Radeon 5600 XT

Jim1348
Jim1348
Joined: 19 Jan 06
Posts: 463
Credit: 257957147
RAC: 0
Topic 220407

Like the other 5000-series Radeon RX cards, the 5600 XT is based on the new RDNA 7nm architecture. AMD boasts huge increases over its predecessor, the RX 590, in two specific categories: Performance and power consumption. Performance is 42% and power is down 33%, which is far larger than a standard generational increase.

https://www.digitaltrends.com/computing/can-amd-radeon-5600-xt-replace-gtx-1060-ces-2020/

Also, the memory bandwidth is increased over the 5500 XT, which appeared to be a limiting factor to me on Einstein.

If some one wants to try it out, I am sure we will all be interested.

 

archae86
archae86
Joined: 6 Dec 05
Posts: 3157
Credit: 7205854931
RAC: 928988

It seems likely that the 5600

It seems likely that the 5600 architecture is similar enough to the 5700 that it would have had 100% failure rates on Einstein GRP work with the versions of the driver and the application available until extremely recently.

On the Navi 10 thread Mumak has reported success running very recently released driver 20.1.1 on a Windows 10 system hosting a 5700 XT.  So far as I know his are the first reports of validation on 5700 series cards except for rare cases where a quorum would accidentally consist of two such cards--which would agree with each other enough to validate.

I suggest anyone considering a 5600 series card for Einstein GRP use monitor closely the Navi 10 thread for ongoing developments.

Jim1348
Jim1348
Joined: 19 Jan 06
Posts: 463
Credit: 257957147
RAC: 0

OK.  It looks like his times

OK.  It looks like his times for the Gamma-ray pulsar binary search #1 are about 350 seconds, which is not quite twice as fast as the RX 570 as I recall.  So the 5600 won't be even that fast.  I would wait until I see a factor of 2 improvement, unless the GW work units tell a different story, but I doubt it.  Thanks.

EDIT: While we are on the subject, AMD rates the "typical board power" for the 5700 XT at 225 watts, and the

Peak Single Precision Compute Performance at 9.75 TFLOPs, or 23 watts/TFLOP, or 0.043 TFLOP/watt.



 

The RX 570 is rated at 150 watts and 5.1 TFLOPs, or 29 watts/TFOP, or 0.034 TFLOP/watt.


That latter number is the efficiency that I normally look to double when upgrading.  I think it will take another generation at this rate.

 

 

 


 

 

Keith Myers
Keith Myers
Joined: 11 Feb 11
Posts: 4957
Credit: 18619857825
RAC: 5614225

Quote:I think it will take

Quote:
I think it will take another generation at this rate.

You may get that generational improvement if "Big Navi" is actually more than hype.

 

 

Jim1348
Jim1348
Joined: 19 Jan 06
Posts: 463
Credit: 257957147
RAC: 0

I hear the name "Arcturus"

I hear the name "Arcturus" too, though that may be more for professional use.  It is all wonderfully vague at the moment.

Mumak
Joined: 26 Feb 13
Posts: 325
Credit: 3496504054
RAC: 2092934

Yeah, I don't see much

Yeah, I don't see much difference in power efficiency between Polaris and Navi10. My measurements show that the Navi10 performs ~2 times faster than the RX 580, but power consumption has doubled as well.
Perhaps one might achieve better results with undervolting I haven't tried this yet, but plan to do so.

-----

cecht
cecht
Joined: 7 Mar 18
Posts: 1525
Credit: 2867856018
RAC: 2060395

archae86 wrote:It seems

archae86 wrote:

It seems likely that the 5600 architecture is similar enough to the 5700 that it would have had 100% failure rates on Einstein GRP work with the versions of the driver and the application available until extremely recently.

On the Navi 10 thread Mumak has reported success running very recently released driver 20.1.1 on a Windows 10 system hosting a 5700 XT.  So far as I know his are the first reports of validation on 5700 series cards except for rare cases where a quorum would accidentally consist of two such cards--which would agree with each other enough to validate.

I suggest anyone considering a 5600 series card for Einstein GRP use monitor closely the Navi 10 thread for ongoing developments.

According to TechPowerUp GPU database, the 5600 uses the same chip as the 5700 XT, just with some shaders disabled.  So, right, same crunching issues would apply to both cards.

Ideas are not fixed, nor should they be; we live in model-dependent reality.

Jim1348
Jim1348
Joined: 19 Jan 06
Posts: 463
Credit: 257957147
RAC: 0

Mumak wrote:Perhaps one might

Mumak wrote:
Perhaps one might achieve better results with undervolting I haven't tried this yet, but plan to do so.

I started out undervolting my RX 570 with the ASRock Phantom utility.  It worked very well in reducing the power and heat, which was my one big concern with AMD cards.  I took it down to 1.010 volts as I recall, or maybe a bit lower, which even allowed me ito increase the clock from 1244 to 1348 MHz on Gamma Ray, keeping the temps reasonable (GW ran even cooler).  That was with the Radeon 18.3.4 drivers under Win7 64-bit. 

However, with the 19.9.2 drivers that I am using now, it does not appear to be necessary to set anything manually.  Or maybe the work units have changed in the interim.  But I now just let the drivers handle it, and it seems to run about as cool.  I found the latest drivers (19.12.2) to be unstable, causing freezes and lockups. 

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.