Discussion in 'Videocards - AMD Radeon' started by OnnA, Jan 20, 2019.
Anything on crossfire?
Radeon VII & Radeon RX Vega 64 Clock-for-Clock Performance
With the variety of changes from the Vega 10 powered RX Vega 64 to the new Radeon VII and its Vega 20 GPU, we wanted to take a look at performance and compute while controlling for clockspeeds.
In this way, we can peek at any substantial improvements or differences in pseudo-IPC.
There's a couple caveats here; obviously, because the RX Vega 64 has 64 CUs while the Radeon VII has only 60 CUs, the comparison is already not exact.
The other thing is that "IPC" is not the exact metric measured here, but more so how much graphics/compute work is done per clock cycle and how that might translate to performance. Isoclock GPU comparisons tend to be less useful when comparing across generations and architectures,
as like in Vega designers often design to add pipeline stages to enable higher clockspeeds, but at the cost of reducing work done per cycle and usually also increasing latency.
For our purposes, the incremental nature of 2nd generation Vega allays some of those concerns, though unfortunately, Wattman was unable to downclock memory at this time, so we couldn't get a set of datapoints for when both cards are configured for comparable memory bandwidth.
While the Vega GPU boost mechanics means there's not a static pinned clockspeed, both cards were set to 1500MHz, and both fluctuated from 1490 to 1500MHZ depending on workload.
All combined, this means that these results should be taken as approximations and lacking granularity, but are useful in spotting significant increases or decreases.
This also means that interpreting the results is trickier, but at a high level, if the Radeon VII outperforms the RX Vega 64 at a given non-memory bound workload,
then we can assume meaningful 'work per cycle' enhancements relatively decoupled from CU count.
I'm a little annoyed at the change in narrative over VRAM amount.
I remember when the Fury GPU's came out with 4 GB of VRAM, AMD had a campaign explaining how it was not a big deal and how memory usage of X doesn't mean you needed that amount.
They were right. And that applies EVEN MORE now. Most modern games fill up VRAM regardless as to how much VRAM they need to properly stream resources. A game might USE 15 GB of VRAM, but literally perform and load the exact same way with only 6 GB.
But now AMD's promotional material has completely done a 180 on this fact. Seemingly to justify 16 GB of VRAM (the typical amount of system ram most PC's have...)
8 GB of HBM2 would have been fine. And the penalty in bandwidth likely would have been negligible. In fact, they likely would have gotten a performance boost if instead they used 64 CU's ... and it would have been cheaper in the bill of materials.
The reason they need to lie to justify expensive 16 GB of HBM2 is because, while a 64 CU 7mm 8GB HBM2 card would have been cheaper and made more sense, it would have cost RND. It was easier to take a workstation card, the MI50 and rebrand it.
NVidia's cards are expensive because they are forcing tech on you that you can't use. But it's there. AMD is just as expensive because they were lazy.
On average Vega VII is ~8FPS faster than my LiQuiD at 1717MHz/1150HBM2 | 1.087v/0.975v
I will test some more to find out the difference between the two, when my LC is set to 1757/1150
SotTR -> 1440p Ultra + TAA (like Hexus test)
Vega VII -> 63/83 (min/Average)
Vega 64 LC -> 61/75 (min/Average)
ZEN set to 4GHz (to avoid CPU bottleneck)
For Gaming i have it at 3.9-3.96GHz
Im waiting for Arcturus
IMO Vega VII is good GPU, we need to wait for price cut to ~600€ -> Then it will be Best of the best.
Ordered mine this morning from AMD site. Now I need the water block for it.
While I love my crossfire Vegas, glad to go back to one card lol.
I see even HERE its started..some numbers being posted.. your like.. wow I cant find those numbers ANY where else..ok. For me $700 is so far over the top. When a OLD card NV 1080Ti is just AS good. When the NV 2080 so far as of right now is better in most games...
You know.. for me its not that. Well $700? WHATS NEW VS VEGA 64? One AMD should have waited. ALL the cards sold look the same. Even now prices are starting to get higher. We look at the NV 2080 and theres a HOPE the price is worth it.. RTX DLSS...if those kick in then YEAH! But you have something NEW.. the VII is ..smaller 7nm but.. needs MORE Power then the 2080
Got my card earlier today. Unfortunately it seems to highly dislike undervolting. I can't seem to go below ~1030mV without hitting stability issues at stock clock and power limits. The frequency sticks to around 1799 MHz.
Edit: Upon further testing, I'm sure drivers are being weird. If I leave the max voltage at 1047mV but I set it to be at 1900 MHz instead of the default 1801 MHz, then 1801 MHz ends up having 992mV specified, and it's perfectly stable. If I change the max frequency back to 1801, and manually set that to 992mV, the card will crash out of 3DMark Time Spy. My score is also then lower. In fact, no matter what I do apart from stock, I will lose score. If I increase or decrease voltage, increase power limits, or move max frequency around, I lose score. If I touch the memory frequency, I lose score. Not sure yet what's going on.
New Overdrive version (8) and a updated driver from 19.2.1 with VII support. With luck the next main release will fare better than this one.
The list of known issues for 19.2.1 VII seems to hint at that at least, hopefully fixes will be quick but who can say. Adrenaline 2019 and now a new GPU with further changes so the driver team has quite a bit of work ahead of them I'd imagine.
I've seen this tool before but haven't run it until now. Has AMD started do to rasterisation differently on the VII?
I believe the only way to achieve the 1TB/s bandwidth the Radeon VII provides was to provide a 4096-bit bus, which requires 16Gb of HBM2. The amount of VRAM and resulting bus-width are tied together. Recall that the Vega 64, with it's 8Gb of HBM2 had "only" a 2048-bit memory bus, with less than half the bandwidth as a result.
Now, will that immense bandwidth the VII possesses be beneficial in the long term? I certainly don't know that answer. Could AMD have tossed the whole existing Instinct design and re-invented the wheel to provide a 4096-bit wide memory bus using only 8Gb of HBM2? Not being an electronics engineer I'll roll the dice and say, "probably."
But, I don't believe I need to be an engineer, or a CFO, to know that the cost and time that effort would take would never have been a serious consideration. That isn't being "lazy." It's being realistic.
You do have some serious issues.
Anyone get a tracking number from AMD.com yet? Mine is still showing "Order in Process."
After further testing, I've found that 1047 mV at 1800 MHz (actual frequency is around 1780 MHz) is stable. A bit disappointed at how poorly my chip handles lower voltage at these frequencies, but we'll see if we can still get some other goodness out of it.
Actually they do sell 2gb hbm2 stacks (edit: or as least planned to sell) , 8gb is possible, though i think mounting height would be a problem, kinda like unmolded vega 64/56 cards
edit:nvm, seems they never went into production oh well,
The radeon 7 is a way to get rid of bottom bin sillicon just like the titan v, if they could be sold as instinct cards they would be, retooling to produce 8gb variants would probably cost more than what ever they get out of the radeon 7.
Driver is still in Early beta stage -> Give them some time
0.950v is possible NP with good driver.
It seems like this is not quite the card I have been waiting for. I will wait on AIB versions to see if they fix the noise issues. I would wait on Navi, but with the rumors saying not until October that's just far to long to wait. Not really sure which way to go at the end of the day. Sadly this launch is disappointing.
@OnnA That secret note from Lisa would make this little boy soo happy lol. To whomever got those, frame it.
As for the noise in my situation its ok. Im going to waterblock it, but i have not seen anything from EK or XSPC yet.
(Vega 64 LiQuiD OC is here and doing Great at 1440p)
No big problems with OC or UV (Yes on the Press driver)
Power (0% Power Slider in WattMan = Max BIOS 264tW)
Also tested with UV to 981mV (Valid V for Radeons)
Power with UV at V56 Levels but with 30%+ more FPS