5 men Syzygy on USB 3.0 Flash Drive

[Laskos]

I am a bit confused. With a lowly flash drive, having such speeds:



I got a statistically significant improvement of 8 Elo points over 5 men Syzygy on HDD. Ultrafast games from endgame starting positions, to magnify the difference, in LittleBlitzer to have all the stats:

Code: Select all

Games Completed = 10000 of 10000 (Avg game length = 13.071 sec)
Settings = Gauntlet/32MB/10000ms+100ms/M 900000cp for 5000 moves, D 180000 moves/EPD:Endgame.epd(3590)
Time = 19299 sec elapsed, 0 sec remaining
 1.  SF Syzygy USB            	5107.5/10000	2962-2747-4291  	(L: m=2747 t=0 i=0 a=0)	(D: r=981 i=685 f=2593 s=32 a=0)	(tpm=171.0 d=21.13 nps=1577415)
 2.  SF Syzygy HDD            	4892.5/10000	2747-2962-4291  	(L: m=2962 t=0 i=0 a=0)	(D: r=981 i=685 f=2593 s=32 a=0)	(tpm=173.3 d=20.53 nps=1469735)

The NPS is also higher by 7%, explaining the Elo difference. I was under impression that 5 men Syzygy on HDD would perform almost identically to flash or even SSD, as all the files are cached into RAM.

[zullil]

I am a bit confused. With a lowly flash drive, having such speeds:



I got a statistically significant improvement of 8 Elo points over 5 men Syzygy on HDD. Ultrafast games from endgame starting positions, to magnify the difference, in LittleBlitzer to have all the stats:

Code: Select all

Games Completed = 10000 of 10000 (Avg game length = 13.071 sec)
Settings = Gauntlet/32MB/10000ms+100ms/M 900000cp for 5000 moves, D 180000 moves/EPD:Endgame.epd(3590)
Time = 19299 sec elapsed, 0 sec remaining
 1.  SF Syzygy USB            	5107.5/10000	2962-2747-4291  	(L: m=2747 t=0 i=0 a=0)	(D: r=981 i=685 f=2593 s=32 a=0)	(tpm=171.0 d=21.13 nps=1577415)
 2.  SF Syzygy HDD            	4892.5/10000	2747-2962-4291  	(L: m=2962 t=0 i=0 a=0)	(D: r=981 i=685 f=2593 s=32 a=0)	(tpm=173.3 d=20.53 nps=1469735)

The NPS is also higher by 7%, explaining the Elo difference. I was under impression that 5 men Syzygy on HDD would perform almost identically to flash or even SSD, as all the files are cached into RAM.

Did you force them to be cached into RAM prior to starting your test? Did you start each test with cached memory cleared?

[Laskos]

Did you force them to be cached into RAM prior to starting your test? Did you start each test with cached memory cleared?

I didn't force them to be cached into RAM before the start. Each game, you mean?

[zullil]

Did you force them to be cached into RAM prior to starting your test? Did you start each test with cached memory cleared?

I didn't force them to be cached into RAM before the start. Each game, you mean?

Sorry, I meant to write "Did you start the match with cached memory cleared?"

[zullil]

Did you force them to be cached into RAM prior to starting your test? Did you start each test with cached memory cleared?

I didn't force them to be cached into RAM before the start. Each game, you mean?

Sorry, I meant to write "Did you start the match with cached memory cleared?"

Out of curiosity, what were the results after the first 1000 games (if you have them)?

[Laskos]

Did you force them to be cached into RAM prior to starting your test? Did you start each test with cached memory cleared?

I didn't force them to be cached into RAM before the start. Each game, you mean?

Sorry, I meant to write "Did you start the match with cached memory cleared?"

Yes.

[syzygy]

The NPS is also higher by 7%, explaining the Elo difference. I was under impression that 5 men Syzygy on HDD would perform almost identically to flash or even SSD, as all the files are cached into RAM.

At very short time control it could make a difference. That should disappear if you type

Code: Select all

type *.rtbw >nul

before the test (obviously from the directory with the files that are going to be accessed by that test).

[ernest]

With a lowly flash drive, having such speeds:

Is the access time of that flash drive, which is really what counts, just as lowly (meaning of the order of 1 ms or more) ?

[Laskos]

With a lowly flash drive, having such speeds:

Is the access time of that flash drive, which is really what counts, just as lowly (meaning of the order of 1 ms or more) ?

0.59 ms

[Laskos]

The NPS is also higher by 7%, explaining the Elo difference. I was under impression that 5 men Syzygy on HDD would perform almost identically to flash or even SSD, as all the files are cached into RAM.

At very short time control it could make a difference. That should disappear if you type

Code: Select all

type *.rtbw >nul

before the test (obviously from the directory with the files that are going to be accessed by that test).

If I am not doing something wrong, still seems 6% faster on flash.