Why do you think it's a miracle? LolFaster performance IF bus width is the same. In this case AMD can't even reach spec sheet 1Ghz at nasty voltages, what makes You think same engineers somehow achieved a miracle of latencies?
I'm not disagreeing with you. So get off of your high horse and stop being so condescending.Do you guys not understand the difference in access latency and bandwidth? I can't believe we're having this discussion in 2017.
For gods sake people, learn the basics before arguing about GPU's. Some people still remaining want to discuss technical merit and not garbage by so called consumers.
On one hand we know the full specs of Threadripper, we know core counts, we even know price. They didn't do "blind" tests with Threadripper, since they knew they had a winning product, and no need to play such games.
On the other hand, Vega RX is cloaked in crazy marketing schemes, no solid information if you want to believe it doesn't have the same specs as the FE, and telling people @PDXLAN that they can't release any information on a unreleased product is untrue--look at Threadripper again.
So, if you are showcasing one unreleased product, yet using smoke & mirrors on the other, what does AMD expect people to think?
This basically means that Raven Ridge if it will have HBM2 on top of it will have to use "different wafers" and potentially would have to have different design.Interposers are not very expensive, so we can please drop that recurring statement. Passive interposers are very simple (copper conductors), made on old processes and I doubt there is a yield problem. A 2012 article.
http://electroiq.com/blog/2012/12/lifting-the-veil-on-silicon-interposer-pricing/
"Sesh Ramaswami, managing director at Applied Materials, showed a cost analysis which resulted in 300mm interposer wafer costs of $500-$650 / wafer. His cost analysis showed the major cost contributors are damascene processing (22%), front pad and backside bumping (20%), and TSV creation (14%).
Ramaswami noted that the dual damascene costs have been optimized for front-end processing, so there is little chance of cost reduction there; whereas cost of backside bump could be lowered by replacing polymer dielectric with oxide, and the cost of TSV formation can be addressed by increasing etch rate, ECD (plating) rate, and increasing PVD step coverage.
Since one can produce ~286 200mm2 die on a 300mm wafer, at $575 (his midpoint cost) per wafer, this results in a $2 200mm2 silicon interposer."
HBM stacks can be costly due to manufacturing difficulties. You have to assemble 5 die at a minimum to very small tolerances. Thousands of micro-bumps at 55 micrometer spacing. Errors at any one stage wastes all 5 die. This is what I believe is restraining quicker adoption.
Not really, you are increasing latency CYCLES generally but latency TIME may have a different trend. I.e. if you increase ram speed from DDR2000 to DDR4000, with CAS latency from 12 to 20 cycles, CAS latency time will go down around 20%. Of course CAS is not the only component of total latency, but the reasoning goes on..Again everything will not be equal, increasing clocks will come at the expense of increasing latency. That's how RAM is designed.
RX Vega has been teased and announced years before TR, and yet we don't know anything about it 7 days before launch. TR is announced like 3 months ago? and yet we have videos of Lisa Su herself testing TR against i9 7900X in Cinebench, with hard numbers and open testing policy. NONE of that happened with RX Vega. Or even Vega FE, which had another blind test vs the TitanXp to obscure it's bad gaming performance. All of this doesn't inspire any confidence in the RX Vega.
Excellent write up! Would have saved me time going through the JEDEC spec myselfThis latency vs clock speed "debate" is one of the things that really annoys me about this forum. There is a severe lack of technical understanding. Not only that there is this intellectual laziness that causes people to argue inane points. People tend to argue in generalizations when they could just go read how a given technology works. You don't need to guess with technology. Engineers already built it. It works the way it does. You can read about exactly how it works and then compare an contrast those SPECIFIC design choices. Access to JEDEC specs is free with sign up.
Anyways, READ this, UNDERSTAND this, and then start talking about HBM2. It's a good HBM2 primer without having to go to the spec sheets.
http://monitorinsider.com/HBM.html
So people had an interesting discussion with an AMD official at PDXLAN, the guy compared to a 500$ FreeSync monitor to 950$ GSync one, alluding that when you can't tell the difference, the freesync option wins. Does that mean RX Vega liquid will cost 650$?
https://www.youtube.com/watch?v=l35VxvWyCyE
So people had an interesting discussion with an AMD official at PDXLAN, the guy compared to a 500$ FreeSync monitor to 950$ GSync one, alluding that when you can't tell the difference, the freesync option wins. Does that mean RX Vega liquid will cost 650$?
https://www.youtube.com/watch?v=l35VxvWyCyE
So it will barley beat a Vega FE at gaming and is priced high. What other reason could there be to even bring up the "total cost" debate. Only reason is your product (GPU) doesn't offer great performance/$ and performance/watt compared to competitor.
Interesting thanks, do you know of- or have information- of the latency of hbm 2 in comparison to ddr4 (3200?).This latency vs clock speed "debate" is one of the things that really annoys me about this forum. There is a severe lack of technical understanding. Not only that there is this intellectual laziness that causes people to argue inane points. People tend to argue in generalizations when they could just go read how a given technology works. You don't need to guess with technology. Engineers already built it. It works the way it does. You can read about exactly how it works and then compare an contrast those SPECIFIC design choices. Access to JEDEC specs is free with sign up.
Anyways, READ this, UNDERSTAND this, and then start talking about HBM2. It's a good HBM2 primer without having to go to the spec sheets.
http://monitorinsider.com/HBM.html
Interesting thanks, do you know of- or have information- of the latency of hbm 2 in comparison to ddr4 (3200?).
Cant find any actual latency charts for hbm memory, would be interesting.
That's not what they're saying. It's basically that within a frame limited setup of freesync/g-sync where it might be say 60-100fps, they're going to perform the same. Without that frame limit, the 1080 may perhaps be 130fps vs possibly 110fps on the Vega. But since we're locking into the adaptive synch environment, they perceptively perform the same.This boils down to, if AMD charges the same amount or slightly lower than the 1080 for people that have the 'normal' monitors, AMD is saying it will cost an additional $300+ or so, before you will get the same results as the "blind" test.
Maybe there aren't any cards yetnot enough information leaking out on these cards... wth China do your job
In your example, how do you know that it is a 20 Hz delta?That's not what they're saying. It's basically that within a frame limited setup of freesync/g-sync where it might be say 60-100fps, they're going to perform the same. Without that frame limit, the 1080 may perhaps be 130fps vs possibly 110fps on the Vega. But since we're locking into the adaptive synch environment, they perceptively perform the same.
Even within the refresh range, there may be 20fps difference between the 2 GPUs in a certain title, but you're not going to notice the difference.
So all things being perceptively equal given the dynamic refresh capabilities of these monitors, you're going to need to spend $300 additional on the g-sync to get the same experience.
In your example, how do you know that it is a 20 Hz delta?
FreeSync has a dynamic refresh rate range of 9–240Hz, so, if Vega was pushing at the low end of the scale, and the 1080 was at the high end side of the Gsync scale, the 'blind' test will still show smooth gameplay between them, but, there will be a huge difference for normal monitor users.
*edit, only 7 days more of speculation!
You don't, that's the point. AMD are trying to show that it doesn't matter if that 20Hz delta exists in the freesync/g-sync environment.In your example, how do you know that it is a 20 Hz delta?
Freesync technology might have that range but the monitors don't. For instance the Asus mx34vq 100hz display has a 48-100Hz range. Still quite a big range but for example 75hz Freesync displays are usually 40-75Hz range.FreeSync has a dynamic refresh rate range of 9–240Hz, so, if Vega was pushing at the low end of the scale, and the 1080 was at the high end side of the Gsync scale, the 'blind' test will still show smooth gameplay between them, but, there will be a huge difference for normal monitor users.
Awesome information thanks.I have you covered. You do have a little work to do to put it all together. Again, JEDEC sign up is free. The specs aren't terrible to read if you know what you are looking for.
SK Hynix Hot Chips presentation. Page 14: Comparison of HBM and other DRAMs.
https://doc.xdevs.com/doc/Memory/HBM/Hynix/HC26.11.310-HBM-Bandwidth-Kim-Hynix-Hot Chips HBM 2014 v7.pdf
Micro Hot Chips Page 9
https://www.hotchips.org/wp-content...-Memory-Horizon-Pawlowski-microni-v1-t1-2.pdf
Didn't know if you saw the embedded GDDR5x deep dive link. It's good stuff also.
http://monitorinsider.com/GDDR5X.html#a-few-basics
This latency vs clock speed "debate" is one of the things that really annoys me about this forum.
As far as the marketing goes it really appears that they are two different companies doing the marketing. It might just be because they are operating from an inferior position with Vega.