I think it will be more than 50%. Probably like 7970 vs 6970. At release 40-50% faster, 1 year later 70% faster.
50% faster than 1x 290X.
50% faster than 1x 290X.
ShintaiDK
Lifer
- Apr 22, 2012
- 20,378
- 146
- 106
There are 2 ways AMD could get to 8 GB HBM. The first method is Hynix being able to deliver 4 Gigabit HBM chips and the other is AMD going for 8 memory stacks. We will soon know which option it is.
The last option you list still requires 8192bit. Hynix solution to increased density on HBM would be an 8 stack high HBM using current tech for 2GB per HBM.
ShintaiDK
Lifer
- Apr 22, 2012
- 20,378
- 146
- 106
Its in relation to 1 of the DX12 features (resource binding). Also pretty much shows its either a clueless intern or its made up by another one that makes money on fake news. Looking on the rest of the slide it begs for the last.
Last edited:
Its in relation to 2 of the DX12 features (resource binding). Also pretty much shows its either a clueless intern or its made up by another one that makes money on fake news. Looking on the rest of the slide it begs for the last.
It could also be related to conservative rasterization tier 3 ...
ShintaiDK
Lifer
- Apr 22, 2012
- 20,378
- 146
- 106
GCN doesnt support it. And why would you list a tier in the first place.
Its quite obvious the tier "inspiration" came from a forum.
Last edited:
@Bold Doesn't matter when we're talking about a new microarchitecture ...
https://msdn.microsoft.com/en-us/library/windows/desktop/dn859364(v=vs.85).aspx
I don't know about you but according to one of Microsoft's early documentation above you can see that there are 3 tiers of support for conservative rasterization ...
ShintaiDK
Lifer
- Apr 22, 2012
- 20,378
- 146
- 106
Are we? How do you know its not GCN 1.2? Same that Carrizo features and Tonga.
Shintai is adamant that it's GCN 1.2, that the watercooler is absolutely required, and that every other negative rumor is true. It's possible that he's right, but we'll see.
ShintaiDK
Lifer
- Apr 22, 2012
- 20,378
- 146
- 106
The main problem is the slide itself screams of being fake.
AMD never used iteration about a generation, only sites did.
Up to 4096, specifying a single product. Sub bullet point for no reason as well.
Mentioning a tier
Posting Tonga benefits taken from sites. (This is how AMD mentions it)
Slide not looking like an AMD slide.
Dated in the future for a march 16th event, what event is AMD holding there?
First GPU designed for VR, utter bollocks as well.
I am quite aware that "news" sites can pretty much sell sand in sahara these days. But it would help on the female type gossip news if people applied some kind of source critisism for once.
AMD never used iteration about a generation, only sites did.
Up to 4096, specifying a single product. Sub bullet point for no reason as well.
Mentioning a tier
Posting Tonga benefits taken from sites. (This is how AMD mentions it)
Slide not looking like an AMD slide.
Dated in the future for a march 16th event, what event is AMD holding there?
First GPU designed for VR, utter bollocks as well.
I am quite aware that "news" sites can pretty much sell sand in sahara these days. But it would help on the female type gossip news if people applied some kind of source critisism for once.
Last edited:
If you read up the VR presentable AMD did, it's not as bollocks as you believe. Their focus is amazing frame latency & high min fps as that's not just good for VR, its required else long play sessions will lead to severe vomiting & nausea. AMD went into great lengths about their technologies in the upcoming GCN to meet those performance/feature targets.
ShintaiDK
Lifer
- Apr 22, 2012
- 20,378
- 146
- 106
What new hardware support is being discussed in LiquidVR that current doesnt support?
How would you know ? Did you design HBM or the R9 390X . AMD did both. They co-designed HBM with Hynix and they designed the R9 390X. Until their design is revealed to the press and public you and I cannot say what is possible or not ? Both of us do not have a shred of knowledge other than what is found on Hynix HBM presentation slides. So lets wait for the actual facts to be revealed.
FatherMurphy
Senior member
- Mar 27, 2014
- 229
- 18
- 81
The word 'functionality' is also misspelled in the bottom bullet of the slide. Doesn't mean it is fake, but everyone knows how fake slides /charts tend to have misspellings.
Last edited:
I can think of a couple of reasons to include typos without much effort.
Weeding out the leaks. It's possible they could release slides with different typos to achieve this.
Keeping the speculation train rolling. Typos cast doubt.
destrekor
Lifer
- Nov 18, 2005
- 28,799
- 359
- 126
This is a fair guess. There are always intentional leaks, sometimes with accurate information, others intent to spread misinformation, all in the effort to feel out who is a source of leaks.
Black Octagon
Golden Member
- Dec 10, 2012
- 1,410
- 2
- 81
destrekor
Lifer
- Nov 18, 2005
- 28,799
- 359
- 126
Intentional as something that the company itself HOPED to happen, as opposed to some employee or other person "in the know" simply deciding to take it upon him/herself to leak it.
A company intentionally leaking it may do so either to help build hype, or to help weed out from where the leak is coming from, by giving different versions of documents to different individuals.
It is very difficult if not impossible to guesstimate performance of R9 390X by just looking at the amount of shaders.
One example from mobile, which I have been an enthusiast for quite some time is:
GTX 880M: 1536 shaders @ 993MHz. 256bit @ 1250MHz
GTX 970M: 1536 shaders @ 1038MHz. 256bit @ 1250MHz
Core/clock devided by core/clock would give you
(1536 * 1038MHz) / (1536 * 993MHz) = 1.04 = +4%
But reality was that GTX 970M was 20% faster than 880M.
The efficiency of the architecture was better than previous one.
So while 4096 shaders @ 1GHz devided on 2816 shaders @ 1GHz (390X WCE vs 290X) give you +45%. More can come out of a new/improved architecture.
One example from mobile, which I have been an enthusiast for quite some time is:
GTX 880M: 1536 shaders @ 993MHz. 256bit @ 1250MHz
GTX 970M: 1536 shaders @ 1038MHz. 256bit @ 1250MHz
Core/clock devided by core/clock would give you
(1536 * 1038MHz) / (1536 * 993MHz) = 1.04 = +4%
But reality was that GTX 970M was 20% faster than 880M.
The efficiency of the architecture was better than previous one.
So while 4096 shaders @ 1GHz devided on 2816 shaders @ 1GHz (390X WCE vs 290X) give you +45%. More can come out of a new/improved architecture.
You're comparing Kepler to Maxwell, which is obviously a lot different from comparing GCN to GCN.
Nobody knows the efficiency gains from GCN 1.1 (290X) to GCN 1.3 (390X) or GCN 1.1 to GCN 2.0 if that happens.
That was my point. Next, there are some big uncertain things like HBM, what that brings out on the table in terms of performance. Second, what about the 28nm SHP process in Global Foundries which AMD was suppose to use? That would in itself give a lot of gain.
Its already been confirmed atleast 10 new mobile cards from the 300 series. Hinting toward a massive change vs previous cards. 1 or 2 rebrands. Of course. But 10 rebrands? Nah, thats not happening. Its a big change in architecture level.
Here they are:
Advanced Micro Devices, Inc. driver update for AMD Radeon (TM) R5 M315
Advanced Micro Devices, Inc. driver update for AMD Radeon (TM) R5 M320
Advanced Micro Devices, Inc. driver update for AMD Radeon (TM) R5 M330
Advanced Micro Devices, Inc. driver update for AMD Radeon (TM) R5 M335
Advanced Micro Devices, Inc. driver update for AMD Radeon (TM) R7 M340
Advanced Micro Devices, Inc. driver update for AMD Radeon (TM) R7 M350
Advanced Micro Devices, Inc. driver update for AMD Radeon (TM) R9 M360
Advanced Micro Devices, Inc. driver update for AMD Radeon (TM) R7 M370
Advanced Micro Devices, Inc. driver update for AMD Radeon (TM) R9 M375
Advanced Micro Devices, Inc. driver update for AMD Radeon (TM) R9 M375X
+
R9 M385 which I have found benchmarks from
Last edited:
ShintaiDK
Lifer
- Apr 22, 2012
- 20,378
- 146
- 106
Just look at the 200 series. 10 SKUs, 9 rebrands.
It doesnt take much more than to open the driver and look to see if its rebrands or not.
And the driver is public avaliable via WSUS/Windows Update.
Last edited:
Low end will for sure be 100% rebrands. There is no value in making such a low power low margin chip.
Oland will continue to reign in mobile. If I had to guess everything 335 and lower is oland.
Oland will continue to reign in mobile. If I had to guess everything 335 and lower is oland.
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 + WCL Discussion Threads
- Started by Tigerick
- Replies: 22K
-
News NVIDIA and Intel to Develop AI Infrastructure and Personal Computing Products- Started by poke01
- Replies: 384
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
Facebook
Twitter