I must beg to differI am not the one with diarrhea of the mouth.
All we've seen of Vega is an ES card with 6+8-pin power running ~10% faster than a 1080 (non-Ti) in a single game that's known to be pretty friendly to AMD's architectures. You must then be assuming either that the ES is massively over-engineered in terms of power delivery, that high-end Vega will need more than 300W of power, or that they somehow will manage to cut power draw by 30-50% or more between Engineering Samples and production cards.
That's possible. The parallel you're indicating between Ryzen ES clocks and Vega ones is false though - raw GHz numbers don't matter in this context, but relative clock difference does. After all, maximum possible clocks are dependent on architecture, even on the same process - otherwise, we'd have 4GHz GPUs by now. Ryzen ES went from ~2.8-3.4GHz (depending on which leaks were real and which weren't) to 4.0GHz two-core turbo (and 3.7 ACT). Depending on the ES in question, that's a 42-17% increase in clock speeds. The 3.4 leaks (and official benchmarks) happened a few months before launch, so it's reasonable to believe they'd had those kinds of speeds running for at least a little while internally, with 2.8 probably being a few months earlier. Going from a 1000MHz Vega ES, given a similar time span before their respectie launches for the 2.8 Zen leaks and 1GHz Vega numbers, let's call that comparable. From that alone, you could extrapolate Vega boost clocks of around 1500MHz, sure. The only problem is, extrapolation like this is pretty much impossible, as clock speed scaling between architectures is essentially incomparable, and we don't have any hard numbers that we're actually sure of to begin with. Nor do we know anything about the voltage scaling of Vega, or really anything else. As such, while not impossible and possibly not even improbable, it's too far into the realm of conjecture for me to be comfortable with, at least.Third option:
- The printed circuit board (PCB) is designed for a 225w card.
- Official early showcase had the engineering sample GPU running at lowish clocks, i would guesstimate/hope ~1000mhz.
- More recent leaks show ES cards running at 1200mhz.
- Finalized (read non-ES) cards could be running at 1500+mhz once released in one and a half months. (and finally using all the available power (225w))
Above could be just as plausible as your lowballing.
Ps. This should be quite telling for whats going on..
Do you remember what clockspeed the Ryzen engineering samples was running at, compared to what it was boosting to once release.. ?
We ended up almost +1ghz above predicted (at one point) clockspeed.
So based on rumors/leaks so far, the 2018 product stack is likely:
RX Vega (full Vega)
RX Vega Nano (cut Vega)
RX 580 (improved 480)
RX 570 (improved 470)
RX 560 (improved 460)
RX 550 (budget/OEM)
Haha, funny that wasn't even intentional. I meant 2017.I see what you did there....
More or less.Just playing a bit of catch-up here. So we've seen an engineering sample that seemed to beat a GTX1080, but we are not sure whether or not that engineering sample is the highest level Vega card, and therefore we're unsure how the entire product stack will line up against Nvidia's offerings. Am I understanding things right? We are certain that the flagship Vega (that would theoretically compete with the 1080ti) will be released in 2017 right?
More or less.
What we know, and have seen:
What's been leaked, but not verified:
- AMD has a Vega card that is at least roughly 10% faster than a 1080 in SW:BF running from an 8+6-pin power connector, cooled by a standard-looking 2-slot blower
- That the Vega cards will be named RX Vega
- That Vega is a big architectural difference from previous GCN chips
- That the new memory controller, the HBCC, has some intriguing implications for low-memory GPUs and large dataset computing, but is as of now an unknown for high end ganing
What's rumored:
- That there are Vega ES samples in circulation with clock speeds up to 1200MHz
- That 'big Vega' has 64CUs, same as the Fury X (although we don't know what's in these CUs, as AMD calls the Vega CU 'NCU' as it's reportedly quite different from before
- That there will be two Vega dice (codenamed 10 and 11)
- That Vega will catch up to Pascal efficiency-wise
- That Vega will clock to or beyond 1500MHz
That's... intriguing. But seven Polaris 12 SKUs as well? Let's see: full and cut-down(?) for desktop, full and cut-down Pro for MacBook Pros, full and cut-down for mobile. What more can there be? There's definitely not room for more than one cut-down size.https://videocardz.com/67524/amd-linux-drivers-lists-seven-vega-10-ids
There are seven Vega 10 SKUs
According to Phoronix, first 100 patches were released to provide initial support for Vega, based on GFX9 architecture, which is more complex than Polaris (GFX8). New architecture required roughly 40 thousand lines of code, which brought support for “new video BIOS interface, new hardware intellectual property, support for video decode using UVD (UVD 7.0), support for video encode using VCE (VCE 4.0), support for 3D via RadeonSI, power management, full display support using DC, and support for SR-IOV virtualization”.
New patches include seven Vega IDs. For comparison, the same file lists only two Fiji IDs. That being said, we are looking at many new models, including those based on Radeon RX, Radeon PRO and Radeon Instinct. It’s worth noting that the list does not include revisions (C1, C3 etc.).
At the same time, seven Polaris 12 IDs were also included, but there is still no trace of Vega 11.
Vega 10 device support:
{0x1002, 0x6860, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},
{0x1002, 0x6861, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},
{0x1002, 0x6862, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},
{0x1002, 0x6863, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},
{0x1002, 0x6867, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},
{0x1002, 0x686c, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},
{0x1002, 0x687f, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_VEGA10},
Polaris 12 device support:
{0x1002, 0x6980, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_POLARIS12},
{0x1002, 0x6981, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_POLARIS12},
{0x1002, 0x6985, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_POLARIS12},
{0x1002, 0x6986, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_POLARIS12},
{0x1002, 0x6987, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_POLARIS12},
{0x1002, 0x6995, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_POLARIS12},
{0x1002, 0x699F, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_POLARIS12},
AMD hasn't said a lot about it, but isn't there also a smaller Vega chip (Vega 11) that was talked about at one point? I think that most of what they've been talking about is the big chip that's the Fury replacement. Unless that got scrapped, it seems like we'll be in a really weird place as they said that Vega will be called Vega, so what are they going to do in order to differentiate the big and little Vega chips in terms of naming. Unless they drop RX and just go with Vega, so Vega 590, Vega 590, Vega Fury, etc. or something along those lines.
Vega cards are confirmed by AMD to be named "RX Vega". Considering that the RX naming scheme is pretty much brand-new (and has been quite successful), abandoning it would be odd. What we don't know is what further denominations they'll use.I'm expecting the "RX" portion to be dropped, and a naming scheme similar to Ryzen to be used. Just as we have Ryzen R7, R5, and R3, it would not be surprising if we see a naming scheme something like Vega V8, V6, and V4 (the latter reserved for iGPUs).
7 Vega 10 SKUs?
Do they think we're gullible?
Nope, only the Vega 10 SKUs. This is list of 7 variants of Vega 10, only.sounds like a lot, but are Raven Ridge APU parts included in that?
Given that APUs have graphics on-die, no. Even the RR iGPU architecture is identical to Vega 10, it will never identify as such.sounds like a lot, but are Raven Ridge APU parts included in that?
Nope, only the Vega 10 SKUs. This is list of 7 variants of Vega 10, only.
Nope, only the Vega 10 SKUs. This is list of 7 variants of Vega 10, only.
That + two tiers each of Radeon Pro and Radeon Instinct would make 7 SKUs and quite a bit of sense in my head. Seems reasonable.I saw some speculation that some of these are also the professional versions of the cards, so it's unlikely we see 7 actual consumer GPUs come of this. Probably 3 at most if they do something like Fury, Fury X, Fury Nano which seems reasonable.
Or are they simply pushing designs out as quick as they can do them, launch scheduling be damned?
That + two tiers each of Radeon Pro and Radeon Instinct would make 7 SKUs and quite a bit of sense in my head. Seems reasonable.
This just leaves the question of Vega 11: what is it, and where is it. If it's smaller than V10, could it be a mobile-oriented chip, and thus launch later (and thus avoiding the mess of more than two RX Vega SKUs)? Or are they simply pushing designs out as quick as they can do them, launch scheduling be damned? Or is V11 bigger, badder, and simply waiting out its time to pounce on GP102?
This just leaves the question of Vega 11: what is it, and where is it. If it's smaller than V10, could it be a mobile-oriented chip, and thus launch later (and thus avoiding the mess of more than two RX Vega SKUs)? Or are they simply pushing designs out as quick as they can do them, launch scheduling be damned? Or is V11 bigger, badder, and simply waiting out its time to pounce on GP102?