loyal customers. It lets NV maximize profits and profit margins for months until R9 390X launches and forces NV to release a faster clocked GM200 card. These profits help to fuel R&D for future NV products; and it lets enthusiasts, esp. those loyal to NV to have the bragging rights/benchmark records for months before any other competing card drops. However, in the grand scheme of things for 99% of the PC gaming market, the Titan X is a 'marketing/meh' videocard.
The PC industry moves forward when performance of cards like 980 falls to $299, and when a card 35-50% faster than an R9 290 is $450-500. A $999 single-chip card's appeal is so limited outside of semi-pro users and benchers, it might as well not even exist. At least the original Titan had a legitimate purpose - CUDA DP performance. That's why no matter what the hype is behind Titan X today, in 6 months, it'll be forgotten as another overpriced NV 'Titan marketing' exercise. The best part about Titan X is approximating where R9 390X and GM200 6GB will land. It also forms a solid basis for the level of performance of a mid-range Pascal/14nm card 2016 (i.e., NV's next gen mid-range card tends to be as fast or faster than last generation's flagship). So look at the Titan X and imagine a $400 14nm card with that level of performance in 1.5 years from now
.The one thing seems to have remained consistent since the slowdown of the GPU industry.
GTX580 --> 780Ti ~ 2X the performance increase in 3 years
HD7970/680/7970Ghz --> Titan X ~ 2X the performance increase in 3 years
Hopefully the move to 14nm GPUs + HBM2 will double the performance over Titan X in a shorter time-frame. I think there is a possibility that by March 2018, we might have a GPU more than 2x faster than the Titan X!
No, you cannot at high resolutions and 1.05Ghz clock speed per the leaked AMD slides. GCN scales nearly linearly and R9 390X will have architectural improvements from R9 280X/7970Ghz based on improvements AMD already made with Tonga. Add HBM, and R9 390X should be at least 2X faster than a 7970Ghz. That alone gets us to Titan X performance right away.
However, we know that Tonga improved pixel fill-rate where despite 1792 SPs + 32 ROPs, it has a higher pixel-fill rate than a 2560SP+64 ROP R9 290, and likely higher than the 2816 SP 290X! That's an incredible in pixel fill-rate performance per mm2 / per ROP that nearly everyone keeps ignoring.
And an incredible 70% higher geometry performance compared to a 2816 SP R9 290X. That means R9 390X will have > 3X the geometry performance of an R9 290X.
Thus, a 4096 SP GCN 1.2/1.3 with 256 TMUs and 500GB/sec+ memory bandwidth should be more than double HD7970Ghz at high resolution gaming, which is WAY higher than 20% over a 980. If you want to cherry-pick low resolutions like 1680x1050 or 1080P, then sure it might only end up 20% faster in older titles but that's not an indication of the actual performance difference in modern games @ high resolutions.
Last edited:
Going by the HFR 285 review, the 390X should be 45.6% faster then 290X.
It's the review I used for this post:http://forums.anandtech.com/showthread.php?p=37248171&highlight=#post37248171
RussianSensation
Elite Member
- Sep 5, 2003
- 19,458
- 765
- 126
This is not correct based on your own analysis. You compared 285 to 290 but the former has 918mhz clocks and the latter has 947mhz clocks. However, 290X has 1Ghz clocks and R9 390X is rumoured at 1.05Ghz.
1.05Ghz x 4096 / (1Ghz x 2816) = 52.7%. In your CU scaling you are ignoring clock speeds. Furthermore, how many reviewers ran R9 290X in normal mode, which meant it was throttling a bit? It's no wonder that 1Ghz after-market R9 290X cards run 5-7% faster in reviews. With WCE, R9 390X should not throttle which means this gives R9 290X a disadvantage in most reviews that are not using Uber mode.
Even if in a review under normal mode R9 290X dipped to 950mhz, we'd already be looking at:
1.05Ghz x 4096 / (0.95Ghz x 2816) = 60.7%.
These tiny 50mhz differences add up on both sides.
Last edited:
Between the 285 and 290 it's almost nothing (0.03%) and going by the fact the the 290 also dropped in frequency when benchmarking, I didn't find it interesting to investigate.
You are right about the 45.6% I posted compared to 290X, I should have added the diff in frequency.
But I don't think multiplying the frequency with the cores is the right thing to do. Going by an other hardware.fr test, the diff between 290 and 290X is around 7%. That's far from the 18% when comparing CU*clocks.
If I find the ODS file I will tweak the numbers.
My point was not to take the 1.05Ghz from the rumors, but only the known max CU's that GCN supports and a scaling factor already known from Tonga.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads
- Started by Tigerick
- Replies: 21K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter