AMD Unveils More Info About Vega

[DocSwag]

56 minutes ago, Kloaked said:

Went from Graphics Core Next to Next-gen Compute Unit.

 

These names, man.

GCN is the architecture as a whole. I'm pretty sure Vega is still based on GCN, it's just that they redesigned the CUs a bit and are now calling the CUs themselves Next-gen Compute Units. The GPU as a whole is still based on GCN.

[MageTank]

1 minute ago, M.Yurizaki said:

According to http://www.anandtech.com/show/6993/intel-iris-pro-5200-graphics-review-core-i74950hq-tested/3, the eDRAM on Iris Pro on Haswell is 50GB/s one way (so 100GB/sec aggregate). The fastest RAM I see on Newegg is about 34GB/sec. o3o

 

 

They say that, but in Aida64's test, the L4 cache only scored the following result:

 

Mine for example, scores this:

 

Perhaps Aida64 can't properly test the L4, I don't know, but using this test, mine seems to beat it in every regard. 

[cj09beira]

15 minutes ago, TheRandomness said:

I vote for it going to have 4096 SPs or more, judging by the core die size appearing a bit smaller (or the same size) than the one on the Fury X combined with the 14nm process. If it has 4480 SPs, then I think it's going to be packing a punch running at around ~1100-1200Mhz

for it to achieve the 12.5 TFPS of fp32 it would need to run at ~1400mhz, if its 4096 it needs ~1525mhz

[Mira Yurizaki]

13 minutes ago, MageTank said:

They say that, but in Aida64's test, the L4 cache only scored the following result:

 

Mine for example, scores this:

 

Perhaps Aida64 can't properly test the L4, I don't know, but using this test, mine seems to beat it in every regard. 

Ah right, dual channeling would double the bandwidth.

 

EDIT: I stewed on it a little more. Anandtech said the eDRAM has 50GB/sec read and write, and when you aggregate them it combines into 100GB/sec total. If AIDA64 is only reading, then writing, then it may not be a good representative of what that memory is capable of. For example, your RAM may only have 50GB/sec of total bandwidth. If AIDA64 is reading, then writing, it's using all of that bandwidth for just that one task. Since L4 has dedicated read and write channels, it would only be using one or the other, not both at the same time.

[TheRandomness]

2 minutes ago, cj09beira said:

for it to achieve the 12.5 TFPS of fp32 it would need to run at ~1400mhz, if its 4096 it needs ~1525mhz

Oh. So that'd place it at 5208 SPs for 1200 MHz and 5682 for 1100 MHz.. Hmmm, technically that could be possible if the die size is the same because of the manufacturing process, but it's more than likely it'll be running somewhere near 1500MHz then for that number.

[MageTank]

52 minutes ago, M.Yurizaki said:

Ah right, dual channeling would double the bandwidth.

 

EDIT: I stewed on it a little more. Anandtech said the eDRAM has 50GB/sec read and write, and when you aggregate them it combines into 100GB/sec total. If AIDA64 is only reading, then writing, then it may not be a good representative of what that memory is capable of. For example, your RAM may only have 50GB/sec of total bandwidth. If AIDA64 is reading, then writing, it's using all of that bandwidth for just that one task. Since L4 has dedicated read and write channels, it would only be using one or the other, not both at the same time.

Isn't read/write aggregation the same as rank interleaving? How does the cache aggregation work?

[Mira Yurizaki]

3 minutes ago, MageTank said:

Isn't read/write aggregation the same as rank interleaving? How does the cache aggregation work?

It's a communication thing. I believe RAM is half-duplex, where as the eDRAM is full-duplex.

[KrumpetPirate]

2 hours ago, lots of unexplainable lag said:

Problem here is that the 1070/1080 have been on the market for months now. People who waited for a Vega card have most likely bought a 1070/1080 right now. Hyping things up and then staying silent will only lose you customers. By the time Vega hits the stores Pascal will have been out for more than a year at this rate. There's no way Vega is going to win the market share back, not with nVidia also having Pascal refreshes planned.

 

AMD's move with the RX480 launching first might have been a smart one on paper, but the money's made in the high-end market, not the mid-range market. AMD has nothing in the high-end save for a last-gen power hungry Fury or Fury X on sale.

My question is whether the improved profit margins for high end cards gets swallowed by mid range numbers. I have never seen anything to indicate that this is the case however.

[MageTank]

40 minutes ago, M.Yurizaki said:

It's a communication thing. I believe RAM is half-duplex, where as the eDRAM is full-duplex.

Yeah, ram is half-duplex. That's why I asked about Rank Interleaving. It "simulates" full-duplex in the sense that you can read from one rank while still writing to the other, but it's not real full-duplex in the sense that it can read/write both ways at the same time. I seriously need to take the time to study this, and I am now curious, lol. The biggest bottleneck in this scenario would be the IMC and the way it handles banks. It can only open one bank per rank at a time. So technically, bandwidth won't improve from this scenario, but latency will. 

 

It's a shame Skylake changed the way the eDRAM worked, as it's now a memory buffer, rather than L4 victim cache. Oh well, this at least gives me an excuse to find a 5775C on sale and run some tests. 

[Mira Yurizaki]

23 minutes ago, KrumpetPirate said:

My question is whether the improved profit margins for high end cards gets swallowed by mid range numbers. I have never seen anything to indicate that this is the case however.

Maybe adding to your post, but if the Steam Hardware Survey is anything to go by, the top 10 GPU are all midrange, if you consider the GTX X70 chips midrange. The first GPU you could consider high-end for its time is the HD 7900, which is 20th in line.

 

So it leads me to believe that midrange cards do most the profits for either company, not high-end cards. Especially when you consider midrange cards are typically high-end cards that failed QC.

[RagnarokDel]

2 hours ago, cj09beira said:

its 2 stacks of hbm2 512GB/s

I'm pretty sure it's 1024GB/s?

[DocSwag]

7 minutes ago, RagnarokDel said:

I'm pretty sure it's 1024GB/s?

That's if you have 4 stacks.

[Inheritance]

 If AMD's top of the line GPU beats my 1070 im going to throw a brick at someone. And by  brick I mean my 1070 brick.

[cj09beira]

6 minutes ago, Zheren159 said:

 If AMD's top of the line GPU beats my 1070 im going to throw a brick at someone. And by  brick I mean my 1070 brick.

i volunteer as a target :-P

[Inheritance]

4 minutes ago, cj09beira said:

i volunteer as a target :-P

Careful its a MSI Seahawk so included in the brick might be another brick attached by wires.

[cj09beira]

1 minute ago, Zheren159 said:

Careful its a MSI Seahawk so included in the brick might be another brick attached by wires.

i would still do it :-p

[TheFlyingSquirrel]

3 hours ago, lots of unexplainable lag said:

Problem here is that the 1070/1080 have been on the market for months now. People who waited for a Vega card have most likely bought a 1070/1080 right now. Hyping things up and then staying silent will only lose you customers. By the time Vega hits the stores Pascal will have been out for more than a year at this rate. There's no way Vega is going to win the market share back, not with nVidia also having Pascal refreshes planned.

 

AMD's move with the RX480 launching first might have been a smart one on paper, but the money's made in the high-end market, not the mid-range market. AMD has nothing in the high-end save for a last-gen power hungry Fury or Fury X on sale.

You assume everyone is a hardcore enthusiast on the same upgrade cycle and has £379+ (UK pricing) to drop on a card.  If AMD position a great 1440p card in the £300-330 void, it is possible to beat the price gourged 1070 in long term sales.

[Doobeedoo]

So Vega on DOOM demo probably used old unoptimized drivers as it was stated before by community? But still it was capped on 60 too.

[raphidy]

50 minutes ago, Zheren159 said:

 If AMD's top of the line GPU beats my 1070 im going to throw a brick at someone. And by  brick I mean my 1070 brick.

It will, 1 year later and will probably cost more than your 1070.

[MageTank]

33 minutes ago, kidanime3d said:

You assume everyone is a hardcore enthusiast on the same upgrade cycle and has £379+ (UK pricing) to drop on a card.  If AMD position a great 1440p card in the £300-330 void, it is possible to beat the price gourged 1070 in long term sales.

If they can release a card that is comparable to the 1070, around the $300-$350 price range, it will be a pretty big win. Especially since you can opt for a 1080p 144hz Freesync monitor for $200, and have a $500-$550 setup that offers a better experience than a single high end GPU for $550+. The card would easily handle high refresh rate 1080p gaming for years to come too, so longevity would be great. 

[Citadelen]

4 hours ago, MageTank said:

-snip-

I seem to recall that the Vega chip demoed was Vega 10, the smaller chip. If it beats a 1080, which it appears to do, then with the major overhaul of the architecture in mind, I very much expect AMD to match Nvidia in the extreme high end, something which I would recently deemed extremely unlikely.

[dalekphalm]

5 hours ago, TheRandomness said:

Oh here's a new thing I'd like to see that AMD could give without revealing too much. Is it going to be stack like:

RX 485 (small cut down Vega), RX 490 (small Vega), RX 495 (big cut down-ish Vega), RX 500 (big Vega)? Or will it skip straight to the 500 series?

To me that would make no sense, especially calling big Vega RX 500 while the rest RX 4xx - they need to either keep everything within the RX 4xx stream, or just switch right to RX 5xx. It also depends on how many chips are gonna be released. We can assume there's Big Vega and Small Vega, but is that it?

 

And we can also assume that there will be cut down variants of both chips, giving us at minimum, 4 distinct model numbers. 4 is hardly enough to fill out an entire generation, so either there are more chips, or they'll be rebranding or refining Polaris to fill out the low end (Which would not really be that bad, if they can keep the power consumption levels to reasonable levels). The only problem I can see with that, is that certain fanboys hate it when AMD re-releases an older card as a lower-end part in a new generational lineup.

[N3v3r3nding_N3wb]

10 hours ago, MageTank said:

If the performance in doom is any inclination on how this thing performs in other titles, this is great. While it won't be enough to answer the Titan XP (or it's rumored cut-down offspring), it will be a definitive answer to the GTX 1080 and 1070. Especially if you compare the results to the previous 4k Vulkan benches:

 

Nvidia: 

 

AMD:

 

That's a solid 28% boost over the Fury X, and 11% over the GTX 1080. Again, this is just in Doom, so results will most definitely vary, but it's still good news nonetheless. We just might see some competition in the high-end before Volta after all. 

And, don't forget, these cards haven't even launched yet.  Drivers are still unpolished and the state of the silicon cannot be guaranteed (although I'd assume that it's a near-release version, since they're showing all this with confidence).  Heck, the 480 has gained 10% since launch.  Pre-launch (possibly months before the launch), the drivers are bound to suck, especially given AMD's aforementioned driver history.

[handymanshandle]

If a discord bot is to be trusted with their 8ball then may, is when we could see a release.

I think when they say later in 1H '17. They probably mean close to Q3 but probably in the middle or near end of Q2.

[N3v3r3nding_N3wb]

5 hours ago, Citadelen said:

I seem to recall that the Vega chip demoed was Vega 10, the smaller chip. If it beats a 1080, which it appears to do, then with the major overhaul of the architecture in mind, I very much expect AMD to match Nvidia in the extreme high end, something which I would recently deemed extremely unlikely.

The numbers (Vega 11 vs. 12) denote the order in which they were conceived, not higher number equals higher performance. I believe that Vega 10 is the bigger one. 

 

I know that this source information-wise is basically irrelevant, but iirc the naming scheme has been confirmed for awhile, so the names it refers to and uses should be accurate. 

https://www.google.com/amp/www.techpowerup.com/226012/amd-vega-10-vega-20-and-vega-11-gpus-detailed%3Famp?client=ms-android-verizon