Register a SA Forums Account here!
JOINING THE SA FORUMS WILL REMOVE THIS BIG AD, THE ANNOYING UNDERLINED ADS, AND STUPID INTERSTITIAL ADS!!!

You can: log in, read the tech support FAQ, or request your lost password. This dumb message (and those ads) will appear on every screen until you register! Get rid of this crap by registering your own SA Forums Account and joining roughly 150,000 Goons, for the one-time price of $9.95! We charge money because it costs us money per month for bills, and since we don't believe in showing ads to our users, we try to make the money back through forum registrations.
 
  • Post
  • Reply
Alereon
Feb 6, 2004

Dehumanize yourself and face to Trumpshed

College Slice

JawnV6 posted:

Sorry, don't have time to learn your new language
"Plain broken" wasn't a comment on the bug being easy or difficult to find, but on the fact that the feature got disabled so customers no longer derive any value from it. What matters is the impact of the bug, if it could get a footnote in an errata document I wouldn't care, but they had to disable TSX so it's a huge fuckup.

wargames posted:

This affects my 4690?
Yes. Though really almost nobody has a use for TSX right now, so this only matters in the abstract sense that a part of the product you paid for is broken, and that now it won't become useful in the future. There's always a lag between when new instruction set features are implemented and when they start getting used, and now that lag will be longer.

Adbot
ADBOT LOVES YOU

Alereon
Feb 6, 2004

Dehumanize yourself and face to Trumpshed

College Slice

SlayVus posted:

It's interesting to note that the 5820k is actually performing better in gaming benchmarks than the other two CPUs. I wonder why that is. (This is looking at Tom's testing).
No other sites are seeing similar behavior (Anandtech, Guru3D so I think Tom's Hardware messed up. They have a pretty bad reputation for the quality of their reviews.

Alereon
Feb 6, 2004

Dehumanize yourself and face to Trumpshed

College Slice

Combat Pretzel posted:

The Ivy Bridge-E CPUs ahead of the Haswell-E in the CPU benchmarks, at 10W less TDP? What am I missing?
Are you looking at the clockspeeds versus core count of the models you're comparing?

Alereon
Feb 6, 2004

Dehumanize yourself and face to Trumpshed

College Slice

The Last Poet posted:

I'm looking to build a new PC soonish and for some reason i want one of the Samsung M.2 drives in it. I also want SLI. Would i need the 5830 or would the 5820 be sufficient (wrt pcie lanes) ?
Why are you not getting a 4790K?

Alereon
Feb 6, 2004

Dehumanize yourself and face to Trumpshed

College Slice

BurritoJustice posted:

A lot of this is wrong. The XP941 is 3.0, it just wasn't marketed as such because until ASRock did 3.0 M.2 it wasn't an option. See Anandtech's review of the Extreme6 specifically addressing this.
The Samsung XP941 is not PCI-Express 3.0

That review clearly states the drive is PCIe 2.0 x4. The point of that review was that most Z97 boards don't have four lanes available from the chipset, so if you want to dedicate four lanes to the M.2 slot, you need to use PCIe 3.0 lanes coming from the CPU.

More broadly, the Samsung XP941 is not a very good SSD for consumers compared to the options that are currently available. Remember that it doesn't support any of the important perfomance-enhancing features of upcoming consumer PCIe SSDs, and since it's an OEM-only product you lose support for all features provided by the Samsung software, most importantly RAPID Mode. For day-to-day usage RAPID Mode provides a very similar performance benefit to high-end PCIe SSDs, and you can get a Samsung 840 Evo 500GB for the price of a 256GB XP941 with money left over. If you don't think the 840 Evo is the drive for you for whatever reason, the 850 Pro still provides compelling benefits over the XP941 at a substantially lower price. More details are available in the SSD Megathread.

My point here is not that the XP941 sucks, but that it's not the straight upgrade over Samsung's consumer SATA SSDs that it might seem like at first glance. That straight upgrade will come with the next generation of drives. With the XP941 you're basically getting 8 channels of MLC on a bus that's twice as wide as SATA600. With a Samsung 840 Evo you're getting 8 channels of SLC (up to the size of the cache, 3-12GB) on a SATA600 bus, along with a memory caching and write combining layer that's designed to alleviate the bottlenecks from that bus. For most typical consumer workloads the Samsung 840 Evo actually works better, especially when you consider it's less than half the price so you can afford a much larger drive. It really is ridiculous how perfectly suited the Samsung 840 Evo is to desktop workloads.

Alereon fucked around with this message at 12:43 on Aug 31, 2014

Alereon
Feb 6, 2004

Dehumanize yourself and face to Trumpshed

College Slice

BurritoJustice posted:



To be clear, I'm not recommending the drive at all here, I just wish to be as factual as possible.


(I love your work in the SSD thread)
I am 99% sure that's a typo, as the review of the drive itself (as do others) says the controller is PCIe 2.0 x4, nowhere else in that review is it mentioned that the drive (versus slot) supports >2GB/sec, and the drive's performance doesn't exceed 1350MB/sec.

So the upshot of all this (so I'm not just being pedantic and telling people they're wrong!) is that you can get full performance out of a Samsung XP941 on a Z97 board if that was a thing you wanted to do by using the PCI-E 2.0 x4 slot provided by the chipset and an adapter card. This is usually your third PCI-E x16 slot on SLI-capable motherboards.

Alereon fucked around with this message at 13:03 on Aug 31, 2014

Alereon
Feb 6, 2004

Dehumanize yourself and face to Trumpshed

College Slice

Interestingly, Anand announced his retirement yesterday, Ryan Smith will now be running the site. I'm not sure I agree with criticisms of their GPU coverage, if you look at their recent reviews they include minimum framerate or framrate over time graphs for most games, and framerate variation graphs for some as well. I'm also not sure of the value of testing Crysis 3 beyond Medium quality at 4K, since you're basically just comparing how not-quite-playable it is on the two highest-end cards that would come the closest. I am a pretty big fan of Tech Report's GPU coverage for the nitty gritty of the kind of experience you can get though, the value of Anandtech's GPU articles is more about how the card actually works and performs rather than the experience in particular games.

Keep in mind also that AMD's APUs are usually pretty good values at launch, the problem comes when they don't evolve in time or when Intel aggressively improves the value of their products. AMD APUs were drat compelling in HTPCs before Intel introduced QuickSync to their Pentiums, and similarly prior to the Pentium Anniversary Edition you had to choose between a decent CPU or decent graphics, and AMD APUs gave you "good enough" for both. I think Anandtech's reviews of the FX series have been appropriately critical, only recommending them narrowly for heavily multi-threaded applications.

Alereon
Feb 6, 2004

Dehumanize yourself and face to Trumpshed

College Slice

Now for something more directly on-topic:

Does Haswell-E make sense for me?

The key benefits for Haswell-E are more cores, support for more PCI-Express 3.0 lanes in higher CPUs, and additional memory bandwidth, as well as support for more than 32GB of RAM (Haswell's memory controller is limited to 4 x 8GB DIMMs). Here are some example basic systems to compare:

High Haswell System:
ASUS ROG MAXIMUS VII HERO Z97 $206
G.Skill Ares DDR3-2133 16GB (2x8GB) $163
Core i7 4790K $340
Total: $709 for 8 x 4.0Ghz threads (4.4Ghz single-core), 16GB RAM, 16 PCIe 3.0 lanes

Max Haswell System:
ASUS Z97-WS $286
G.Skill Ares DDR3-2133 32GB (4x8GB) $326
Core i7 4790K $340
Total: $952 for 8 x 4.0Ghz threads (4.4Ghz single-core), 32GB RAM, 32 PCIe 3.0 lanes (16GB/sec shared)

Mini Haswell-E System:
Lower-end Asus board launching later this year ($300)
G.Skill RipJaws DDR4-2133 16GB (4x4GB) $260
Core i7 5820K $400
Total: $960 for 12 x 3.3Ghz threads (3.6Ghz single-core), 16GB RAM, 28 PCIe 3.0 lanes

Basic Haswell-E System:
Asus X99 Deluxe $399
G.Skill RipJaws DDR4-2133 32GB (4x8GB) $480
Core i7 5820K $400
Total: $1279 for 12 x 3.3Ghz threads (3.6Ghz single-core), 32GB RAM, 28 PCIe 3.0 lanes

Mid Haswell-E System:
Core i7 5930K $590
Asus X99 Deluxe $399
G.Skill RipJaws DDR4-2133 32GB (4x8GB) $480
Total: $1469 for 12 x 3.5Ghz threads (3.7Ghz single-core), 32GB RAM, 40 PCIe 3.0 lanes

High Haswell-E System:
Core i7 5960X $1050
Asus X99 Deluxe $399
G.Skill RipJaws DDR4-2133 32GB (4x8GB) $480
Total: $1929 for 16 x 3.0Ghz threads (3.5Ghz single-core), 32GB RAM, 40 PCIe 3.0 lanes

For most people reading this thread, particularly gamers, that first High Haswell system makes the most sense. It has high-clocked cores needed for good framerates without hitching, and 8 x 4Ghz threads is still a lot of multi-threaded CPU performance. If you really want more PCIe lanes, you can upgrade to a motherboard with a PLX switch chip, which multiplies the 16 lanes from the CPU into 32. They still share 16GB/sec of bandwidth, but each card is uncapped, and more than two-way SLI is allowed. You can also max the system out with 32GB of RAM if desired.

The Basic Haswell-E system only provides 24% more CPU performance best-case in heavily multi-threaded apps than the 4790K due to the lower clockspeed, so I don't see how it's worth the extra $570 over the High Haswell System, or even the $325 over the Max Haswell system with the same amount of RAM and more electrical PCIe lanes. This is the cheapest way to ridiculous CPU performance if you are overclocking, however, and that would benefit from the additional memory bandwidth.

The Mid Haswell-E system at least has a kind of value proposition, as it offers a full 40 PCIe 3.0 lanes and a 31% CPU performance boost over the 4790k, along with doubled memory bandwidth.

The Max Haswell-E system has a ridiculous price tag, but this is actually the system that makes the most sense in many ways. With a 50% CPU performance boost over the 4790K in heavily-threaded applications that scales to over 100% when overclocked, if you can use the cores nothing will come close.

Edit: Added a "Mini Haswell-E" system, which represents a lower-end build using the boards launching later this year. It makes a lot of sense compared to the Max Haswell system if you don't need more RAM, since DDR4 will be going down in price unlike DDR3, but you still give up a lot of per-thread performance and spend an extra $250 over the High Haswell system for 24% more multi-threaded CPU performance. Even overclocked it doesn't seem to make sense unless you really need the extra cores without needing RAM, which is odd.

Alereon fucked around with this message at 16:58 on Aug 31, 2014

Alereon
Feb 6, 2004

Dehumanize yourself and face to Trumpshed

College Slice

Combat Pretzel posted:

--edit: ^^^ If you can forego a bunch of the options on the deluxe board and go with the X99-A instead and go for 16GB of RAM, you can get way cheaper for the basic Haswell-E. I'm currently at 748 for a 5820K, 4x4GB of DDR4-2400 and that X99-A.
Yeah cheaper boards are launching later this year, though the X99-Deluxe is Asus's mainstream LGA2011-3 board. I added a system like that to the list called "Mini Haswell-E", though it seems to occupy an even narrower window. I can fill 16GB of RAM pretty easily when exercising 8 threads so I feel like you need to go 32GB if you're bumping up to Haswell-E, though of course that depends on your workload. At least DDR4 is getting cheaper while DDR3 gets more expensive.

Alereon
Feb 6, 2004

Dehumanize yourself and face to Trumpshed

College Slice

The Lord Bude posted:

You could also buy Asrock boards instead of paying more for Asus stuff with fewer features. Asrock X99 extreme 3 is $209; and there are a whole raft of asrock boards out at different price points above that.
Those cheap Asrock boards aren't any good, though. There are a few factors that go into whether a motherboard is good or not, not all of them are easy to measure or market, and features are the least important. Here's the basic order of importance of considerations for a motherboard:

1. Design, trace layout, and overall engineering
2. Component selection and quality
3. BIOS engineering and optimization
4. Features

Audio quality is the best, most easily measured example of #1. When you measure the audio output quality of a motherboard you're measuring how well the designer isolated the analog audio signals from the other traces and sources of interference on the board. Poor isolation causes audible hiss and noise in the background of audio that changes based on system activity, which can be very annoying. The audio section from the Asrock Z97 Extreme 6 upthread illustrates this well, all the boards with the same audio chip get similar dynamic range scores, but despite being a higher-end board the Asrock Z97 Extreme 6 has THD+N scores similar to mid-range mothebroards. Their less expensive boards would be further down that chart.

Even if you don't care about audio, those same factors apply to the signal traces you DO care about, particularly those between the RAM and the CPU. The quality of the signal between the RAM and CPU is the single greatest factor affecting system stability, and it's why RAM that works just fine may fail in a lower-quality board. It's hard to measure this directly, but memory overclocking results are a good indicator, and this is an area where Asus has excelled for a few generations.

Power delivery quality is a combination of the first and second factors. The number of VRM phases and the quality of the capacitors selected is a simple component choice issue, using thick enough traces to carry power from the top of the board to the PCI-Express slots is more layout and design. Cheap boards like that have extra power connectors near the PCI-Express slots so they can cheap out on those traces. Like all things this is a matter of degree, Gigabyte boards go so cheap you always have to have the extra power cable connected for stable operation, that may not be the case with only one or two cards on that Asrock board.

Some other component selection choices are things like the network adapter. Good boards use Intel, low-end boards use Realtek, garbage boards use KillerNIC chipsets. The audio chipset itself is pretty standard these days and doesn't make much difference.

BIOS quality comes down to three main factors, POST time, DPC latency, and power usage. POST time is how long it takes the system to start loading the operating system from the time you press the power button. Better motherboards initialize faster, though boards with more hardware take longer to initialize (though usually you can get this time back by disabling that hardware). DPC latency is how long (in microseconds) the system takes to respond to an interrupt, and depends on component selection, drivers, and BIOS optimizations. A board with poor DPC latency can feel noticeably laggy. Asrock has actually been pretty good compared to other brands on both POST time and DPC latency in recent generations, though these new boards remain to be tested. BIOS optimizations also play the biggest role in power consumption, though this isn't very relevant for desktop users.

Only now do we get to the features that manufacturers use to segment their products. There's some overlap here with component selection for things like VRM phases, as the number of phases is usually a selling feature for the board. Similarly, the presence of ports or controller chips is both a component and feature selection. I personally think the fan control that Asus offers on its boards is pretty drat compelling, and I like Asus's UEFI firmware, but these aren't really the actual reason to buy the boards. Then again Asrock does have a dehumidifier

https://www.youtube.com/watch?v=-mQfFVT4tT8
(The theory is that the hot air in your case can hold more moisture than the cooler air outside, so if you shut your system down moisture could condense when the hot air cools. This runs the fans to blow the hot air out of the system before it cools. It can also run daily if your computer was turned off, to ventilate the system as the day cools down for the same reason. You decide whether this is crazy-talk!)

Alereon fucked around with this message at 18:40 on Aug 31, 2014

Alereon
Feb 6, 2004

Dehumanize yourself and face to Trumpshed

College Slice

go3 posted:

Just remember to buy for what you need now, not what you want later.
The key is to look at total cost of ownership, and to be realistic about what you're getting when you spend extra money, as well as what you're losing when you don't. It seems pretty obvious that getting a Core i5 over a Core i3 will extend the useful life of the system by years since you're doubling the CPU performance, so spending an extra $50-$100 on that is a no-brainer. It might even be the case that jumping up to a Core i7 4790K for another $100 could make sense from a value perspective if it allows you to push off an upgrade for another cycle, though this is much more speculative. On the other hand, I don't see a way that going up to Haswell-E could make sense unless you really can use the cores for some application where your time really is money. In that kind of situation a dual-socket box with Haswell-EP might even start to make financial sense, when available.

Alereon
Feb 6, 2004

Dehumanize yourself and face to Trumpshed

College Slice

Combat Pretzel posted:

What does the BIOS have to do with anything beyond initializing device parameters during POST? Once the NT kernel's booting, it is entirely out of the picture.
Have you seen the wealth of power-related options on Haswell systems with UEFI? You can override many of them with the Intel Extreme Tuning Utility, but unless you take the time the defaults set by the motherboard manufacturer will apply. Microsoft sure as hell isn't tweaking the VRM frequency.

Note that in some cases default settings may not even match Intel's specs, for example the Multi-Core Enhancement on most overclocking boards where the CPU defaults to its highest single-core turbo bin at all load levels.

Alereon fucked around with this message at 19:30 on Aug 31, 2014

Alereon
Feb 6, 2004

Dehumanize yourself and face to Trumpshed

College Slice

Combat Pretzel posted:

OK, but these things happen and are set up during POST. The kernel, once running, is not going to call into the UEFI BIOS though. So how do DPC latencies factor into this? If the Xonar driver drama taught me anything, it's seems mostly a driver issue.
There are a number of settings that affect DPC, some of which may or may not be exposed to the user. There are definitely event timer settings, and there's likely settings similar to pollrates for integrated peripherals that let the motherboard manufacturer balance performance and latency. You can see measurable and meaningful differences in DPC latency between boards with similar integrated peripherals. This did get a lot better with Z97 boards, and it will be interesting to see how X99 boards stack up with eachother.

Adbot
ADBOT LOVES YOU

Alereon
Feb 6, 2004

Dehumanize yourself and face to Trumpshed

College Slice

The Lord Bude posted:

This is excellent information. Have there been any reviews of asrock boards compared to Asus boards that analyse these issues at that level? I ask more for 1150 boards than anything else, because we tend to steer people towards asrock in the parts picking thread because you tend to save money and get higher grade audio chips in asrock boards vs similar Asus boards and I'm wondering whether we should rethink that. I've never seen an asrock board requiring extra PCIe power connectors like gigabyte do though.
Anandtech's reviews are pretty good, as are TechPowerUp's. They do DPC latency, audio quality, and POST time benchmarks that can be hard to find elsewhere, and Anandtech in particular has done reasonably good vDroop tests, including discovering Gigabyte's fake voltage readings. I think choosing a board comes down to picking a range that suits your intended purpose, then picking a sweet spot within that range that best matches your desired features and price. For example, most Asus channel boards have pretty similar audio section layout with good isolation, but you can clearly see a decline in the number and size of capacitors as you move down in price.

BurritoJustice posted:

I don't think we should stop recommending ASRock based entirely off speculation and anecdotes, ignoring the multitudes of objective reviews showing them to be good, excellent even. The only spec in that review that the ASRock is not fantastic is a single audio benchmark where it is middle off the pack instead of the absolute best (like it is for DPC latency.
Like I said above, the point is that the quality of isolation of the audio traces tells you a lot about the quality of the board in general. That's a board loaded with typically higher-end features, but with a mid-range price corners have been cut to get there. Some of those, like the limited number of VRM phases, are pretty reasonable. Poor quality trace layout and isolation, which you see as that poor THD+N result, is not. The difference between "good" and "crap" is making the right tradeoffs, and lower quality for more features is almost never the right trade.

quote:

The added connector for PCI-E power is not an indicator of PCI-E trace quality, it's a feature. It's useful for stability when you have a whole pile of add in cards each requesting 75w from the slot. It's entirely optional. You know what else has one? The ~Asus~ Rampage V Extreme super ultra gamer edition.
Extreme high-end boards typically do have at least one additional 8-pin power connector to handle a large number of cards. Very low-end boards on which a lot of corners have been cut have a SATA power connector or a molex connector. The $400 Asus X99 Deluxe does not have any additional power connectors and it is meant to take four PCI-E 3.0 x16 cards, plus an x4. It has the build quality to power its slots without additional power connectors, the Asrock board does not. Consider that Gigabyte boards with a SATA power connector require it to be connected for stable operation with even a single videocard, despite the manual claiming it is for multi-GPU.

  • 1
  • 2
  • 3
  • 4
  • 5
  • Post
  • Reply