Buildzoid does a breakdown of the “nano” Vega PCB:
Cheers, will take a look. I did play with the config file briefly, but its only mining part time so haven’t spent a lot of time optimising. But thanks again.
definitely helped. i haven’t increased the intensity but definitely doubling the threads i’m getting a lot more.
Vega 20 early benchmark results
I find it quite amusing that the judgement on an Instinct card is about gaming. On what is essentially an Alpha product no less. That HBM speed is interesting though.
Pretty dumb leak; shouldn’t have happened.
they just wanted to show it’s alive. nothing to do with scores, except that they wanted to show it can perform at roughly the same level.
Pre-release engineering hardware gets low score in irrelevant benchmark with pre-release driver!
The drivers for Vega 10 aren’t even really properly functional yet…
I like how they assume the clock speed was “obviously read incorrectly”.
Maybe it does get that performance running at 1Ghz, and it was running at that nominal clockspeed because they haven’t performed any sort of binning process yet.
Hmm. New driver. 18.4.1
Hash rate on 2x Vega 64s for cryptonightv7 dropped from 3800-4k hashes to 660.
Something definitely changed
Ah… crossfire turned itself back on. Forgot that driver updates reset it back to default = crossfire enabled.
Turned it back off and back to normal.
The video is worth a watch in full length but this specifically is relevant here.
Finally got to put a Vega card in my machine today. The Sapphire Vega 56 Nitro that I had ordered in January was just not showing up, so I switched to a PowerColor Red Devil Vega 56 that became available.
It’s a fairly similar beast on the outside with a basically three slot design and a couple of different BIOS level profiles for easier management of the “power envelope”. I picked the quiet option and found that using that as a base there’s still a lot of room in the AMD driver settings to push the power usage from around 220 to 300-ish if you feel like. And the cooler is really effective at staying quiet while keeping the temperature down.
The Sapphire Nitro would have come with an extra support bracket which is a neat idea for such a sizeable card. Now in the picture below one can see that the slouching isn’t too bad as the back plate keeps it fairly straight. And while the motherboard also has one of those strengthened PCIe slots, I noted that there is definitely a possibility to tie a strap to the far right end of the backplate and secure it somewhere around the drive cage.
When it comes to gaming performance for what I can say after testing half a dozen games this evening, this Vega 56 is being held back by my modestly clocked Ryzen 1700 at 3.8 GHz in most games even at 1440p, to a varying degree. On the other hand, compared to the previous R9 Fury card I can now comfortably play with higher settings.
So is it worth the money? Probably not at todays prices. But I needed something better and for all the rumors I don’t see AMD coming out with a more powerful gaming graphics card until next year. And as far as this rig goes I don’t see anything else that really warrants an upgrade for the rest of this year.
With a slight maaybe on the Threadripper 2, but probably not even that.
That’s alright. Just mine with the card while you’re not using the computer so it can slowly begin paying itself back
So, looky here…
WX8200 workstation Vega. Similar price to FE. But only 56 CUs. But with 8 GB (at 2Gbps).
But… SR-IOV support…!
The IO bandwidth and boost clock of 1500mhz should mean this thing performs pretty decently?
I wish someone would test one with it
I don’t think the WX series have SR-IOV support. Only the Radeon Instinct cards have it: https://www.amd.com/en/graphics/servers-radeon-instinct-mi
I don’t know. But what’s the difference between the cards? Just clock speed or CU count?
If it’s something like that i’d be tempted to take a punt. I very much doubt sapphire would go to the trouble of designing a totally different PCB for a slightly different model of card, even if things like the cooler are a little different. It’s just un-neccessary work that would cost them extra money.
More likely they’d use the same PCB and leave a couple of components off it (e.g., a power connector) or use a slightly crappier cooler or slightly lower binned GPU die. Or a GPU with lesser HBM on it or whatever.
But I’d take a punt that the PCB is the same.
I mean the reference PCBs are exactly the same between Vega 56/64/64-liquid and FE.
It just doesn’t make sense to go making another one if you can avoid it, and i reckon they could avoid it…
I thought the WX series cards did.
I could be mistaken, attempting to confirm :-\
Found this from @wendell, looks like i false-alarmed, sorry guys.
That is probably something for a separate topic since it is pretty specific.
The WX pcbs are different. The only visible detail is the pcie 8pin in a different location.
Though there is a lack of high res pictures of those cards, otherwise i’dbe looking into modding a vega64 sriov capable.
Yeah, sorry i was referring to a post (since deleted) regarding two varieties of RX vega by Sapphire… (another subtopic in this thread)
Not WX series cards vs. RX vega