Yeah, I saw the Anand article with the mention of the actual FP64 ratio and checked the stock immediately. Newegg was already out of stock then on all models. I've only checked once since then as I assumed the rumors of limited supply were real and they were being price gauged on eBay.
These cards are putting up some big numbers at E@H and MW. I haven't seen Collatz results but it'll be a great performer there as well.
My first Radeon VII bought on AMD's website is a dud. The default voltage is 1146mV@1801MHz. When undervolted to 1039mV it can finish Superposition Benchmark. But if I use it on E@H, it goes to zombie state quickly. The card maintains the boost clock but the temperature drops significantly. The utilization still shows 100% but the crunching will take forever. The core is not working. It also happens when the GPU doesn't get enough CPU resources. You can't tell it's working or not by the core clock or utilization. You can only tell by the temperature.
The driver doesn't crash when this happens. If you suspend the current WUs and feed it with new WUs, it would work at full power again. But it will soon go back to the zombie state. To use it on E@H I have to apply at least 1062mV@1801MHz.
This is super baffling to me because on Nvidia cards if you fail an overclock the driver just crashes.
My Asus branded Radeon VII bought on Newegg seems much better. It can run E@H with 961mV@1801MHz. Initially I put the AMD card in the first PCIe slot and the Asus one in the third PCIe slot. The AMD branded one thermal throttled and went into zombie state even at 1062mV@1801MHz profile. I could only switched both cards and applied a -20% power limit on the top Asus card to keep the junction temperatures in the 90s. Even then the Asus card still maintains the 1800MHz boost clock.
In summary the throttling behavior of Radeon VII baffles me.
But if I use it on E@H, it goes to zombie state quickly. The card maintains the boost clock but the temperature drops significantly. The utilization still shows 100% but the crunching will take forever. The core is not working. It also happens when the GPU doesn't get enough CPU resources. You can't tell it's working or not by the core clock or utilization. You can only tell by the temperature.
I've also experienced this zombie state, with the same indicators of very high WU times and low temperatures. This has happened to me when I've tried to adjust the memory clock manually. I've tried a few times to raise the memory from 1000MHz to 1050MHz (which is what I run my Vega64 memory at), but it goes into the zombie state within a few minutes. I've since given up on adjusting memory clocks for the time being.
I've been running stable for a few days with 980mV@1800MHz and a -20% power limit. I'll try to push a little harder, but I'll be surprised if I can match your 961mV, that seems very good. I got mine direct from AMD. Something does seem off with your other card, unless you're also playing with the memory clocks like I was and that's causing the issue. Can you try putting it in another machine to see if you get the same results?
Both of my cards have a 1080MHz memory clock. I will see if I can undervolt further by defaulting the memory clock. I will also try the new 19.2.3 driver. WattMan doesn't monitor core voltage. HWiNFO shows 938mV and 887mV but seems too low. I will see what newest GPU-Z and Afterburner give. I don't think the worse card is eligible for return though. It just can't really overclock right now.
The NowInstock Radeon VII page reported the PowerColor flavor to be for sale at NewEgg for three hours today at list price.
I've noticed that the photos displayed in Amazon listings for Radeon VII cards from Asus, Gigabyte, MSI, PowerColor, Saffire, and XFX look remarkably similar to one another.
Is it likely these are all to an identical standardized design from AMD, and that differentiated models will come out later? A bit like the reference cards at Nvidia a couple of generations ago?
Also: it seems the designation on these is uniformly Radeon VII, not Vega VII.
... the designation on these is uniformly Radeon VII, not Vega VII.
Yes, AMD always seem to end up with multiple and confusing naming systems. However, in this case, they seem to always refer to it as Radeon VII, particularly for marketing purposes.
Since it is an updated/improved 'Vega' architecture, together with a die shrink, and since internally, it seems to be referred to as 'Vega 20', it's not all that surprising that some people have been calling it Vega VII. I don't think I've seen 'Radeon 20' yet, though :-).
I've chosen to always try to refer to it as Radeon VII, so as not to unnecessarily add to any existing confusion :-).
The NowInstock Radeon VII page reported the PowerColor flavor to be for sale at NewEgg for three hours today at list price.
I've noticed that the photos displayed in Amazon listings for Radeon VII cards from Asus, Gigabyte, MSI, PowerColor, Saffire, and XFX look remarkably similar to one another.
Is it likely these are all to an identical standardized design from AMD, and that differentiated models will come out later? A bit like the reference cards at Nvidia a couple of generations ago?
Also: it seems the designation on these is uniformly Radeon VII, not Vega VII.
As you surmised they're all reference designs. The rumor mill at release time was that the output volume of the cards was so low that no non-reference designs were to be made and I can't find any leaks of one since then.
As you surmised they're all reference designs. The rumor mill at release time was that the output volume of the cards was so low that no non-reference designs were to be made and I can't find any leaks of one since then.
After watching NowInStock for a couple of days, and reviewing the back record portrayed there of brief windows of pre-order or order availability, and considering the limited supply situation implied by the tweaktown story, I swallowed my dislike of paying extra and facilitating scalpers, and bought one of the allegedly new cards available on eBay for around $850.
It seems likely that this specific card model has a huge differential advantage in running Einstein Gamma-Ray Pulsar work under the current application relative to the performance it gives in game play. In cold economic logic, that makes the card worth more to me than to a game player (the target market).
I should be able to report my experience next week, and have started reducing the work queue on my main machine by way of preparation.
My biggest current concern is fan noise, and my hope is that the card will respond to power restriction and fan curve specification by Afterburner to give a satisfactory operating point. While I'm not a no-walls case person, the case this card will go into is unusually well ventilated (fans on five of the six surfaces!) and reviews do say the thermal solution is very capable--just that the default fan control is very aggressive in response to game loads. So I'm optimistic.
It looks like the Radeon VIIs
)
It looks like the Radeon VIIs are also either out of stock, or selling above their list price at around $1000 in the U.S.
So there is not much to compare at the moment anyway.
Yeah, I saw the Anand article
)
Yeah, I saw the Anand article with the mention of the actual FP64 ratio and checked the stock immediately. Newegg was already out of stock then on all models. I've only checked once since then as I assumed the rumors of limited supply were real and they were being price gauged on eBay.
These cards are putting up some big numbers at E@H and MW. I haven't seen Collatz results but it'll be a great performer there as well.
My first Radeon VII bought on
)
My first Radeon VII bought on AMD's website is a dud. The default voltage is 1146mV@1801MHz. When undervolted to 1039mV it can finish Superposition Benchmark. But if I use it on E@H, it goes to zombie state quickly. The card maintains the boost clock but the temperature drops significantly. The utilization still shows 100% but the crunching will take forever. The core is not working. It also happens when the GPU doesn't get enough CPU resources. You can't tell it's working or not by the core clock or utilization. You can only tell by the temperature.
The driver doesn't crash when this happens. If you suspend the current WUs and feed it with new WUs, it would work at full power again. But it will soon go back to the zombie state. To use it on E@H I have to apply at least 1062mV@1801MHz.
This is super baffling to me because on Nvidia cards if you fail an overclock the driver just crashes.
My Asus branded Radeon VII bought on Newegg seems much better. It can run E@H with 961mV@1801MHz. Initially I put the AMD card in the first PCIe slot and the Asus one in the third PCIe slot. The AMD branded one thermal throttled and went into zombie state even at 1062mV@1801MHz profile. I could only switched both cards and applied a -20% power limit on the top Asus card to keep the junction temperatures in the 90s. Even then the Asus card still maintains the 1800MHz boost clock.
In summary the throttling behavior of Radeon VII baffles me.
shuhui1990 wrote:But if I use
)
I've also experienced this zombie state, with the same indicators of very high WU times and low temperatures. This has happened to me when I've tried to adjust the memory clock manually. I've tried a few times to raise the memory from 1000MHz to 1050MHz (which is what I run my Vega64 memory at), but it goes into the zombie state within a few minutes. I've since given up on adjusting memory clocks for the time being.
I've been running stable for a few days with 980mV@1800MHz and a -20% power limit. I'll try to push a little harder, but I'll be surprised if I can match your 961mV, that seems very good. I got mine direct from AMD. Something does seem off with your other card, unless you're also playing with the memory clocks like I was and that's causing the issue. Can you try putting it in another machine to see if you get the same results?
Both of my cards have a
)
Both of my cards have a 1080MHz memory clock. I will see if I can undervolt further by defaulting the memory clock. I will also try the new 19.2.3 driver. WattMan doesn't monitor core voltage. HWiNFO shows 938mV and 887mV but seems too low. I will see what newest GPU-Z and Afterburner give. I don't think the worse card is eligible for return though. It just can't really overclock right now.
I'm glad you guys are
)
I'm glad you guys are finding/working out all the bugs! :D
The ASUS card is selling for 22% above the RRP here in Aus. No thx.
The NowInstock Radeon VII
)
The NowInstock Radeon VII page reported the PowerColor flavor to be for sale at NewEgg for three hours today at list price.
I've noticed that the photos displayed in Amazon listings for Radeon VII cards from Asus, Gigabyte, MSI, PowerColor, Saffire, and XFX look remarkably similar to one another.
Is it likely these are all to an identical standardized design from AMD, and that differentiated models will come out later? A bit like the reference cards at Nvidia a couple of generations ago?
Also: it seems the designation on these is uniformly Radeon VII, not Vega VII.
archae86 wrote:... the
)
Yes, AMD always seem to end up with multiple and confusing naming systems. However, in this case, they seem to always refer to it as Radeon VII, particularly for marketing purposes.
Since it is an updated/improved 'Vega' architecture, together with a die shrink, and since internally, it seems to be referred to as 'Vega 20', it's not all that surprising that some people have been calling it Vega VII. I don't think I've seen 'Radeon 20' yet, though :-).
I've chosen to always try to refer to it as Radeon VII, so as not to unnecessarily add to any existing confusion :-).
Cheers,
Gary.
archae86 wrote:The NowInstock
)
As you surmised they're all reference designs. The rumor mill at release time was that the output volume of the cards was so low that no non-reference designs were to be made and I can't find any leaks of one since then.
https://www.tweaktown.com/news/64501/amd-radeon-vii-less-5000-available-custom-cards/index.html
DanNeely wrote:As you
)
After watching NowInStock for a couple of days, and reviewing the back record portrayed there of brief windows of pre-order or order availability, and considering the limited supply situation implied by the tweaktown story, I swallowed my dislike of paying extra and facilitating scalpers, and bought one of the allegedly new cards available on eBay for around $850.
It seems likely that this specific card model has a huge differential advantage in running Einstein Gamma-Ray Pulsar work under the current application relative to the performance it gives in game play. In cold economic logic, that makes the card worth more to me than to a game player (the target market).
I should be able to report my experience next week, and have started reducing the work queue on my main machine by way of preparation.
My biggest current concern is fan noise, and my hope is that the card will respond to power restriction and fan curve specification by Afterburner to give a satisfactory operating point. While I'm not a no-walls case person, the case this card will go into is unusually well ventilated (fans on five of the six surfaces!) and reviews do say the thermal solution is very capable--just that the default fan control is very aggressive in response to game loads. So I'm optimistic.