No you cannot! You will have a small, minor, speed increase but ONLY once sales start to fall on the current boards. Please understand that we must max the $$$$.
>a 1x10^17 bit error rate > a Silent Bit Error Rate (SBER) lower than 10^23
1x10^17 is just 10^17 - this info is utterly meaningless. As is the 10^23 number. I presume you meant to say 1 in 10^17, and 1 in 10^23.
Either sloppy cut'n'paste from the marketing, or poor understanding of what the bit error rate numbers represent, either of which isn't good for a tech site author. :(
Actually the article said 1×10^-17. See the minus sign? Which implies 1/(10^17) Which implies 1 in 10^17. The article is right.
But the table should not have the minus sign for the power as it says "1 bit per ...".
It doesn't look at all to me a poor understanding. The table error is obviously an oversight. Excuse them for being mere humans that sometimes make small mistakes.
Nah, the edit has made things worse. The principle of a bit error rate is that it is quoted as less than one error in (hopefully) a *very* large number. Putting minus signs in the exponent makes the number a miniscule fraction instead. It's just factually wrong, and the article text hasn't been corrected for the original mistake in quoting the number alone (without the "one error in" info).
Clearly the author doesn't understand at all what he's regurgitating.
The minus says that the chance of a bit getting corrupt is 0.00000000000000001%
If there is one corrupt bit every 10^17 bits, then the odds of any one bit getting corrupt is 1 / 10^17 or interchangeably - 10^-17
So either leave it "1 per 10^17" or just "10^-17"
Because right now as it is - "1 bit per 10^-17 bits read" - means that each and every bit read will have 10^17 errors in it LOL. That's immensely worse than having 100% error rate LOL. If you have 1 error bit per each 10 attobits read, you will mount 100 peta errors until you are done reading a single whole bit.
Precisely. The apparent point of these tech promo articles is to convey correct marketing (uh, information) from the manufacturers, without much editorial in the way. If the only "editorial" input from Shilov is to change a few numbers and words so as to make it read less like the original press release, but then totally mess up what the numbers mean, it's not doing either the site or the manufacturer any favours.
And while your maths is OK, I still take issue with the idea of representing the BER as a probability stated as a fraction. It's universally described by manufacturers (as in the datasheet) as "one error in X bits read", not as a pure number, fractional or otherwise. That's why the article text is factually wrong as it stands, even if the intent is clear (but currently wrong due to the negative exponents) in the tables.
Yet again, the shocking lack of a proper in-house editor at Anandtech is exposed. :(
I have something for sale too. If you have to ask what it is then it isn't for you. If you have to ask how much then you can't afford it. Just give me all the money and trust ;)
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
19 Comments
Back to Article
R0H1T - Saturday, June 24, 2017 - link
I definitely want one, or two. Memblaze can probably send a few for (home) server reviews ^_^Bullwinkle J Moose - Saturday, June 24, 2017 - link
Can we haz PCIe 4.0 X 16 next year and PCIe 5 in 2019 ?cnom now....FIGHT FIGHT FIGHT FIGHT!
damianrobertjones - Monday, June 26, 2017 - link
No you cannot! You will have a small, minor, speed increase but ONLY once sales start to fall on the current boards. Please understand that we must max the $$$$.asmian - Saturday, June 24, 2017 - link
>a 1x10^17 bit error rate> a Silent Bit Error Rate (SBER) lower than 10^23
1x10^17 is just 10^17 - this info is utterly meaningless. As is the 10^23 number. I presume you meant to say 1 in 10^17, and 1 in 10^23.
Either sloppy cut'n'paste from the marketing, or poor understanding of what the bit error rate numbers represent, either of which isn't good for a tech site author. :(
XZerg - Saturday, June 24, 2017 - link
now 10^-17 or 10^-23remove the "-", it should just be 10^17 or 10^23
rems - Saturday, June 24, 2017 - link
Think so too!Tadashi130 - Saturday, June 24, 2017 - link
Actually the article said 1×10^-17. See the minus sign? Which implies 1/(10^17) Which implies 1 in 10^17. The article is right.But the table should not have the minus sign for the power as it says "1 bit per ...".
It doesn't look at all to me a poor understanding. The table error is obviously an oversight. Excuse them for being mere humans that sometimes make small mistakes.
asmian - Sunday, June 25, 2017 - link
Nah, the edit has made things worse. The principle of a bit error rate is that it is quoted as less than one error in (hopefully) a *very* large number. Putting minus signs in the exponent makes the number a miniscule fraction instead. It's just factually wrong, and the article text hasn't been corrected for the original mistake in quoting the number alone (without the "one error in" info).Clearly the author doesn't understand at all what he's regurgitating.
ddriver - Monday, June 26, 2017 - link
1 out of 10 is the same as 0.1The minus says that the chance of a bit getting corrupt is 0.00000000000000001%
If there is one corrupt bit every 10^17 bits, then the odds of any one bit getting corrupt is 1 / 10^17 or interchangeably - 10^-17
So either leave it "1 per 10^17" or just "10^-17"
Because right now as it is - "1 bit per 10^-17 bits read" - means that each and every bit read will have 10^17 errors in it LOL. That's immensely worse than having 100% error rate LOL. If you have 1 error bit per each 10 attobits read, you will mount 100 peta errors until you are done reading a single whole bit.
ddriver - Monday, June 26, 2017 - link
Ops, that should be 0.000000000000001%ddriver - Monday, June 26, 2017 - link
Lastly, thou shalt have copieth from teh manufacturer charts, which clearly and adequately state " < 1 bit per 10^17 bits read "asmian - Monday, June 26, 2017 - link
Precisely. The apparent point of these tech promo articles is to convey correct marketing (uh, information) from the manufacturers, without much editorial in the way. If the only "editorial" input from Shilov is to change a few numbers and words so as to make it read less like the original press release, but then totally mess up what the numbers mean, it's not doing either the site or the manufacturer any favours.And while your maths is OK, I still take issue with the idea of representing the BER as a probability stated as a fraction. It's universally described by manufacturers (as in the datasheet) as "one error in X bits read", not as a pure number, fractional or otherwise. That's why the article text is factually wrong as it stands, even if the intent is clear (but currently wrong due to the negative exponents) in the tables.
Yet again, the shocking lack of a proper in-house editor at Anandtech is exposed. :(
shabby - Saturday, June 24, 2017 - link
Please post prices... oh nevermind.kaidenshi - Saturday, June 24, 2017 - link
I'm pretty sure this is in the "if you have to ask, you can't afford it" range.CaedenV - Saturday, June 24, 2017 - link
I have something for sale too.If you have to ask what it is then it isn't for you. If you have to ask how much then you can't afford it. Just give me all the money and trust ;)
kaidenshi - Saturday, June 24, 2017 - link
Whoosh.PCTC2 - Saturday, June 24, 2017 - link
"and comes with further priority que management optimizations over the previous version."que -> queue.
JoeyJoJo123 - Monday, June 26, 2017 - link
Unless I'm mistaken, you should probably search and replace all "DPWD" with "DWPD", because I believe you're referring to:Drive
Write
Per
Day
ddriver - Monday, June 26, 2017 - link
Nah, that's a whole different metric - Drives Per Write Day, that's how many drives will fail in a day of writing.Consistent with the 1 quadrillion percent error rate listed in the chart.