Radeon Vega Frontier Edition LCS and 4096 Shader Processors

#5432737

Denial

2017-05-17 15:37

Huh? I'm confused. How is HBM1 faster than HBM2? R9 Fury X = HBM1 512 GB/s whereas Vega frontier = HBM2 480 GB/s?????

Four stacks vs two stacks.

#5432745

slyphnier

2017-05-17 16:08

not things to concern but those "R" on the corner led is for Radeon ? anyone agree they should just go with AMD logo or Vega logo or even just let it blank ? imo R letter on the led feel so ackward

#5432757

labidas

2017-05-17 16:40

So at same clock fast as fury analogue to 9xx and 10xx nvidia series nothing revolutionary except H265 encoder and less power.

so, the same as pascal v maxwell. nothing revolutionary... :no:

#5432759

isidore

2017-05-17 16:43

Interesting. Almost same specs as the FuryX at twice the speed 😀. Can't wait to see this baby in cation. If the consumer end Vega Nova is sold at 599 and it will at the 1080Ti level, **** it's a bargain. Only thing AMD has lost is time. Releasing this at the 1080Ti launch date would have crippled nvidia's sales. But that's AMD people.

#5432768

LiviuTM

2017-05-17 17:17

Huh? I'm confused. How is HBM1 faster than HBM2? R9 Fury X = HBM1 512 GB/s whereas Vega frontier = HBM2 480 GB/s?????

HBM1 used with Fury had a 4096-bit wide memory interface. HBM2 used with Vega hasa 2048-bit interface; with a similar 4096-bit interface it should pull a respectable 960 GB/s, almost double compared to HBM1. There maybe also other under the hood improvements, such as reduced latency. Cheers

#5432773

holler

2017-05-17 17:31

So they still havn't figured out the Heat problem, seeing it needs Watercooling 😛uke2:

uh no, they released two card actually. one is air cooled the other the same card with water cooling, so you have the option but its not neccessary. Fury had air cooling too so your comment is completely inaccurate.

#5432782

xendrome

2017-05-17 17:44

I cringe looking at the comments with people thinking this is a consumer side/gaming card..

#5432783

pharma

2017-05-17 17:45

uh no, they released two card actually. one is air cooled the other the same card with water cooling, so you have the option but its not neccessary. Fury had air cooling too so your comment is completely inaccurate.

Two cards - one with 8-pin + 8-pin, one with 8-pin + 6-pin.

#5432791

LiviuTM

2017-05-17 18:30

I cringe looking at the comments with people thinking this is a consumer side/gaming card..

😀

#5432822

jortego128

2017-05-17 19:32

---"""

Interesting. Almost same specs as the FuryX at twice the speed 😀. Can't wait to see this baby in cation. If the consumer end Vega Nova is sold at 599 and it will at the 1080Ti level, **** it's a bargain. Only thing AMD has lost is time. Releasing this at the 1080Ti launch date would have crippled nvidia's sales. But that's AMD people.

---""" Only its not even close to 2x speed. FuryX= 1050MHz, Vega=1600Mhz I hope they made more changes than just clock speed, cause 50% upclock alone wont cut it against 1080(Ti)...

#5432825

schilperoordbas

2017-05-17 19:34

What a sexy card

#5432860

Fox2232

2017-05-17 20:16

13 tflops it has to run 1550 or 1600 for that one

Shader cores for math have been redesigned so, till someone confirms how much, one may not be 100% sure about calculations per clock per SP. And there is another leak stating 90 GPixel/s and with 64 ROPs it stands as 1.4GHz. (It may be wrong, but I would not overestimate my guesses here. It may be able to do 1.6GHz as shown in Sandra or may not.)

#5432865

PrMinisterGR

2017-05-17 20:30

I don't understand while we still are in doubt about Vega's clocks. Whatever they will be in the final version for consumers, the chip gets at least 1.6GHz.

#5432870

Fox2232

2017-05-17 20:37

---"""---""" Only its not even close to 2x speed. FuryX= 1050MHz, Vega=1600Mhz I hope they made more changes than just clock speed, cause 50% upclock alone wont cut it against 1080(Ti)...

	Fiji	Polaris 10	Diff [%]
SP	4096	2304		-44	
TMU	256	144		-44
ROP	64	32		-50
MemBand	512	256		-50%
Clock	1050	1266		+21

That means for Polaris 10: Pixel Fillrate 40% lower. Texture Fillrate 32% lower. Compute power 32% lower. And it still sits just a bit under Fiji and in certain games it touches it. That's because Polaris have seen some changes, And Vega changes... Since you do not look like you know changes in Polaris, yet alone Vega... Look at it from Polaris 10 start point. Vega has almost everything doubled in comparison to Polaris 10 and on top of that there is Clock uplift and uArch improvements.

#5432877

Crazy Serb

2017-05-17 21:00

I hope these cards will improve reliability of wooden PCs loading online games within last couple seconds before disconnect. Full wooden version looks more reliable to me. Pointless cards otherwise 😉

#5432894

jortego128

2017-05-17 22:01

	Fiji	Polaris 10	Diff [%]
SP	4096	2304		-44	
TMU	256	144		-44
ROP	64	32		-50
MemBand	512	256		-50%
Clock	1050	1266		+21
That means for Polaris 10: Pixel Fillrate 40% lower. Texture Fillrate 32% lower. Compute power 32% lower. And it still sits just a bit under Fiji and in certain games it touches it. That's because Polaris have seen some changes, And Vega changes... Since you do not look like you know changes in Polaris, yet alone Vega... Look at it from Polaris 10 start point. Vega has almost everything doubled in comparison to Polaris 10 and on top of that there is Clock uplift and uArch improvements.

I understand Polaris is a completely different uArch. Vega should be as well. That doesn't change the fact that someone posted that Vega looks extremely similar to Fury X with twice the speed. Its not twice the speed, more like 50% more speed (clockrate wise). That is not debatable. My point is, they better hope they picked up about a 50% throughput per clock in games to go along with that 50% uptick in clock if they plan to touch the 1080Ti and TitanXp, because those cards are roughly 100% faster in most games than FuryX. I sincerely hope that happens, but seeing the lack of a proper gaming version of Vega and a severe lack of game demos, I don't think AMD have very good news on that front. I hope I'm wrong...

#5432897

PrMinisterGR

2017-05-17 22:16

I understand Polaris is a completely different uArch. Vega should be as well. That doesn't change the fact that someone posted that Vega looks extremely similar to Fury X with twice the speed. Its not twice the speed, more like 50% more speed (clockrate wise). That is not debatable. My point is, they better hope they picked up about a 50% throughput per clock in games to go along with that 50% uptick in clock if they plan to touch the 1080Ti and TitanXp, because those cards are roughly 100% faster in most games than FuryX. I sincerely hope that happens, but seeing the lack of a proper gaming version of Vega and a severe lack of game demos, I don't think AMD have very good news on that front. I hope I'm wrong...

The only honest-to-god issue that GCN has is memory bus efficiency and triangle performance. Already with Polaris this seems to have improved very much, Vega (according to what AMD tells us) will improve this much further. Just the clocks alone could improve the Fury X enough to go vs the 1080 almost easily I would say. The memory controller/tessellator and tiled-based rasterization will make a big difference.

#5432901

Fox2232

2017-05-17 22:19

I understand Polaris is a completely different uArch. Vega should be as well. That doesn't change the fact that someone posted that Vega looks extremely similar to Fury X with twice the speed. Its not twice the speed, more like 50% more speed (clockrate wise). That is not debatable. My point is, they better hope they picked up about a 50% throughput per clock in games to go along with that 50% uptick in clock if they plan to touch the 1080Ti and TitanXp, because those cards are roughly 100% faster in most games than FuryX. I sincerely hope that happens, but seeing the lack of a proper gaming version of Vega and a severe lack of game demos, I don't think AMD have very good news on that front. I hope I'm wrong...

There's really no need. This Vega introduced to us indeed has very similar specs as Fiji (In rough numbers). But Fiji is 8.9B transistor chip. Mentioned Titan XP/1080 Ti has 12B transistors. That makes it quite more expensive as it increases percentage of defective GPUs per wafer. From Manufacturing perspective, this Vega will cost something in between GTX 1080 and GTX 1080Ti. And if it competes with Ti, then it will be very financially successful. If it competes with GTX 1080, it will be disappointing. But one should just expect it to sit somewhere in between. Edit: Btw. if that card introduced to us has really 1600MHz, then it has only 56ROPs as Pixel Fillrate ~90 Gpixels/sec is official value. (less transistors)

#5432913

pharma

2017-05-17 23:26

I don't understand while we still are in doubt about Vega's clocks. Whatever they will be in the final version for consumers, the chip gets at least 1.6GHz.

Was there a press release about final clocks? The compubench info tab for that unknown AMD card with 16GB states 1600 is the max freq. That's with many variables unknown ... whether sustainable, underwater, suicide run, etc.. Compubench.com info tab states: „CL_DEVICE_MAX_CLOCK_FREQUENCY 1600“

#5432941

PrMinisterGR

2017-05-18 00:53

There are press conferences about FP32 performance and the shader configuration of the part. (4096 * 2) x FREQUENCY = 13TFLOP. Frequency = 1600MHz. It's not rocket science. We don't even have to look at the Linux driver to know that at this point. He said that they have two models, but that they both get ~13Tflop, so that means that the boost (?) clock of 1.6GHz holds for both, with the watercooled part more likely to sustain it for longer.