|
HalloKitty posted:8350 is OK to a certain degree, it seems, but is still ridiculous in terms of power consumption, and has almost no overclocking headroom on air unlike the Intel gear. 8350 is Piledriver, and to say the bulldozer didn't have any OC potential on air is complete bull. 5GHz (well 4.97) on air using a Zalman CNPS 9900 and Artic MX-4 on an ASUS Sabertooth 990FX. Idle is 29c load is 56c with an 8120 bulldozer - all 8 cores still enabled. It has a lot of headroom, mine had a 3.2ghz base clock for example. For what I payed for the board and processor I definitely got my monies worth (245$ on sale), and would not have done it any other way. How did you end up with no clue about what you are talking about in the first place? (USER WAS PUT ON PROBATION FOR THIS POST)
|
#
?
Oct 27, 2012 17:54
|
|
|
|
| # ? May 11, 2024 17:40 |
|
|
Foil posted:8350 is Piledriver, and to say the bulldozer didn't have any OC potential on air is complete bull. 5GHz (well 4.97) on air using a Zalman CNPS 9900 and Artic MX-4 on an ASUS Sabertooth 990FX. Idle is 29c load is 56c with an 8120 bulldozer - all 8 cores still enabled. http://www.anandtech.com/show/6396/the-vishera-review-amd-fx8350-fx8320-fx6300-and-fx4300-tested/8 I'm confused - I never said it had no OC potential (after all, the highest current OC record is on Bulldozer with liquid nitrogen), and I wasn't commenting on oc'ing the 8120. I was just going by review numbers for Piledriver - and those load numbers look pretty brutal still, which is meaningful when using the machine for constant loads such as those with video encoding. I didn't even say Piledriver was bad - it has potential in certain loads, and AMD is headed in the right direction. It would certainly be a good drop-in replacement for an older Bulldozer chip. HalloKitty fucked around with this message at 18:18 on Oct 27, 2012 |
|
#
?
Oct 27, 2012 18:15
|
|
|
HalloKitty posted:8350 is OK to a certain degree, it seems, but is still ridiculous in terms of power consumption, and has almost no overclocking headroom on air unlike the Intel gear. I had a decent Gateway desktop with an i7 860, but I somehow fried the motherboard when I tried charging my bluetooth speaker phone in the USB port. I had just replaced the harddrive and Power Supply, so I found a cheap desktop with the FX 6100 and bought it without really looking into it that much.
|
|
#
?
Oct 27, 2012 18:23
|
|
|
Mill Village posted:I had a decent Gateway desktop with an i7 860, but I somehow fried the motherboard when I tried charging my bluetooth speaker phone in the USB port. I had just replaced the harddrive and Power Supply, so I found a cheap desktop with the FX 6100 and bought it without really looking into it that much. Ah fair enough, I've seen some good deals on AMD based systems recently. In which case as long as the PSU handles it, the 8350 would provide a good upgrade path. It does seem to do especially well in video encoding as per the AnandTech review. HalloKitty fucked around with this message at 18:29 on Oct 27, 2012 |
|
#
?
Oct 27, 2012 18:24
|
|
|
Foil posted:How did you end up with no clue about what you are talking about in the first place? A living, breathing AMD fanboy? Didn't y'all go extinct around '09?
|
|
#
?
Oct 28, 2012 18:56
|
|
|
JawnV6 posted:A living, breathing AMD fanboy? Didn't y'all go extinct around '09? You don't have to be a fanboy to find places where the product works well, in this case, the highly multithreaded world of VMware.
|
|
#
?
Oct 28, 2012 19:02
|
|
|
adorai posted:When we did our recent VMware refresh, we nearly went with bulldozer. We had a single application that had terrible performance on the AMD proc though, and given it's importance we spent the extra and went with Intel. With bulldozer the aggregate performance in our testing exceeded what we are getting with the E5s we bought, for about 20% less for the whole platform. Analysing specific use cases ≠ being a fanboy
|
|
#
?
Oct 28, 2012 19:04
|
|
|
Debating if I should jump onto the FX-8320 I currently have a T1055 @ 3.2, seeing how the FX-8320 is 20 off and I am doing quite a bit of heavy vmware work, it might be a nice upgrade 2 more cores and a bit more L2/L3 cache. I am tossing this up against just getting some Xeon E5-2620, I am sure the E5 wins however it is much more expensive. Does look like a nice improvement... http://www.anandtech.com/bench/Product/698?vs=147
|
|
#
?
Oct 29, 2012 02:44
|
|
|
AMD expected to unveil ARM-based server on Monday. (via SeaMicro) Will definitely be watching this closely, I've been watching ARM server news pretty closely and there have been a long line of CEO's (EMC the biggest one) saying they'll never get market penetration. I'm incredibly curious to see benchmarks and the price point it comes in at.
|
|
#
?
Oct 29, 2012 03:22
|
|
|
Re: above, AMD moving ARM-based server CPUs to production by 2014, with some details on planned market segmentation and a bit of high-level backbone design.
|
|
#
?
Oct 29, 2012 23:12
|
|
|
I'm gonna stick this here because there isn't really a better place right now: AnandTech got a reporter in to cover the upgrade of the Oak Ridge National Laboratory's Jaguar supercomputer (18,688 12-core Opterons) into the Titan (18,688 16-core Opterons plus 18,688 Nvidia Tesla K20 coprocessors). The Jaguar --> Titan upgrade actually provides this 33% x86 core increase and roughly 50 million FP32 units within the same power budget, which is good because that budget is already as large as practicable at 9 megawatts. The RAM has also been upgraded, at 32 GB DDR3 and 6 GB of GDDR5 per node for a total of 710 TB of RAM. Cray provided the hardware, but they didn't act as a full integrator, so the ORNL's poor saps had to install 18,688 CPUs and 18,688 GPUs by hand:  Read the article; it's super awesome and only gets better.
|
|
#
?
Oct 31, 2012 13:20
|
|
|
But isn't that from 12 cores to 8 bulldozer modules? not even clocked very high - I can't imagine single thread performance budged at all. The real upgrade is the huge amount of GPUs. Sad AMD couldn't get in on that action, since GCN is powerful in compute. Wonder what they do with all those old 12 core opterons? Would still make a hell of a datacentre!
|
|
#
?
Oct 31, 2012 14:33
|
|
|
HalloKitty posted:But isn't that from 12 cores to 8 bulldozer modules? not even clocked very high - I can't imagine single thread performance budged at all. The real upgrade is the huge amount of GPUs. Sad AMD couldn't get in on that action, since GCN is powerful in compute. Cray has a deal set up to take care of the old hardware.
|
|
#
?
Oct 31, 2012 15:04
|
|
|
HalloKitty posted:But isn't that from 12 cores to 8 bulldozer modules? not even clocked very high - I can't imagine single thread performance budged at all. The real upgrade is the huge amount of GPUs. Sad AMD couldn't get in on that action, since GCN is powerful in compute. Supercomputers give absolutely no fucks about single-threaded performance*. And, by all accounts, Bulldozer-based designs can perform well under certain circumstances. For you and me, that doesn't matter all that much; we're not going to be rewriting off-the-shelf software packages to target our hardware. For a giant supercomputer in a lab where even the janitors probably have comp sci doctorates, that equation shifts somewhat. *well, ok, it's not quite that clear cut. Per-core and per-thread performance is always nice. But, raw parallel power is the main goal.
|
|
#
?
Oct 31, 2012 16:10
|
|
|
The article addresses CPU choice a bit: that's just what the original Jaguar system used, since they were the better choice at its inception in 2005, and the Cray XK line has kept AMD offerings since. Nvidia K20s are used instead of Southern Islands for roughly the same reason: the decision to use GPGPU acceleration was made three years ago, and only Nvidia had real GPGPU chops and roadmaps at the time, so that's who Cray contracted with.
|
|
#
?
Oct 31, 2012 16:46
|
|
|
Tom's ran some tests to see if Windows 8 ran any better on Piledriver than Windows 7: 
|
|
#
?
Oct 31, 2012 17:59
|
|
|
HalloKitty posted:But isn't that from 12 cores to 8 bulldozer modules? not even clocked very high - I can't imagine single thread performance budged at all. When you've got a cluster of 299,008 cores, if you ever care about the performance of a single thread, you're doing it wrong.
|
|
#
?
Oct 31, 2012 19:20
|
|
|
Zhentar posted:When you've got a cluster of 299,008 cores, if you ever care about the performance of a single thread, you're doing it wrong. Oh I realise the single-threaded performance isn't the focus, but rather I'm wondering what the overall effect of that CPU upgrade is. Edit: Top 500 has the newest configuration, it seems: http://i.top500.org/system/176544 But the previously submitted configuration used 2.6GHz 6-cores. Looks like this beast has had plenty of upgrades. HalloKitty fucked around with this message at 19:49 on Oct 31, 2012 |
|
#
?
Oct 31, 2012 19:38
|
|
|
If anyone is wondering why the Top500 CPU core count differs from what we've been talking about here, some of the CPUs are dedicated solely to OS and hardware interrupts, so that those interrupts aren't constantly cascading out and hiccuping the entire machine. E: The ranks above the Oak Ridge computers are, in order: 5) 186k Westmere Xeon cores @ 2.93 GHz with Fermi Tesla, 22% less power. 4) 146k Sandy Bridge Xeon cores @ 2.7 GHz with 8.2k Westmere cores, no GPGPU, 33% less power 3 through 1) IBM Power and SPARC with ridiculous core count per watt It looks like the Linpack and power rankings only count main cores, not coprocessors like the Tesla gear. Factory Factory fucked around with this message at 20:11 on Oct 31, 2012 |
|
#
?
Oct 31, 2012 19:57
|
|
|
Factory Factory posted:If anyone is wondering why the Top500 CPU core count differs from what we've been talking about here, some of the CPUs are dedicated solely to OS and hardware interrupts, so that those interrupts aren't constantly cascading out and hiccuping the entire machine. Yeah, that's super critical for HPC and real-time applications. We apply a similar kernel patch to our machines that keep the OS on 1-2 cores, and leave the rest open for scheduling / usage by the real-time co-kernel. Nothing like trashing your realtime performance because some OS task with good intentions pops up for attention. I haven't fully understood the voodoo that goes to make that happen, but I'm sure that voodoo + no real SW department is why we backport poo poo to Linux 2.6.24 
|
|
#
?
Oct 31, 2012 19:59
|
|
|
movax posted:Yeah, that's super critical for HPC and real-time applications. We apply a similar kernel patch to our machines that keep the OS on 1-2 cores, and leave the rest open for scheduling / usage by the real-time co-kernel. Nothing like trashing your realtime performance because some OS task with good intentions pops up for attention. I have no sense of scale of the CPU load for thread scheduling or anything is for massive systems like this; could you compare it to a more typical desktop load? Also, congrats on your new mod status.
|
|
#
?
Oct 31, 2012 20:20
|
|
|
We've come a long way from supercomputers that used a scant 9,000 Intel Pentium Pros edit: Maybe, maybe not - sure, the cores have gone way up but there's still only double the amount of CPU's Bob Morales fucked around with this message at 20:26 on Oct 31, 2012 |
|
#
?
Oct 31, 2012 20:23
|
|
|
unpronounceable posted:I have no sense of scale of the CPU load for thread scheduling or anything is for massive systems like this; could you compare it to a more typical desktop load? Thanks! Ours is actually somewhat close to a desktop load; a single one of our boards generally has a quad-core Opteron on it (can be expanded if needed though). We foist Linux kernel + various system tasks onto Core 0 and leave the other cores free for a real-time co-kernel like Xenomai which is designed for real-time use. Real-time OSes have very strong concepts of task priority, which is important when you need your tasks to run in a deterministic manner. I.E., I want to make sure my high-priority Task A gets all the attention he deserves, but slightly-lower priority tasks B and C are welcome to CPU time if A doesn't need it. Remember that even on our awesome dual-core/quad-core CPUs, your OS is still busy furiously switching between all the myriad tasks making sure they keep doing their thing. Windows/OS X/Linux (stock) aren't real-time and don't give two fucks about guaranteeing latency/timing for a task. On said real-time system I might have a FTP daemon, and the regular Linux housekeeping tasks running. These aren't performance critical, but I don't want vsftpd (FTP server) stealing cycles at a bad time, so I try to isolate all that to core 0, leaving 1-3 open for work. So long story short, you shove (the best you can) all that poo poo onto some number of cores, and leave the rest open for usage/scheduling by something a team of CS geniuses wrote.
|
|
#
?
Oct 31, 2012 21:01
|
|
|
movax posted:Is a moderator now! This has been a good year for new blue stars.
|
|
#
?
Nov 1, 2012 02:07
|
|
|
Factory Factory posted:The article addresses CPU choice a bit: that's just what the original Jaguar system used, since they were the better choice at its inception in 2005, and the Cray XK line has kept AMD offerings since. GPU: AMD never had the software strategy to actually go after this market. It's a lot harder than "build a driver, call it a day," (support for third party debuggers and profilers, out-of-band monitoring, dealing nicely with exotic interconnects, etc--lots of stuff that never shows up at all in the non-HPC side of things) and they simply never staffed OpenCL/compute up enough to ever show up in these bids. Xeon Phi is the first real competing accelerator that NV has had in the HPC space. Professor Science fucked around with this message at 04:30 on Nov 1, 2012 |
|
#
?
Nov 1, 2012 04:24
|
|
|
Yeah, it needs to be stressed that the primary reason for going AMD years ago was because of Hypertransport beating the pants off Intel's interconnect since that's such a big factor in large-scale HPC workloads (the joke is that supercomputers turn cpu-bound tasks into I/O bound ones)
|
|
#
?
Nov 1, 2012 06:32
|
|
|
At the time it was easier to implement your own HT devices as well; the University of Mannheim developed a workable HT IP core that you could implement on FPGAs with the appropriate transceivers. You enjoy some pretty low latency when your I/O enjoys a HT link to an Opteron. Intel was rather quiet about QPI at the time, I think. It's unfortunate that even right now, AMD is still selling three-chip solutions in the server market
|
|
#
?
Nov 1, 2012 06:39
|
|
|
necrobobsledder posted:Yeah, it needs to be stressed that the primary reason for going AMD years ago was because of Hypertransport beating the pants off Intel's interconnect since that's such a big factor in large-scale HPC workloads (the joke is that supercomputers turn cpu-bound tasks into I/O bound ones) the only equivalent device that I know of that sits on QPI is the interconnect on the SGI UV machines.
|
|
#
?
Nov 1, 2012 06:51
|
|
|
Professor Science posted:no no, this isn't a QPI or FSB versus HT performance/multi-socket thing (although I'm sure that didn't hurt at the time)--Gemini is actually a device that hangs off the HyperTransport bus. besides the occasional Torrenza FPGA, Gemini is the only device I know of on HT like this. Me and my friend Bob (he was working on the GPGPU side of the house) in 2003 were looking for a high-performance bus to solve both our I/O problems on COTS HPC efforts and wound up with Hypertransport being literally the only thing that could suffice for anything that could call itself COTS HPC (although he was quite peeved given there was no HT-based video card available). I thought that the FPGA nodes wouldn't have made it in after the disappointing results I showed but apparently someone thought it was worth it, hrm. Unfortunately, it's highly doubtful that going forward AMD will see any more of that sweet, sweet DoD / DoE HPC money because Intel's going all-out for Aries apparently and owns Infiniband tech outright. QPI is being skipped entirely for Infiniband it seems. More bad AMD news in the AMD thread? Par for the course. unpronounceable posted:I have no sense of scale of the CPU load for thread scheduling or anything is for massive systems like this; could you compare it to a more typical desktop load?  . Sort of like how Google's clusters encounter constant hardware failures at all times, this sort of commercial cluster has similar problems to deal with. Think of it like ECC for entire machines.
|
|
#
?
Nov 1, 2012 15:07
|
|
|
On another topic: Aren't all the new next-generations consoles using AMD platform chipsets? I wouldn't count them on going out of business anytime soon.
|
|
#
?
Nov 9, 2012 02:11
|
|
|
Job Truniht posted:On another topic: Aren't all the new next-generations consoles using AMD platform chipsets? I wouldn't count them on going out of business anytime soon. There has been no consistent proof of this. There's been stuff released of next-xbox devkits that had nvidia and intel chips instead.
|
|
#
?
Nov 9, 2012 02:23
|
|
|
AMD's not going any lower right? I'm safe to go long on them now? 
|
|
#
?
Nov 9, 2012 02:36
|
|
|
As far as I am aware, we have confirmation that all three next-gen consoles use AMD graphics, but the CPU portion is still up in the air.
|
|
#
?
Nov 9, 2012 02:40
|
|
|
I'd bet you that AMD management is grooming the company for sale. I'd be surprised if they aren't acquired/merged in the next 2 years.
|
|
#
?
Nov 9, 2012 04:28
|
|
|
Star War Sex Parrot posted:AMD's not going any lower right? I'm safe to go long on them now? I remember when their stock first fell to $7, tons of pretend stock market gurus on hardware boards were telling everyone to buy it up because that was a bargain price and had no where to go but up 
|
|
#
?
Nov 9, 2012 06:36
|
|
|
http://allthingsd.com/20121113/amd-exploring-options-including-breakup-sale/quote:Shares of the chipmaker Advanced Micro Devices are up by more than 14 percent on a report by Reuters saying that the company has hired J.P. Morgan to “explore options” that could include a disposition of its patents or an outright sale. Well things seem to keep getting better for AMD. I'm assuming they are looking to sell off some of their assets. Wonder how much longer they can keep going at this rate WhyteRyce fucked around with this message at 05:04 on Nov 15, 2012 |
|
#
?
Nov 15, 2012 04:50
|
|
|
canyoneer posted:I'd bet you that AMD management is grooming the company for sale. I'd be surprised if they aren't acquired/merged in the next 2 years. I can't think of anyone who would want them at this point, unless the x86 license was part of the deal. I sort of have to assume it is not, because their market cap is only $1.3B with $2B in debt. (also, their long term debt has gone from $1.5B to $2B last quarter, which is the exact opposite of what companies looking to be bought out usually do) Chuu fucked around with this message at 09:09 on Nov 15, 2012 |
|
#
?
Nov 15, 2012 09:04
|
|
|
Chuu posted:I can't think of anyone who would want them at this point, unless the x86 license was part of the deal. I sort of have to assume it is not, because their market cap is only $1.3B with $2B in debt. Apple has been developing their own chips for a while now (for mobile devices anyways), and were even rumored to have had an AMD MacBook Air prototype in the works a while back. Would they benefit anything from purchasing AMD?
|
|
#
?
Nov 15, 2012 11:00
|
|
|
If AMD still had fabs, probably.
|
|
#
?
Nov 15, 2012 11:09
|
|
|
|
| # ? May 11, 2024 17:40 |
|
|
Edit: I can't read.
|
|
#
?
Nov 15, 2012 12:42
|
|
