You can build a 48-core Opteron server with 512GB of RAM for under $8000. Going over 512GB in a single server gets a lot more expensive (you either need expensive high-density modules or expensive 8-socket servers - or both) but if you can run some sort of cluster that's not a problem.
It's a mix. We use enterprise drives for the really heavy stuff, and mainstream drives for data that's either read-only, read-mostly, or is in a database that does sequential writes like TokuDB or Cassandra.
We've been using SSDs in our servers since late 2008, starting with Fusion-io ioDrives and Intel drives since then - X25-E and X25-M, then 320, 520 and 710, and now planning to deploy a stack of S3700 and S3500 drives. Our main cluster of 10 servers has 24 SSDs each, we have another 40 drives on a dedicated search server, and smaller numbers elsewhere.
What we've found:
* Read performance is consistently brilliant. There's simply no going back. * Random write performance on the 710 series is not great (compared to the SLC-based X25-E or ioDrives), and sustained random write performance on the mainstream drives isn't great either, but a single drive can still outperform a RAID-10 array of 15k rpm disks. The S3700 looks much better, but we haven't deployed them yet. * SSDs can and do die without warning. One moment 100% good, next moment completely non-functional. Always use RAID if you love your data. (1, 10, 5, or 6, depending on your application.) * Unlike disks, RAID-5 or 50 works pretty well for database workloads. * We have noted the leading edge of the bathtub curve (infant mortality), but so far, no trailing edge as older drives start to wear out. Once in place, they just keep humming along. * That said, we do match drives to workloads - SLC or enterprise MLC for random write loads (InnoDB, MongoDB) and MLC for sequential write/random read loads (TokuDB, CouchDB, Cassandra).
We have hundreds of SSDs in production servers. We couldn't survive without them. For heavy database workloads, they are the silver bullet to I/O problems, so much so that running a database on regular disk has become almost unimaginable. Why would you even try to do that?
Two years after buying this machine, Apple will release a newer version. The newer version will be so much better (faster bus, etc), that the older one will be left in the dust and on ebay for $499.
Three year old Mac Pro: Up to 12 cores, 64GB RAM.
Not even released yet Mac Pro: Up to 12 cores, 64GB RAM.
People don't seem to realize that Thunderbolt is external PCIe, it is not USB or Firewire or SATA. 6 ports gives you 6 PCIe 4x slots at 20Gbs which places it between PCIe 3 and 4
One PCIe 3 x16 slot is 128GT/s using 128b/130b encoding (i.e. data transfer can use >98% of the theoretical bandwidth). So ~126Gbps.
Six Thunderbolt 2 connectors are 6 x 20GT/s using 8b/10b encoding (80% efficient), so 96Gbps.
Well, points for learning from the best. And to be fair, from the brief look I got before their site went offline, they provide a more in-depth simulation of the game development process. Game Dev Story is charming but pretty simplistic.
Chiropractic is bogus. It wasn't designed to correct posture or "force tight muscles to release", it was designed to cure completely unrelated illnesses. Which it doesn't do. Any success it may have in treating minor back pain is accidental.
They didn't use the gas directly; they dissolved it in water, at a concentration of 1 part per billion (ppb). 5ppb concentration in air is detectable by smell; 10 ppm, i.e. 10,000ppb, is the safety limit for extended exposure set by OSHA. So the concentrations used in the experiment are quite safe. That said, in higher concentrations it is seriously nasty.
SoftLayer have a history of being very aggressive on DMCA takedown orders - you may get as little as four hours to remove the content or have your server taken offline. Even if it's a shared server hosting many accounts.
So, 4 cores at 2.5-3.5GHz, 384 shaders, and dual-channel DDR3-1866 RAM at 35W.
If AMD were to double everything they'd have a really nice 70W desktop chip. Not sure what the die size is for Richland, so a doubled chip might not be cost-effective - though the PS4 APU has 8 cores and 1152 shaders, so it's at least possible.
Likewise for our database clusters. We use open-source software (MongoDB, Redis, Riak) so hardware cost matters - if you use Oracle or something like that, software costs dominate. If you want large 4-socket servers, AMD offers much better value than Intel. And if you want lots of small 1S servers, AMD wins again, because the E3 Xeons only support 32GB RAM.
DARPA runs a lot of these research seed programs, putting a couple of million dollars into a bunch of different but related research projects. In this case the program budget is $100 million in total, and Continuum got $3 million for their Python work (Numba, Blaze, etc). Some of the program money may have gone to R as well; there's a couple of dozen research groups, but I don't have a full list.
I've only spent half a day with it, but I'd have to say, no. It's as resource-intensive as ever, the UI still freezes intermittently when there's any processing going on at all, and it crashed completely when trying to play a short video.
On the other hand, it managed to resurrect my copy of Dr Horrible's Sing-Along Blog which iTunes 10 ate. (The player said I hadn't downloaded it and wouldn't play, and the store said that I had downloaded it and wouldn't download. iTunes11 re-downloaded it for me. And crashed a short while later.)
In contrast, there was 9/11 - sure, most people think it was the Saudis, but there are too many questions unanswered, like the lack of debris, lack of video, lack of an airplane at the Pentagon
Lack of debris? Lack of airplane? If you believe that, I'd seek a second opinion if you said the sky is blue.
Piledriver is the architecture, like Intel's Ivy Bridge is the architecture.
These are server chips. Best case, these are finally faster than their pre-Bulldozer parts in real, consumer desktop use. They will not beat an 8 core Sandy Bridge Xeon in FP-heavy applications, and power consumption is, at best, on the same level as the Xeons.
That's true. A 16-core Opteron has the same FP width as an 8-core Xeon, and a higher TDP for a given clock.
On the other hand, we buy almost all AMD because it lets us build cheap 1U or 2U 4-socket servers with 512GB of RAM each. 4-socket Intel chips (E5-4600 or E7) are much more expensive; mid-range servers work out to 50% more for Intel, and high-end servers about 80% more for equivalent speed.
I'm not saying that the United States in general and the Bay Area in particular don't have their foibles - like every other place humans have ever lived. But at least those people don't post vacuous twaddle on Slashdot bemoaning the blighted state of the souls of entire populations.
Slashdot Mirror
Because if write wear is the prime failure mode and you're running RAID, you're likely to lose multiple SSDs in a relatively short interval.
You can build a 48-core Opteron server with 512GB of RAM for under $8000. Going over 512GB in a single server gets a lot more expensive (you either need expensive high-density modules or expensive 8-socket servers - or both) but if you can run some sort of cluster that's not a problem.
It's a mix. We use enterprise drives for the really heavy stuff, and mainstream drives for data that's either read-only, read-mostly, or is in a database that does sequential writes like TokuDB or Cassandra.
We've been using SSDs in our servers since late 2008, starting with Fusion-io ioDrives and Intel drives since then - X25-E and X25-M, then 320, 520 and 710, and now planning to deploy a stack of S3700 and S3500 drives. Our main cluster of 10 servers has 24 SSDs each, we have another 40 drives on a dedicated search server, and smaller numbers elsewhere.
What we've found:
* Read performance is consistently brilliant. There's simply no going back.
* Random write performance on the 710 series is not great (compared to the SLC-based X25-E or ioDrives), and sustained random write performance on the mainstream drives isn't great either, but a single drive can still outperform a RAID-10 array of 15k rpm disks. The S3700 looks much better, but we haven't deployed them yet.
* SSDs can and do die without warning. One moment 100% good, next moment completely non-functional. Always use RAID if you love your data. (1, 10, 5, or 6, depending on your application.)
* Unlike disks, RAID-5 or 50 works pretty well for database workloads.
* We have noted the leading edge of the bathtub curve (infant mortality), but so far, no trailing edge as older drives start to wear out. Once in place, they just keep humming along.
* That said, we do match drives to workloads - SLC or enterprise MLC for random write loads (InnoDB, MongoDB) and MLC for sequential write/random read loads (TokuDB, CouchDB, Cassandra).
We have hundreds of SSDs in production servers. We couldn't survive without them. For heavy database workloads, they are the silver bullet to I/O problems, so much so that running a database on regular disk has become almost unimaginable. Why would you even try to do that?
Two years after buying this machine, Apple will release a newer version. The newer version will be so much better (faster bus, etc), that the older one will be left in the dust and on ebay for $499.
Three year old Mac Pro: Up to 12 cores, 64GB RAM.
Not even released yet Mac Pro: Up to 12 cores, 64GB RAM.
So... No.
People don't seem to realize that Thunderbolt is external PCIe, it is not USB or Firewire or SATA. 6 ports gives you 6 PCIe 4x slots at 20Gbs which places it between PCIe 3 and 4
One PCIe 3 x16 slot is 128GT/s using 128b/130b encoding (i.e. data transfer can use >98% of the theoretical bandwidth). So ~126Gbps.
Six Thunderbolt 2 connectors are 6 x 20GT/s using 8b/10b encoding (80% efficient), so 96Gbps.
The bad ones have "Are you sure?" The awful ones have "Don't ask this again." The good ones have "Undo."
Well, points for learning from the best. And to be fair, from the brief look I got before their site went offline, they provide a more in-depth simulation of the game development process. Game Dev Story is charming but pretty simplistic.
Chiropractic is bogus. It wasn't designed to correct posture or "force tight muscles to release", it was designed to cure completely unrelated illnesses. Which it doesn't do. Any success it may have in treating minor back pain is accidental.
They didn't use the gas directly; they dissolved it in water, at a concentration of 1 part per billion (ppb). 5ppb concentration in air is detectable by smell; 10 ppm, i.e. 10,000ppb, is the safety limit for extended exposure set by OSHA. So the concentrations used in the experiment are quite safe. That said, in higher concentrations it is seriously nasty.
The problem was, it didn't actually answer the questions.
SoftLayer have a history of being very aggressive on DMCA takedown orders - you may get as little as four hours to remove the content or have your server taken offline. Even if it's a shared server hosting many accounts.
They're releasing the laptop chips first.
Socket 2011 has quad-channel memory off one die, and AMD's G34 server socket has quad-channel off two. Again, it's definitely possible.
With 768 shaders off just dual-channel DDR3 you'd be seriously bandwidth-starved, so I don't see it working well without four memory channels.
So, 4 cores at 2.5-3.5GHz, 384 shaders, and dual-channel DDR3-1866 RAM at 35W.
If AMD were to double everything they'd have a really nice 70W desktop chip. Not sure what the die size is for Richland, so a doubled chip might not be cost-effective - though the PS4 APU has 8 cores and 1152 shaders, so it's at least possible.
Likewise for our database clusters. We use open-source software (MongoDB, Redis, Riak) so hardware cost matters - if you use Oracle or something like that, software costs dominate. If you want large 4-socket servers, AMD offers much better value than Intel. And if you want lots of small 1S servers, AMD wins again, because the E3 Xeons only support 32GB RAM.
DARPA runs a lot of these research seed programs, putting a couple of million dollars into a bunch of different but related research projects. In this case the program budget is $100 million in total, and Continuum got $3 million for their Python work (Numba, Blaze, etc). Some of the program money may have gone to R as well; there's a couple of dozen research groups, but I don't have a full list.
Both. The prebuilt "Anaconda" distro defaults to Python 2.7, but it also works with 3.3 and 2.6.
I've only spent half a day with it, but I'd have to say, no. It's as resource-intensive as ever, the UI still freezes intermittently when there's any processing going on at all, and it crashed completely when trying to play a short video.
On the other hand, it managed to resurrect my copy of Dr Horrible's Sing-Along Blog which iTunes 10 ate. (The player said I hadn't downloaded it and wouldn't play, and the store said that I had downloaded it and wouldn't download. iTunes11 re-downloaded it for me. And crashed a short while later.)
Great idea. Wrong implementation. There are many pitfalls with making science degrees cheaper, like for example what happens when you switch majors?
After you have a couple of semesters of credits in science and math? Mission accomplished.
In contrast, there was 9/11 - sure, most people think it was the Saudis, but there are too many questions unanswered, like the lack of debris, lack of video, lack of an airplane at the Pentagon
Lack of debris? Lack of airplane? If you believe that, I'd seek a second opinion if you said the sky is blue.
Piledriver is the architecture, like Intel's Ivy Bridge is the architecture.
These are server chips. Best case, these are finally faster than their pre-Bulldozer parts in real, consumer desktop use. They will not beat an 8 core Sandy Bridge Xeon in FP-heavy applications, and power consumption is, at best, on the same level as the Xeons.
That's true. A 16-core Opteron has the same FP width as an 8-core Xeon, and a higher TDP for a given clock.
On the other hand, we buy almost all AMD because it lets us build cheap 1U or 2U 4-socket servers with 512GB of RAM each. 4-socket Intel chips (E5-4600 or E7) are much more expensive; mid-range servers work out to 50% more for Intel, and high-end servers about 80% more for equivalent speed.
I'm not saying that the United States in general and the Bay Area in particular don't have their foibles - like every other place humans have ever lived. But at least those people don't post vacuous twaddle on Slashdot bemoaning the blighted state of the souls of entire populations.
Also, souls don't exist.
So what you're saying is nothing in his comment meant enough to reflect on, so you flicked it away with an "it's human nature" bromide?
Yes.