Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Are the RE3 / ES.2 versions worth the price diff? (added 2nd Q!)
- Thread starter Loa
- Start date
- Sort by reaction score
Hello,
Come on man... The Workbench graph you want us to look at represents the number of I/O operations. Not data transfers, nor maximum throughput.
If you want to see how the drives perform under real world usage, flip back a page in the article. (http://www.tomshardware.com/reviews/intel-x25-e-ssd,2158-4.html)
Look at the read, write and maximum data throughput.
On the write the Cheetah is neck and neck with Intel's 800$ SSD, and even faster than Intel's cheaper alternative. And maximum drive interface bandwidth: the same.
Overall: 2 of the Cheetahs in RAID0 will go faster than Intel's 800$ SSD.
Loa
Come on man... The Workbench graph you want us to look at represents the number of I/O operations. Not data transfers, nor maximum throughput.
If you want to see how the drives perform under real world usage, flip back a page in the article. (http://www.tomshardware.com/reviews/intel-x25-e-ssd,2158-4.html)
Look at the read, write and maximum data throughput.
On the write the Cheetah is neck and neck with Intel's 800$ SSD, and even faster than Intel's cheaper alternative. And maximum drive interface bandwidth: the same.
Overall: 2 of the Cheetahs in RAID0 will go faster than Intel's 800$ SSD.
Loa
Hello,
Come on man... The Workbench graph you want us to look at represents the number of I/O operations. Not data transfers, nor maximum throughput.
If you want to see how the drives perform under real world usage, flip back a page in the article. (http://www.tomshardware.com/reviews/intel-x25-e-ssd,2158-4.html)
Look at the read, write and maximum data throughput.
On the write the Cheetah is neck and neck with Intel's 800$ SSD, and even faster than Intel's cheaper alternative. And maximum drive interface bandwidth: the same.
Overall: 2 of the Cheetahs in RAID0 will go faster than Intel's 800$ SSD.
Loa
Aren't those sequential speeds?
http://www.yougamers.com/forum/showthread.php?t=96682
http://www.legitreviews.com/article/788/9/
http://www.pcper.com/article.php?aid=616&type=expert&pid=7
PCMark is a better bench for real world use, seq isn't accurate..
http://www.legitreviews.com/article/788/9/
http://www.pcper.com/article.php?aid=616&type=expert&pid=7
PCMark is a better bench for real world use, seq isn't accurate..
Yes. And mine is set up with EcoGreen drives for crying out loud.
But here's something to think about. They have 500GB platters now right? This density affords great speed! Are there any single platter 500GB drives out there? If there are then that is totally the way I would go! If theere are I bet they'd hall-ass.
If there aren't any right now I wonder how long it'll be till there are?Anyone know the answers to theses questions?
PS:
Sequential is plenty accurate for testing sequential operations. Many wares use sequential operations. For Random Operations it depends on the Data size just like Loa has been saying. Here my Green drives being tested with Random I/O operations and various sizes:
And remember these are Green Drives!!! Loa is right on! SSD just isn't up to snuff yet - not for the price per gig for sure!
☯
Man, we're running in circles now. As far as I know, only your OS will do a "lot" of random reads.
For everything else, and even for a boot drive, nothing I've seen has changed my mind: 3+ RAID0 sets will be faster. And cheaper. And give you more space.
Loa
Yes that's right. Random I/O tests are not 100% accurate either. Even for OS loading. Most (all?) random tests use a butterfly pattern across the entire platter surface. Almost everything you do will be some mixture of random and sequential. For most application data the majority will be of a very sequential nature - and large enough that rotational media drives can shine - whooping SSDs...
Ok, you found some results showing that a SSD is better than 2 VR.
That's a good thing. Seriously. But it would be a lot more interesting for if they were Mac tests, with applications that actually required fast drives like PS, or any type of video editing.
I'm also very curious about their results: the WindowMediaPlayer test, wouldn't that be completely sequential in nature? Reading a video?
I *want* SSDs to be good. But if you recall correctly at the start of this discussion, we put some fundamental problems aside: cost per GB and maximum disk size.
These problems remain: how can you have a main HD that's only ~100GB??? It's all very fine for a boot drive, but what about the rest? If I want a TB of very fast data, I buy 12 of them and pay 4000$??? Using Tesselator's prices (ahem) he gets 4.5TB of data for the price a single 80GB X25M SSD.
Even IF a SSD was faster than a 4 drive RAID0, which I'm still not convinced of, those two problems still need to be addressed.
Loa
Loa and Tesselator are on the money here, IMO.
But there's another issue that gets ignored from what I've seen with SSD's. They're great for Random Access reads, but the write reliablity of flash just isn't there for high write environments. A UBE (Unrecoverable Bit Error) of 1E4 or 1E5 just can't compete with that of even a consumer drive of 1E14, and enterprise at 1E15. Despite the additional capacity added for remapping. The actual capacity would have to be much greater. At least 3x on 1E5 flash. (1E5)^3 (wear leveling technique) to equal an enterprise mechanical drive.
Then there's the issues with TRIM (non existant for OS X), and of course the costs.
All in all, SSD's are nice, but are too immature ATM to be considered reliable. And that's one of the most important aspects of RAID. RAID 0 is actually the exception to the rule, not the standard.
Until SSD's mature, mechanical drives seem to hold the overall advantage. At least for reliablilty, capacity, and cost. And the performance differences aren't that staggering to me.
But there's another issue that gets ignored from what I've seen with SSD's. They're great for Random Access reads, but the write reliablity of flash just isn't there for high write environments. A UBE (Unrecoverable Bit Error) of 1E4 or 1E5 just can't compete with that of even a consumer drive of 1E14, and enterprise at 1E15. Despite the additional capacity added for remapping. The actual capacity would have to be much greater. At least 3x on 1E5 flash. (1E5)^3 (wear leveling technique) to equal an enterprise mechanical drive.
Then there's the issues with TRIM (non existant for OS X), and of course the costs.
All in all, SSD's are nice, but are too immature ATM to be considered reliable. And that's one of the most important aspects of RAID. RAID 0 is actually the exception to the rule, not the standard.
Until SSD's mature, mechanical drives seem to hold the overall advantage. At least for reliablilty, capacity, and cost. And the performance differences aren't that staggering to me.
Loa and Tesselator are on the money here, IMO.
But there's another issue that gets ignored from what I've seen with SSD's. They're great for Random Access reads, but the write reliablity of flash just isn't there for high write environments. A UBE (Unrecoverable Bit Error) of 1E4 or 1E5 just can't compete with that of even a consumer drive of 1E14, and enterprise at 1E15. Despite the additional capacity added for remapping. The actual capacity would have to be much greater. At least 3x on 1E5 flash. (1E5)^3 (wear leveling technique) to equal an enterprise mechanical drive.
Then there's the issues with TRIM (non existant for OS X), and of course the costs.
All in all, SSD's are nice, but are too immature ATM to be considered reliable. And that's one of the most important aspects of RAID. RAID 0 is actually the exception to the rule, not the standard.
Until SSD's mature, mechanical drives seem to hold the overall advantage. At least for reliablilty, capacity, and cost. And the performance differences aren't that staggering to me.
Where does your UBE number come from?
http://download.intel.com/design/flash/nand/extreme/319984.pdf
http://download.intel.com/design/flash/nand/mainstream/mainstream-sata-ssd-datasheet.pdf
Both specify a 1x10^15 BER.
Also..though they aren't 12 vs 1..
http://smartbbs.info/x25-m.html
http://download.intel.com/design/flash/nand/extreme/OLTP_Performance_Comparison_Solid-State.pdf
http://download.intel.com/design/fl...dia_Load_Simulator_Performance_Comparison.pdf
http://download.intel.com/design/flash/nand/extreme/Photoshop_CS4_Performance_Comparison.pdf
http://download.intel.com/design/flash/nand/extreme/ESRP_Performance_Comparison.pdf
http://www2.electronicproducts.com/...t_matter-article-FAJH_Intel_Apr2009-html.aspx
http://www.intel.com/performance/mobile/sata/sata.htm
I have a Mac Pro running 3 Vertex drives, it maxes out the ICH10R at 750/650 read/writes.
Also it's only sequential data if it isn't fragmented..
http://it.anandtech.com/IT/showdoc.aspx?i=3532&p=1
MLC = 1E4 for a single unrecoverable bit error
SLC = 1E5 for a single unrecoverable bit error
The above is the design limitation of each type of NAND flash, and is the write failure, not read. This is my issue with flash. The read error rates are much better than their write failure rates, unlike mechanical drives.
In the case of BER, it's not a single bit. Simply put, it's the number of erroneous bits recieved divided by the total number of bits transmitted. Usually, this applies to packet data. In this case, they've equated an entire sector with a packet for the calculation.
From the link given, BER = 1 sector per 1E15 bits read.
Hopefully, this makes sense.
MLC = 1E4 for a single unrecoverable bit error
SLC = 1E5 for a single unrecoverable bit error
The above is the design limitation of each type of NAND flash, and is the write failure, not read. This is my issue with flash. The read error rates are much better than their write failure rates, unlike mechanical drives.
In the case of BER, it's not a single bit. Simply put, it's the number of erroneous bits recieved divided by the total number of bits transmitted. Usually, this applies to packet data. In this case, they've equated an entire sector with a packet for the calculation.
From the link given, BER = 1 sector per 1E15 bits read.
Hopefully, this makes sense.
I still don't get where your numbers come from. Source?
Also how big is a sector? And on HDDs it's per sector as well...
Partially true. The main difference between what benchmarkers call sequential I/O and random I/O tests are seeks. If there is very minimal fragmentation to a 30gig file and you copy it, read it, or write it (to minimally fragmented areas of available space) you will get nearly identical results to the sequential I/O benchmarks.
Additionally Mac OS and it's file system do not seem to fragment much at all like Windows does so this may not ever even be a concern at all. I recently defragged files and the volume on a 3-drive 1TB RAID0 that had never been defragged in 3 years and which was about 75% full. There were ~12,000 fragmented files out of 4,031,455 total files. That's like 0.003% file fragmentation.
And the volume fragmentation while split into 2 distinct halves at either end of the volume was very very good with just about the entire 25% free-space all continuously empty - even though it was all in the middle. Hello,
Question that kind'a got lost in the debate about SSDs:
I've opened up my optical bay to be able to see what I need to fit 2 HD in there. I found noise suppressing brackets at OWC, but I still can't find the Y-cable. Can anyone point me in the right direction?
Thanks
Loa
Question that kind'a got lost in the debate about SSDs:
I've opened up my optical bay to be able to see what I need to fit 2 HD in there. I found noise suppressing brackets at OWC, but I still can't find the Y-cable. Can anyone point me in the right direction?
Thanks
Loa
Maybe you'll like this better.
http://www.maxupgrades.com/istore/index.cfm?fuseaction=Product.display&product_id=187
Hello,
Thx for the link, but I wonder what the benefits of such a system over a simple double mount (25$) and split cable.
Or maybe I can't simply use a Y cable?
Loa
Yes, provided you remove the Optical drive.
http://eshop.macsales.com/item/Noiseblocker/HDC65/
This is only to physically mount the 2 drives in one of the 2 optical bays. The SATA connections still need to be completed, and that's currently the problem I need to solve. One connection hangs there in the second optical bay, but if you want to have 2, I don't know how.
Loa
Here's a simple one from Wiki. But if you check out NAND flash data sheets, it's usually listed. IIRC, it's been buried in some SSD articles on Tom's and AnandTech as well.
SSD = 128k
HDD = 4k
But on HDD's, the UBE is calculated to the bit, rather than an entire sector with BER.
1 vs. 128k...Hmm...
SSDs are using different flash tech than CompactFlash cards or is it the same? If it's the same as CF cards and that's what we're talking about then OMG! For data integrity I think a cucumber is a better storage device than CF... Well close!
Same!SSDs are using different flash tech than CompactFlash cards or is it the same? If it's the same as CF cards and that's what we're talking about then OMG! For data integrity I think a cucumber is a better storage device than CF... Well close!
CF started with NOR based flash, but has since moved to NAND for higher densities. So it does use the same technology these days.
Same!
CF started with NOR based flash, but has since moved to NAND for higher densities. So it does use the same technology these days.
WOW! Now that information puts me off SSDs 100% and completely!
Let's start taking bets on how many months it will be till we start seeing threads posted almost daily named: "HELP! SSD lost 1/2 my files and it's IMPOSSIBLE to recover.
" LOL... seriously.We'll see them at some point.WOW! Now that information puts me off SSDs 100% and completely!
Let's start taking bets on how many months it will be till we start seeing threads posted almost daily named: "HELP! SSD lost 1/2 my files and it's IMPOSSIBLE to recover.
I'll wait until the flash technology is upgraded to something more in line with reliable use in a drive. That's why I've mentioned FeRAM. It's already capable of 1E16 or better write UBE.
It needs a little more time for capacity (8Mb on a 130nm process) and cost though, and there's some competition. PMC, MRAM, PCM, and a couple of others.
We'll see them at some point.
I'll wait until the flash technology is upgraded to something more in line with reliable use in a drive. That's why I've mentioned FeRAM. It's already capable of 1E16 or better write UBE.
Yeah, when I said "completely 100%" I meant as it is currently - NAND or whatever. It's good to hear there's competition and readied techs to replace that awful stuff. Who knows what it'll be huh? Maybe they'll even have a breakthrough in microfluidic bubble tech.
That would be awesome!⤴