Overclock.net › Forums › Industry News › Rumors and Unconfirmed Articles › [Videocardz] AMD Vega with 64 Compute Units spotted
New Posts  All Forums:Forum Nav:

[Videocardz] AMD Vega with 64 Compute Units spotted - Page 45

post #441 of 566
Quote:
Originally Posted by xzamples View Post

From MI25's TFLOP rating we can say it will be at clocked around 1525 MHz, is this safe to say?
That's what many have figured out, though it's a 300W TDP server card. Still, the server stuff is usually fairly conservatively clocked and power rated so 1550MHz+ boost clocks with Vega wouldn't surprise me, especially with Polaris already reaching 1400MHz+ on AIB cards out of the box.

Even at 1500 MHz full Vega should be much closer to 1080 ti than 1080 unless a game is really poorly optimised for AMD. At 1600MHz the performance should be close to equal.
post #442 of 566
Quote:
Originally Posted by Tojara View Post

That's what many have figured out, though it's a 300W TDP server card. Still, the server stuff is usually fairly conservatively clocked and power rated so 1550MHz+ boost clocks with Vega wouldn't surprise me, especially with Polaris already reaching 1400MHz+ on AIB cards out of the box.

Even at 1500 MHz full Vega should be much closer to 1080 ti than 1080 unless a game is really poorly optimised for AMD. At 1600MHz the performance should be close to equal.

wait what? 300w server card? can you link me any information please.
post #443 of 566
Quote:
Originally Posted by xzamples View Post

From MI25's TFLOP rating we can say it will be at clocked around 1525 MHz, is this safe to say?

At least 1525MHz probably based on the Videocardz estimation. If you look at workstation cards they generally aren't as aggressively clocked in comparison to consumer ones with aftermarket coolers.

For example,
Radeon Pro WX 7100 (RX 480) = 1188 MHz and ​1243 MHz boost , 7 Gbps VRAM (8GHz)
Radeon Pro WX 5100 (RX470D) = ​713 MHz and 1086 MHz boost, ​5 Gbps VRAM (6.6GHz)
AMD FirePro W9100 (R9 290X) = 930MHz , 5GHz GDDR5
AMD FirePro W8100 (R9 290) = 824MHz , 5GHz GDDR5
AMD FirePro W7100 (R9 285) = 920MHz , 5GHz GDDR5
AMD FirePro W9000 (HD7970) = 975MHz , 5.5GHz GDDR5
AMD FirePro W8000 (HD7950) = 900MHz , 5.5GHz GDDR5
Edited by AlphaC - 4/16/17 at 12:15pm
Workstation stuff
(407 photos)
SpecViewperf 12.0.1
(117 photos)
PGA 1331
(13 items)
 
CPUMotherboardGraphicsRAM
AMD Zen SR7 octocore (Ryzen 7 1700) Overclockable AM4 motherboard X370 To be determined , AMD Vega? 2x8GB DDR4 low-profile or heatsink-less 
Hard DriveHard DriveCoolingCooling
Samsung 950 Pro / 960 Evo / 960 Pro 256GB or 51... Samsung 850 Evo 1TB SSD Storage Black or black+white Twin tower air cooler or s... EK Vardar F2-140 140mm, Phanteks PH-F140SP 140m... 
CoolingOSMonitorPower
Fractal Design Dynamic GP14 (included with case) Win 10 Pro 64 bit 4K monitor with Freesync EVGA Supernova G3/P2 750W or 850W 
Case
Fractal Design Define R5 Blackout edition 
  hide details  
Reply
Workstation stuff
(407 photos)
SpecViewperf 12.0.1
(117 photos)
PGA 1331
(13 items)
 
CPUMotherboardGraphicsRAM
AMD Zen SR7 octocore (Ryzen 7 1700) Overclockable AM4 motherboard X370 To be determined , AMD Vega? 2x8GB DDR4 low-profile or heatsink-less 
Hard DriveHard DriveCoolingCooling
Samsung 950 Pro / 960 Evo / 960 Pro 256GB or 51... Samsung 850 Evo 1TB SSD Storage Black or black+white Twin tower air cooler or s... EK Vardar F2-140 140mm, Phanteks PH-F140SP 140m... 
CoolingOSMonitorPower
Fractal Design Dynamic GP14 (included with case) Win 10 Pro 64 bit 4K monitor with Freesync EVGA Supernova G3/P2 750W or 850W 
Case
Fractal Design Define R5 Blackout edition 
  hide details  
Reply
post #444 of 566
Quote:
Originally Posted by caswow View Post

wait what? 300w server card? can you link me any information please.
https://videocardz.com/64677/amd-announces-first-vega-accelerator-radeon-instinct-mi25-for-deep-learning
They should be out in H1 2017, so similar timeframe to Vega release.
Quote:
Originally Posted by AlphaC View Post

At least 1525MHz probably based on the Videocardz estimation. If you look at workstation cards they generally aren't as aggressively clocked in comparison to consumer ones with aftermarket coolers.

For example,
Radeon Pro WX 7100 (RX 480) = 1188 MHz and ​1243 MHz boost , 7 Gbps VRAM (8GHz)
Radeon Pro WX 5100 (RX470D) = ​713 MHz and 1086 MHz boost, ​5 Gbps VRAM (6.6GHz)
AMD FirePro W9100 (R9 290X) = 930MHz , 5GHz GDDR5
AMD FirePro W8100 (R9 290) = 824MHz , 5GHz GDDR5
AMD FirePro W7100 (R9 285) = 920MHz , 5GHz GDDR5
AMD FirePro W9000 (HD7970) = 975MHz , 5.5GHz GDDR5
AMD FirePro W8000 (HD7950) = 900MHz , 5.5GHz GDDR5
And as a sidenote, while the W9000 and W8000 are clocked higher than the 7970 and 7950 they were release almost an year after those, and they come at a higher TDP.
post #445 of 566
Quote:
Originally Posted by Lee Patekar View Post

Anyone who buys something at inflated prices will feel compelled to justify the cost. Basic human nature and a major part of the console vs pc never-ending arguments.
Yes because Maxwell was such a leap over Kepler, and Pascal is such a leap over Maxwell. Perhaps Volta is truly a new architecture instead of a refresh with minor improvements. We'll see.

You didn't answer the original argument to your post. That being you have to criticize AMD if your going to criticize Nvidia about the pricing increase that generation.

The basis of the argument is AMD also increase the price of their chip the same way as Nvidia did and they fired the first shot. This is true. You cannot make your original statement without dinging AMD. The 7970 was not that much bigger than a gtx 680. And it wasn't the only high priced chip relative to the die size. Let me remind you AMD charged $350 for a 211mm2 pitcairns chip in the 7870. That's smaller than polaris!!! That makes today's pricing look angelic beside the titan series.

That is more expensive than any chip today when you consider how small of a performance jump it was relative to the 6970 and gtx 570 and how small the chip was.

There was no justification for the price increase. The 7870 was something like 5-10% faster than a gtx 570, a card that launched at $350. There was no shift in the pricing dynamics and you didn't see the extreme fall of pricing. This just showed how expensive AMD launched this series.





The second image is from AMD BTW.

And you can't justify it with the increase in cost with 28nm wafers. There was still a big drop in cost per transistor. Compare this to the cost per transistor at 14/16nm(no savings). What the same cost per transistor shows is that per the die area, not taking yields into account, when transistor density doubles, cost per area also doubled. Hence if there was any time for cost per die to increase, it's this one. Not 28nm.

Taking this into account, how can you explain polaris can be sold for 200 to 240, but the 7870 was selling for 350 dollars on a much cheaper node. It was an overpriced chip and add in this was an AMD card and people just didn't want to get it.

What made this move both incredibly stupid compared to what Nvidia is doing now was Nvidia was on the verge on launching their own cards against AMD 7xxx series. This means price to high and people will wait and people did. Nvidia launched cards with better performance, cheaper and were more efficient. Add in Nvidia still had brand value at the time and AMD lost all it's good will it built up during the 4xxx and 5xxx and 6xxx generation. AMD marketshare collapsed.

With Nvidia this generation, AMD essentially said it was not releasing cards in the high end. So with this knowledge in your hand and knowing you won't have competition for a year, what would you do when you consider your the brand leader in the marketplace?

While the founders edition was a despicable move, it wasn't much different than the 7970 and 7870 pricing. And there was still options for cheaper variants 1080/1070. Considering the lack of competition, the none FE editions pricing would be actually generous. The 1070 in particular. The 1070 brought 650 dollar card performance down to $379. Something only the r9 290 managed last generation. The 1070 was probably the card that made AMD price their rx480 pricing the way they did and why we are not getting much of a price drop with the rx580.
Edited by tajoh111 - 4/17/17 at 1:29am
post #446 of 566
Looking at that videocardz benchmark vega kick titan pascal pretty hard in DeepBench benchmark. Possible to guess gaming performance from here, or is totally different workload?
post #447 of 566
Quote:
Originally Posted by Caradine View Post

AMD always did it to a degree that's justifiable by lower yields when the new manufacturing process is still young,

Oh so you are saying that when the processing manufacturer is young (like the brand new 14/16nm manufacturing process), and when AMD are doing it, it is absolutely fine, but nvidia are doing it, oh hell no they are the devil, right..

Sure. I'm convinced...
Quote:
Originally Posted by Lee Patekar View Post

Yes because Maxwell was such a leap over Kepler, and Pascal is such a leap over Maxwell. Perhaps Volta is truly a new architecture instead of a refresh with minor improvements. We'll see.

I see.
So you are saying that 40% increase in performance from maxwell to pascal is a rebrand and a minor improvement?
I guess that is why you are still rocking the 7970 from AMD. Because its performance is still very much on par with the mid range cards by being maybe 20% of their performance as of today.
Main system
(16 items)
 
Editing PC
(8 items)
 
 
CPUGraphicsGraphicsRAM
E5-1680v2 AMD FirePro D700 AMD FirePro D700 64GB 1866mhz 
Hard DriveOSMonitorCase
1TB PCIE SSD OSX 10.10.x Dell U2713H Mac Pro 
  hide details  
Reply
Main system
(16 items)
 
Editing PC
(8 items)
 
 
CPUGraphicsGraphicsRAM
E5-1680v2 AMD FirePro D700 AMD FirePro D700 64GB 1866mhz 
Hard DriveOSMonitorCase
1TB PCIE SSD OSX 10.10.x Dell U2713H Mac Pro 
  hide details  
Reply
post #448 of 566
Quote:
Originally Posted by Defoler View Post

I see.
So you are saying that 40% increase in performance from maxwell to pascal is a rebrand and a minor improvement?
I guess that is why you are still rocking the 7970 from AMD. Because its performance is still very much on par with the mid range cards by being maybe 20% of their performance as of today.

Forgetting the process node are we?
post #449 of 566
Quote:
Originally Posted by tajoh111 View Post

You didn't answer the original argument to your post. That being you have to criticize AMD if your going to criticize Nvidia about the pricing increase that generation.

Its an elegant post you wrote which, surprisingly agrees with what I've been saying. Why did AMD initially price the 7970 and 7870 so high? For the same reason nVidia is pricing their cards so high today. Lack of competition at the time. AMD was first to market with the fastest flagship and they tried to capitalize on it. nVidia eventually responded with their 680 GTX, which was a mid-sized die like the 7970 and it outperformed the 7970 on launch. They priced it according to the competition and held back their 500+ mm^2 die. Traditionally the mid-sized die would be the xx70 card and the 500+ mm^2 the xx80 card, but things changed that day. AMD could no longer compete.

The issue you're having, and why you think you're disagreeing with me now when you're not, is that I'm not loyal to any one brand. I buy the best bang for the bucks that meets my needs. I skipped the 7970 on launch, it was too pricey. I waited for nVidia to respond and bring some competition. They did, partially. Unfortunately AMD had nothing else than the 7970.. so they overclocked it and re-branded it.. and nvidia withheld their 500+ mm^2 die. I eventually picked up the 7970 when the price came down, and that card held up the test of time quite well. (it outperforms the 680 today).

The high prices aren't due to the process. Especially in the later years once 28 nm was mature, that argument falls flat on its face. Its lack of competition and nothing else.

And yes, it works both ways.
post #450 of 566
Quote:
Originally Posted by Defoler View Post

Oh so you are saying that when the processing manufacturer is young (like the brand new 14/16nm manufacturing process), and when AMD are doing it, it is absolutely fine, but nvidia are doing it, oh hell no they are the devil, right..

Sure. I'm convinced...
If you not only read what I said but also understood the words, you wouldn't be saying things like "when AMD are doing it, it is absolutely fine", because you would find out that what I said was that even though AMD and nvidia are doing the same thing on a new process (raising the price per mm^2 die size), the thing that is fine to me is that AMD is doing it to a ~3x lesser degree.

You can think about it like this:
AMD: 1 unit of overpricification
nvidia: 3 units of overpricification
I'm just talking about the two additional ones.

techpowerup has all the die sizes and prices laid out for you so you don't need to be convinced by me.
It was so significant that nvidia had to create the "titan" just so that it doesn't seem like GTX X80 suddenly got 3 times more expensive. And guess what, Pascal improves on the overpricification with the fact that the titan isn't even the biggest chip.
Quote:
Originally Posted by Defoler View Post

So you are saying that 40% increase in performance from maxwell to pascal is a rebrand and a minor improvement?
I guess that is why you are still rocking the 7970 from AMD. Because its performance is still very much on par with the mid range cards by being maybe 20% of their performance as of today.
And here you need to understand what was being talked about before that statement was made - R&D, architectural improvements? yes.
If AMD took VLIW5, redid it on 14nm, and threw more shaders at the problem it would also beat the 5870.
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Rumors and Unconfirmed Articles
Overclock.net › Forums › Industry News › Rumors and Unconfirmed Articles › [Videocardz] AMD Vega with 64 Compute Units spotted