Radeon VII (2nd gen consumer Vega GPU) - Page 3

preda · 2019-01-17, 12:05

Quote:

Originally Posted by SELROC

I am procrastinating the buy a new more powerful gpu, do you have any plans to optimize gpuowl for large numbers ?

I don't have any clear optimization ideas at this stage. (aside from going down the hand-assembly path, which is not realistic for me because it's a lot of work)

What large numbers do you have in mind? Do you think of some specific optimizations?

2019-01-17, 12:13

Quote:

Originally Posted by preda

I don't have any clear optimization ideas at this stage. (aside from going down the hand-assembly path, which is not realistic for me because it's a lot of work)

What large numbers do you have in mind? Do you think of some specific optimizations?

The 300M to 500M exponents.

A 332M exponent took 2 months of gpu work on the RX580.
https://www.mersenne.org/report_expo...2412937&full=1

as a side note: it seems it is now assigned to someone else.

nomead · 2019-01-17, 19:57

https://www.pcgamer.com/amd-scoffs-a...short-supply//

It's truly silly season now...

2019-01-18, 16:08

Quote:

Originally Posted by M344587487

It's a GFX9 card so no: https://github.com/RadeonOpenCompute/ROCm

Quote:

Originally Posted by preda

2x would be amazing. In practice I would be very happy if I see a 50% speedup.

About memory, it is my impression that the latency did not improve much, but the bandwidth doubled. But to take advantage of this, better occupancy would be required (double the number of memory operations in flight), and this is not easily achievable because of other limiting resources: LDS memory and nb. of registers (VGPRs) that remain unchanged I guess.

About compute, the parts that aren't DP (e.g. pointer arithmetic, other integer e.g. carry, logic) remain unchanged, and this will reduce the observed speedup.

IMO another limiting factor for GCN performance is still the compiler, after so many years: the compiler does a rather poor job at generating highly efficient code (not an easy task I agree).

OTOH the better cooling will help, and allow the card to be higher clocked without thermal throttling (which is a problem on Vega64 blower cooler)

https://www.phoronix.com/scan.php?pa...se-AMD-GPU-TDP

Mark Rose · 2019-01-23, 03:03

.

mackerel · 2019-02-07, 14:28

Haven't had time to dig yet, saw at Anandtech that AMD changed their minds yet again and FP64 is now 1/4 rate.

https://www.anandtech.com/show/13923...eon-vii-review

Mark Rose · 2019-02-07, 14:32

I just came here to post that

Quote:

The Radeon VII graphics card was created for gamers and creators, enthusiasts and early adopters. Given the broader market Radeon VII is targeting, we were considering different levels of FP64 performance. We previously communicated that Radeon VII provides 0.88 TFLOPS (DP=1/16 SP). However based on customer interest and feedback we wanted to let you know that we have decided to increase double precision compute performance to 3.52 TFLOPS (DP=1/4SP).

M344587487 · 2019-02-07, 15:13

Quote:

Originally Posted by Mark Rose

I just came here to post that

Just to make sure all the bases are covered this guy says that FP64 is ~1.7 aka DP=1/8 SP in his gaming review although he's probably parroting old information: https://www.youtube.com/watch?v=6jP3tetYnVI

I was expecting £700 but it's £650 in the UK. Tried to buy one but my bank decided I was trying to steal from myself and now they're out of stock so that's nice.

mackerel · 2019-02-07, 15:31

I saw it in stock at Scan at £650, but as I was looking at other stores it sold out. Some places are indicating stock arriving tomorrow so assume shipments are ongoing. OCUK have some in stock at £800. I don't want it enough to pay £150 premium for what looks identical to the £650 ones.

This could be the GPU to make the largest known prime not be a mersenne. Over at PrimeGrid they just started "do you feel lucky" project which are GFN22 at a high enough level to exceed largest known prime. Fastest GPUs so far are doing about one a day and the code is FP64. If this card could do several units a day, that would help a lot.

Edit: forget that last part. Just been pointed out to me that specific project can't use FP64. Regular GFN21/22 could still see a significant benefit.

nomead · 2019-02-07, 16:14

Interesting. How late in the product cycle can they make these decisions on how much FP64 to include? Is it configuration fuses on the die, microcode update, driver limitation, or what? And of course... could it be hacked afterwards

tServo · 2019-02-07, 16:26

I'll buy 1 ( eventually ) to test but 3 thoughts:

(1) The board is going to be difficult to "live with":
High power requirements and blasts lots of heat IN THE CASE. Reviewers have noted its fans are obnoxiously loud.

(2) It's impossible to get, of course. It will be interesting to see how soon AMD can alleviate this situation. Is this a result of poor 7nm yields?

(3) With such impressive specs, I would have thought that it would absolutely CRUSH other boards in toe-to-toe comparison tests. It wins a lot, but not as many and not by as much.

2019-01-23, 03:03	#27
Mark Rose "/X\(‘-‘)/X\" Jan 2013 https://pedan.tech/ C70₁₆ Posts	. Last fiddled with by Mark Rose on 2019-01-23 at 03:04

2019-02-07, 15:31	#31
mackerel Feb 2016 UK 111000000₂ Posts	I saw it in stock at Scan at £650, but as I was looking at other stores it sold out. Some places are indicating stock arriving tomorrow so assume shipments are ongoing. OCUK have some in stock at £800. I don't want it enough to pay £150 premium for what looks identical to the £650 ones. This could be the GPU to make the largest known prime not be a mersenne. Over at PrimeGrid they just started "do you feel lucky" project which are GFN22 at a high enough level to exceed largest known prime. Fastest GPUs so far are doing about one a day and the code is FP64. If this card could do several units a day, that would help a lot. Edit: forget that last part. Just been pointed out to me that specific project can't use FP64. Regular GFN21/22 could still see a significant benefit. Last fiddled with by mackerel on 2019-02-07 at 16:14 Reason: updated info

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Vega 20 announced with 7.64 TFlops of FP64	M344587487	GPU Computing	4	2018-11-08 16:56
GTX 1180 Mars Volta consumer card specs leaked	tServo	GPU Computing	20	2018-06-24 08:04
RX Vega performance	xx005fs	GPU Computing	5	2018-01-17 00:22
Radeon Pro Duo	0PolarBearsHere	GPU Computing	0	2016-03-15 01:32
AMD Radeon R9 295X2	firejuggler	GPU Computing	33	2014-09-03 21:42

2019-01-17, 19:57	#25
nomead "Sam Laur" Dec 2018 Turku, Finland 475₈ Posts	https://www.pcgamer.com/amd-scoffs-a...short-supply// It's truly silly season now...

2019-02-07, 14:28	#28
mackerel Feb 2016 UK 1C0₁₆ Posts	Haven't had time to dig yet, saw at Anandtech that AMD changed their minds yet again and FP64 is now 1/4 rate. https://www.anandtech.com/show/13923...eon-vii-review

2019-02-07, 16:14	#32
nomead "Sam Laur" Dec 2018 Turku, Finland 13D₁₆ Posts	Interesting. How late in the product cycle can they make these decisions on how much FP64 to include? Is it configuration fuses on the die, microcode update, driver limitation, or what? And of course... could it be hacked afterwards

2019-02-07, 16:26	#33
tServo "Marv" May 2009 near the Tannhäuser Gate 327₁₆ Posts	I'll buy 1 ( eventually ) to test but 3 thoughts: (1) The board is going to be difficult to "live with": High power requirements and blasts lots of heat IN THE CASE. Reviewers have noted its fans are obnoxiously loud. (2) It's impossible to get, of course. It will be interesting to see how soon AMD can alleviate this situation. Is this a result of poor 7nm yields? (3) With such impressive specs, I would have thought that it would absolutely CRUSH other boards in toe-to-toe comparison tests. It wins a lot, but not as many and not by as much.