Slashdot Mirror
Retail Radeon R9 290X Graphics Cards Slower Than AMD's Press Samples
crookedvulture writes "AMD's recently introduced Radeon R9 290X is one of the fastest graphics cards around. However, the cards sent to reviewers differ somewhat from the retail units available for purchase. The press samples run at higher clock speeds and deliver better performance as a result. There's some variance in clock speeds between different press and retail cards, too. Part of the problem appears to be AMD's PowerTune mechanism, which dynamically adjusts GPU frequencies in response to temperature and power limits. AMD doesn't guarantee a base clock speed, saying only that the 290X runs at 'up to 1GHz.' Real-world clock speeds are a fair bit lower than that, and the retail cards suffer more than the press samples. Cooling seems to be a contributing factor. AMD issued a driver update that raises fan speeds, and that helps the performance of some retail cards. Retail units remain slower than the cards seeded to the press, though. Flashing retail cards with the press firmware raises clock speeds slightly, but it doesn't entirely close the gap, either. AMD hasn't explained why the retail cards are slower than expected, and it's possible the company cherry-picked the samples sent to the press. At the very least, it's clear that the 290X exhibits more card-to-card variance than we're used to seeing in a PC graphics product."
This has been discussed in many places like toms hardware. Essentially they found the cards volting is determined in the bios and the fan speeds can be altered. change the bios which many have released and the undervolting which occurs at lower temps is solved. Sapphire already released a new bios for the card to make these changes to keep them consistent yet keeping them from going above 95 degrees.
AMD cards are faster for GPGPU. They crunch numbers faster than nVIDIA cards, at the cost of more power and flakier drivers. Also AMD has better free software drivers.
Because they aren't the same thing when dynamic clocking is enabled, and if you're serious enough to care you disable it. Duh.
If you glance at the Newegg reviews for the R9 280X you'll note someone found that the voltage was locked.
And you are like the people that always eat the same pizza, while complaining that there's no variation.
Cherry-picking would be a bad thing; but if it turns out that the junior thermal past application technicians get less attentive once the first production batch is finished and the people who've been babying the project leave, that wouldn't be a total surprise.
This is becoming a habit for AMD. You can't even trust their FRAPS measured frame-rates. Seriously. I will never ever own another AMD card. Their graphics cards are nothing but empty promises and bad drivers.
Oh gosh I hope this doesn't result in some poor sap attempting to play Battlefield 4.7 and while thinking they should achieve a pure 115fps they only hit a measly 92fps and their lives are ruined forever. The consequences will never be the same.
This probably happens more often than we think. Frankly, it makes sense to validate that a card is going to run solid for someone before you send it to them if they're going to be blabbering all over the interwebs about it. It's just in this case they got burned (and justifiably) due to the fact that the software/driver adjusts the frequency independently instead of being a static clock speed (something they should have disclaimed to the reviewers).
I'm a huge fan of market competition and AMD. If this is not not a slashvertismemnt then I'm a noob.
Really, I know what I'm doing...Ohhhh, look at the shiny buttons!
It seems like this is a recurring problem for AMD. I recall a similar issue with the bulldozer cpu's
Somewhere, something incredible is waiting to be known. -Carl Sagan
Toms Hardware covered this pretty extensively a month ago.
The short story is that AMD is throttling clock speeds to hold within a temperature limit. They learned the hard way that 40% PWM does not equal 40% fan speed, especially across all fans the OEMS used. There's a driver fix for that now measures fan speed and adjusts accordingly when in quiet mode that eliminates most of this performance discrepancy (retail cards can now see higher performance in line with review samples).
Remaining differences between cards may be due to different heatsink grease, also already examined by replacing the grease on a retail card for a significant performance gain.
Someone obviously didn't buy the turbo liquid nitrogen supply option.
I believe the entire R9 series of cards are little more than rebranded versions of older cards AMD just discontinued. The R9 280X for example? It's just a 7970 card. Sure, it may have a few BIOS tweaks and such, but you can even peel the label off the circuit board of many of them and find 7970 etched on the board underneath the R9 280X label.
Personally, I think AMD should have waited until it had legitimate improvements to offer before releasing anything, rather than trying to fool people with the R9 series.
I use Intel/Nvidia with FreeBSD because they have better driver support than AMD/ATI.
I would expect the retail NVidia cards vary considerably from the press review cards. Each vendor has their own tweaks and adjustments.
Nope. The 290/290X is a much larger chip - similar architecture, but bigger (and mildly improved).
And cherry picks the best units to send out to reviewers.
Can't seem to care too hard. Unless this wasn't an industry standard way of doing things... But it is.
Not just in computer related things either. The stuff sent to be reviewed is always the absolute best and pre-tested before the reviewers get it.
Video card manufacturers have been cheating for more than 20 years. Sending hardware and drivers to reviewers that has been tweaked and optimized far beyond what is available at retail.
I used to be an Nvidia guy until last week. Just hopped onto the crypto coin mining bandwagon and my new 7950 card should pay for itself in ~2-3 weeks.
Nvidia are junk for this, unfortunately.
And it shows, you try to find a new or 2nd hand 79xx card anywhere for a reasonable price. They have all been snapped up by the miners :(
Benchmarking: If you want it done right, you have to do it yourself.
What a surprise that Wasson couldn't be bothered to figure out what's actually going on this time, either.
Silicon varies. A lot, really. Some chips run faster and cooler than others, simply by chance. This is why they're sold at different speeds, core counts, etc. The process of separating the chips into these categories is called 'binning', but it's not perfect in that there will still be some variation within a bin. That variation won't go lower than the bin's minimum standards, but it will go higher. With the number of review cards AMD - or any other chipmaker - ships out, it would follow that several of them would be of this better quality. The only way to avoid that, in fact, would be to specifically select the minimally-performing cards, which would be silly.
The reason that we haven't seen these chips in retail boards - and, to be fair, you wouldn't expect to find one in a sample of two, more like one in ten or twenty or even fifty, and Nvidia would surely know this - is that the board manufacturers that buy the chips from AMD hold those best chips back for their special, overclocked cards that have fancy cooling solutions. Rumors point to those coming out sometime in the range from Christmas to January for the R9 290/290X.
More specifically to the 290-series, this chip variance will show up more readily in that better chips will throttle a little less - but it's usually not going to be as dramatic a difference as, say, the variation in fan speed that they had to solve (whether this was AMD's mistake in not considering variation in fan performance or the manufacturers using cheap, crappy fans is uncertain). Or using crappy thermal paste. The neat thing about this method of dynamic frequency is that, aside from the limitation that a chip can never draw more power than its maximum safe limit, performance is entirely controlled by cooling power. Want to never throttle? Well, nothing can be done about Furmark - it draws too much power and you'd burn your chip out. But for every game out there, just turning up your fan speed will suffice, and the cards even have a built-in switch to do that if you don't want to simply adjust a slider in the control software. And when the custom cards with those nice, top-grade chips turn up, they'll run at full speed and be nearly whisper-quiet, too. Except full speed could be between 1050 and 1100 megahertz. Or you could use water cooling - running at 1300 megahertz on water isn't unusual from what I've heard, though you would also need to overclock to do that.
In the end, this is just Nvidia using a shoddy journalist to do a little bit of mud-slinging.
The 280X is just the 7970 and it's not really a secret. The reviews I read pointed this out (I also just bought a 280X two weeks ago). The 280X even shows up in your device manager as 7970.
um yea coin mining now days, you will barley break even vs cost of card and electric needed to mine them. you missed the bandwagon by about 1-2 years.
There are lots of alt-coins in addition of bitcoin.
Some are easier to mine on GPU than other.
You need to pick up a coin that plays well with it.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
You're joking but that's more or less what's happening.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
If you bothered to RTFA (I know!), you'd see that they indeed checked this out. They flashed the BIOS of their sample card onto their worst performing retail card. There was a small difference, but far from enough to make up for the gap between that card and the sample unit they received from AMD
Not that BIOS. As other have pointed in the thread, the variation in performance is more or less linked to the variance of thermal management.
Not all PWM Fan behave the same. There's a *newer* BIOS version (not as in "use the one that came with the sample" but as in "download the latest version that was made available on the manufacturer website, betwen when you bought it and now").
This version of BIOS is better at computing what signal it should send to the fan to have better cooling.
And once the cooling is improvent, the card will automatically scale up its speed.
Also, there can be difference in thermal grease, etc.
At this point it's reasonable to assume that AMD cherry-picked the cards they sent reviewers to make sure they were as good as they could be.
Or, instead of cherry-picking, maybe there's some build quality between the first engineering sample sent by AMD, and the mass-produced card by NONAME asian manufacturer ? (Or even mass-produced cards by very popular brands that have to fulfill lots of orders ?)
Difference in quality of the fans (NONAME will pick whatever is currently the cheapest, and even with popular big-names, there's going to be some variance, depending on where the current batch was sourced).
Difference in quality of thermal conduction of the interface. Difference of quality of thermal grease (NONAME will pick the cheapest, bigname might have variation in batches, specially if they source batch from several manufacturer to keep up with the pace). Difference in quality of work (NONAME might even do a sloppy job in applying the thermal medium to the radiator).
You end up with exactly the same chip produced by AMD, but vastly different thermal condition, all this with a firmware and a driver which isn't yet best at fan throttling, and you end-up with some measurable difference in output. ...BUT...
Pick up Nvidia cards, and you're going to see exactly the same effect.
Either card that vary in their performance (or that have big variation in temperature, depending on how the current firmware throttles the card)
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
but if it turns out that the junior thermal past application technicians get less attentive once the first production batch is finished and the people who've been babying the project leave, that wouldn't be a total surprise.
Now add in the mix that some parts, like fans, might be sourced from several different manufacturer (and according to source, BIOS wasn't until latest update so good at operating them), add also that there might be variation in quality between the different batches of thermal paste (which got very probably sourced from several productors) and the output variation is clearly expected.
But also fixable (newest BIOS to compensate fans with better throttling, manually replace thermal paste. Now cards works as good as benchmarking sample or even better)
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
due to the fact that the software/driver adjusts the frequency independently instead of being a static clock speed (something they should have disclaimed to the reviewers).
It's well known that these cards operate at a fixed temperature and push the clock and voltage as high as they can within these thermal limits.
It's so well known among professionnal, that some like Tom's are giving advice about BIOS replacement (newer have better and more consistent throttling or fan accross all the varied parts) or thermal paste replacement (to improve cooling and thus performance).
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
Intel/Nvidia are more expensive; AMD is better price/performance ratio. Budget-to-midrange is cheaper on AMD parts.
Personally, I think AMD should have waited until it had legitimate improvements to offer before releasing anything, rather than trying to fool people with the R9 series.
The problem, is that AMD got too busy doing legitimate improvement under contract for the coming generation of consoles (Xbox One, Play Station 4, and some of the upcoming Steam Box, all run the same combo of AMD CPUs and GPUs).
With that work, there was going to be some delay for their PC's Radeon HD 8xxx serie.
So it was either:
- have absolutely nothing to sell.
- do some small upgrade on the older board (R2 270/280 are simply older board slightly upgraded) and older chips (R2 290(X) are GCN1.1 chips, slightly newer than the GCN1.0) and have something to sell until the real new gen comes, while still taking advantage of the time to add some improvements.
Note that this was pretty much announced that way, and well known for people in the field.
now look at the bright side of the things:
even if they are coming with some delay, that means that next year, you're going to finally see the newer GCN2.0 chip based card, that have taken advangtage of all the R&D done for consoles to improve the performance and quality for radeons.
A little bit later, but a little bit more R&D money poured on the steps leading to them (specially the latest step)
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
ok, so I checked the price of the Radeon R9 290X. It is about $550. wow. Not trying to start an argument but do people really need a $500 video card? Just asking. I thought a $200 would be enough for most users including heavy gamers. I remember when gamers bought $200 AGP cards and were satisfied.
The latest ones crash ALOT if you read maximumpc.com or tomshardware.com and my Nvidia fanboys I raid with.
AMD has better quality hardware with less flaky voltage regulators. I went through 2 nvidia cards over 8 years that failed and switched all AMD/ATI in 2010 with the phenom II (better that steamroller per clock tick sadly ) and an ATI 5750.
Had one bizaare issue with black edges on the screen when switching to HDMI. That problem went away after I went into the drivers and configured my screen to not do underscanning.
NO issues for 2 years! Just upgraded to an ATI 7850 and it runs GREAT. No black screens, no overheating, no wierd driver issues at all. This was true in 2002 with the ill fated ATI Rage Pro 128 TV adapter. The Mac version was the only one that worked ok.
But damn that was 11 years ago!
I admit I do not like steamroller, but over all I got a nice $600 system that is 6core, runs VMWare accelerated, back in 2010 back when I would need an Intel extreme edition for $700 without anything else but the damn chip!
What they do not tell you is Intel cripples the bios and turns off hyperthreading and Virtualization support to force you into the more expensive units. AMD chipsets never do this.
IF AMD gets their act together and not create underperforming space heaters I will consider AMD again for a CPU purchase. But what I have works great and the phenom II was a pretty competitive chip for its time.
Raedons are great GPUs and are very competitive with Nvidia. Drivers are worse today with Nvidia unless they clean things up.
http://saveie6.com/
My cheap AMD/ATI system is the most stable computer I ever owned.
Only BSOD was from an unproper shutdown and file corruption that I did.
http://saveie6.com/
or whether review cards come from the initial run, probably the one where AMD's people are mostly tightly and nervously observing the process, rather than the potentially more variable just-another-day-slapping-chips-on-cards production that follows.
I would indeed agree with your first post and this part. To me, a big conspiracy to manipulate results is far less likely than simply slopiness of a mass-produced good, where speed of production counts, in order to quickly meet the demand.
To quote a variant of Hanlon's Razor (often attributed without sources to Napoleon Bonaparte):
"Never ascribe to malice what can adequately be explained by incompetence."
Merely variation is only inconvenient, and may well mean that the usual 3rd-party overkill heatsinks actually help significantly.
Yup very probably. Specially with modern card that try to go as fast as they can, while still within target thermal envelope.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
For some values of GPGPU. They are great for MHash(Bit coin last year) but poor for some Boinc projects. Cuda (prior to some specific changes made for AMD to make it par) crunched 5-10x faster than its AMD equivalent.
while(1) attack(People.Sandy);
perhaps, but this is typical of AMD and ATI products ranging all the way back to my experiences with Athlons and the ATI Rage 128 graphics card.
Their hardware just isn't as stable as Intel or Nvidia. And that's my experience with 4 different AMD systems over the years. When you put an intel computer on the same power supply, stability just happens.
They're using their grammar skills there.
bwahahaha. If you were a serious miner, you'd have an ASIC rig.
GPU miners don't cover the cost of their electrical bills.
They're using their grammar skills there.
Maybe AMD's open source drivers are better but Nvidia's drivers are better suited for use to do renders in Blender because of their support of CUDA.
While few will need to buy a graphics card for Blender 3D in particular it does show that you have to consider what you wish to use the card for when you want to make a purchase to get the performance you need out of it.
Ignorance means more money for me =)
I listened to a Maximum PC podcast about this a couple of weeks ago. Is this a slow news day?
Also AMD has better free software drivers.
Yeah, but their proprietary Linux drivers are simply atrocious. Performance isn't so bad but stability is awful.
The Gospel according to lolcat
Hardware wise, I've had way more issues with NVIDIA. Software/driver wise, they've both had their fair share of issues. Don't go near AMD Enduro systems...
He didn't say he was mining Bitcoin.
He could be mining Litecoin or other scrypt-based coins.
I'm talking about all the above being GCN 1.x chips.
The GCN 2.x chips (which was initially what was going to be inside HD 8000, before they delayed everything) will be here early next year.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
Will gamers see that much of an improvement? The PS4 and Xbone being x86 hardware is nice as the excuses on why a port to PCs can't happen, but both consoles are pretty sad when compared to current mid-range PCs let alone a high-end rig. The Xbone one is struggling to hit 1080p while the PS4 is hitting it, but at 30fps. This is matching or lower than the performance of a current mid-range PC and the performance gap will only widen.
...all this done with intergrated GPU. That's the key point. These performance are pulled using just an embed chip, that pulls minimal amount of power.
Now scale this thing up, move one generation next ( to GCN 2.x ) and the discrete card that will be getting next year from AMD are going to be quite interesting.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
Yeah, and that's the main exchange where I trade mine.
A few other thing:
- much smaller scale, but cryptsy has almost the whole zoo.
- lots of trader are playing around and speculating with minor coin.
- That means there's a lot of exchange traffic between major and minor alt coins.
- That means it's easier to exchange whatever you mined with whatever is more usefull to do transaction for you.
- LTC is starting to grow big enough to get some indenpendance from BTC. ...thus it's starting to get more acceptance (more payment processor are accepting it too) ...its fluctuation vs. fiat isn't so thighly couple to BTC as before.
-
-
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
And according to Tom's, that exactly what the last BIOS update was all about. Taking better into account the tacho feedback.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
I've just ordered a R9 290X, before I saw this (oh no!). If this effect is reducing the 290X performance, and 290X is just a higher clocked 290, could I expect similar performance on the 290 as the retail 290Xs? If so I just want to return the 290X and recoup 25 % of the cost... Advice greatly appreciated
And me :)
Yeah I thought I missed the bandwagon, but then I discovered scrypt based coins and it turns out I might just be able to catch the tail end of this ride. My first card I just bought should pay for itself in ~16 days. *IF* that happens, I will be upgrading to a bigger rig.
It's not too late.
Mine for WorldCoin by the way. It's still one of the best longterm options for mining.
I will either be wrong, in which case I have a perfectly sellable graphics card I can move on (Or game with) or I am right and BWUAAHAHAHAAAA
I think that's called win:win :)
Just for your info, I covered the extra electricity charges for the coming month in under 24 hours.
And there are no ASIC chips for scrypt based mining yet so it's currently GPU mining all the way (to tha moon, lol).
I'm not 100% sure yet, but I also think that due to the very nature of the scrypt algorithm, ASIC mining is unlikely to come soon as it;s really reliant on having a lot of fast memory (Basically, high end video cards). There is one company that claims it's close to making one, but they have not actually shown anyone a working product yet so we'll see.
A GPU is a type of ASIC connected to a bunch of fast memory. There is absolutely no reason why you can't design a dedicated mining ASIC for any coin crypto algorithm even if it needs lots of memory. It costs extra to to license a DDR memory controller core, and get it working, so the competence of the ASIC design team is an issue (this is a big deal given the low quality of coin mining ASICs to date), but in the big picture that doesn't matter -- if any of these coins had a real future someone would design an ASIC miner to mine them.
They don't have a future, of course, except to be mini speculative bubbles popping alongside the big bitcoin speculative bubble.
What's sad is the amount of irreplacable fossil fuel being burned to power mining all these intrinsically valueless "coins". At least when it was tulips, creating more tulips involved a renewable-energy process (photosynthesis) and sequestered atmospheric carbon. And they had at least some value, in that they were pleasant to look at once they bloomed. The cryptocurrency bubble, it's got nothin'.