AMD 2xx series how did they get it so wrong?

[pgi947]

The only thing I would say AMD got 'wrong' so to speak was the length of time it took to release. iirc AMD supposedly had the chip design ready for months before it was put into production. I suppose BF4 forced their hand a little. Better late then never I suppose.

 

[EdgeDs]

The hot it should wrong the fx rred miley tom seasonic the it peoject and release a New card

 

[Goose]

The hot it should wrong the fx rred miley tom seasonic the it peoject and release a New card

Quoted for truth.

 

[Freddie1980]

^ My heads in a deep spin after trying to make sense of that.

 

[IT Troll]

Aside from the usual naff AMD release cooler, I don't see how they did get anything else wrong. The card is designed to handle those temps so no worries there either.

The trouble is it can't handle those high temps and so throttles badly, which is why AMD have released two driver updates this week modifying the fan profile in an attempt to stabilize performance.

If AMD had released the 290 at launch with a "Titan cooler" for £350, these discussions would look very different and AMD wouldn't be able to churn them out fast enough. That is the missed opportunity.

 

[TheRealDeal]

^ My heads in a deep spin after trying to make sense of that.

I thought it was some kind of joke so kept on trying to get the meaning.

 

[arc@css]

^ My heads in a deep spin after trying to make sense of that.

Yep.

 

[GoogalyMoogaly]

The trouble is it can't handle those high temps and so throttles badly, which is why AMD have released two driver updates this week modifying the fan profile in an attempt to stabilize performance.

It is disappointing that they've had to sort this out with a driver update (which I believe isn't automatic either and requires you to reset to defaults in CCC?) and couldn't get it right at launch. How hard would it have been to have tested a few cards?
On the plus side though, they did fix the problem, it only took a driver update (no BIOS flash that could go wrong or asking people to return the cards for a replacement) and it didn't take that long to get it sorted.
Although it doesn't seem there are any official drivers available for download for the R9 cards, just beta ones. Are there really no official drivers out yet for these cards? Cuz that's a bit poor, not everyone wants to run BETA drivers.

 

[EdgeDs]

So sorry am nearly blind irl and did not release that the program i got too help me was changing everything i typed

 

[Steampunk]

Although it doesn't seem there are any official drivers available for download for the R9 cards, just beta ones. Are there really no official drivers out yet for these cards? Cuz that's a bit poor, not everyone wants to run BETA drivers.

This all was talked about when AMD stopped doing monthly WHQL drivers. Long story short, WHQL doesn't mean a lot, and it allows AMD to spend their resources doing quick/useful drivers like the several updates we've seen popping up in the last couple of weeks.

 

[TheRealDeal]

So sorry am nearly blind irl and did not release that the program i got too help me was changing everything i typed

No worries.

 

[drunkenmaster]

The only thing I would say AMD got 'wrong' so to speak was the length of time it took to release. iirc AMD supposedly had the chip design ready for months before it was put into production. I suppose BF4 forced their hand a little. Better late then never I suppose.

This isn't how chip production makes, first of all it's rare to impossible to go from a tape out(chip is essentially ready for production) to release within 6 months, NO gpu's do this, 7970, Titan, nothing, that is how long it takes to make several batches, validate, make pcb's, test, assemble and ship.

It takes, or I should say a few years ago it would take 6 weeks from the time a silicon wafer into production to come out a finished wafer of chips. With each shrink this time is increasing as things like double patterning significantly increase the time in production and is one of the single biggest reasons for drastically increasing cost from one node to the next, with 10nm projected to cost 4x per wafer than 28nm.

The move to 450mm wafers(over current 300mm) is one way to combat that, time is still longer but basically bigger machines and bigger wafers doing increasing chips done in that longer period.

AMD released this chip when ready, nothing more, nothing less, they didn't have it ready and waited, there is no reason to do that at all.

The trouble is it can't handle those high temps and so throttles badly, which is why AMD have released two driver updates this week modifying the fan profile in an attempt to stabilize performance.

If AMD had released the 290 at launch with a "Titan cooler" for £350, these discussions would look very different and AMD wouldn't be able to churn them out fast enough. That is the missed opportunity.

No they wouldn't, because firstly it DOES NOT throttle badly, Tom's the only person to mention a problem had 10% less performance.. shocking, secondly it's a teething issue, they didn't increase the fan speed from press samples, the stabilised the production models to all have the same speed as PWM was giving less stable fan speeds, nothing more, nothing less.

The people who are making up this crap about it being much worse than Titan/780gtx, are trolls who are, I'll say it again, making it up. Titan throttles when it costs £800, fine, AMD's card throttles, not fine? Titan runs cooler, doesn't throttle and overclocks further with third party coolers, fine, AMD card does the same... terrible.

Even ignoring that it is literally harder to cool something with the same heat output(let alone more) when the source is smaller, it's doing no worse a job than Titan. The fan boys would post troll threads even if the card never throttled, cost £200, and was 50% faster, trolls be trolls, they'd just pick on something else, one person in a 1000 has their card turn up DOA... we'd have threads about how they all fail. We already have the "ridiculously noisy" posts based on a fan speed of 100% which NEVER happens at stock under any circumstances in gaming, doesn't stop fan boys bleating on about the 100% fan noise and posting videos comparing a THROTTLING Titan at stock to a 100% fan noise 290x and saying noise is insane.

Fan boys be fan boys, the quality of the product doesn't make any difference.

 

[IT Troll]

No they wouldn't, because firstly it DOES NOT throttle badly, Tom's the only person to mention a problem had 10% less performance.. shocking, secondly it's a teething issue, they didn't increase the fan speed from press samples, the stabilised the production models to all have the same speed as PWM was giving less stable fan speeds, nothing more, nothing less.

Haven't there been two changes? The first was a global increase to the fan speed to reduce throttling / increase performance. Which led to launch delay and press review updates. Then the second more recent one to address the differences in PWM which was causing low fan speeds and throttling in certain cards.

It is good that AMD are on the ball and actively addressing the issue but fundamentally the cooler is not fit for purpose and is running at the limit of its capability. I really hope we see a better cooler design next time and/or a cooler running chip.

 

[ubersonic]

No they wouldn't, because firstly it DOES NOT throttle badly, Tom's the only person to mention a problem had 10% less performance.. shocking, secondly it's a teething issue, they didn't increase the fan speed from press samples, the stabilised the production models to all have the same speed as PWM was giving less stable fan speeds, nothing more, nothing less.

Before I say anything I would like to say the 290's are fantastic cards and I am in no way hating, just adding a couple of corrections. Firstly Tom's actually got a 18% performance variation between a bunch of R290X cards, the non X retail card they tested was 23% slower than the press card, and they weren't the only ones to mention a problem they were the first, it was replicated by others including AMD themselves, however in both cases the issue has been fixed by the new driver AMD released in response to the problem, so all is now well.

Secondly, they have actually increased the fan speed from what the press samples had, it's now standardised ~150RPM above press card speed, as a consequence performance on R290/290X is now up slightly over the original reviews so it's a win all round, Toms got a great scoop and kudos for pinpointing the issue, AMD gets kudos for fixing it super fast and the consumer gets more performance

 

[EdGey]

Performance is great... it's the cooler which lets the card down in my eyes.

 

[Goose]

So sorry am nearly blind irl and did not release that the program i got too help me was changing everything i typed

No need to apologise dude, I shouldn't have made fun. My apologies.

 

[mmj_uk]

Before I say anything I would like to say the 290's are fantastic cards and I am in no way hating, just adding a couple of corrections. Firstly Tom's actually got a 18% performance variation between a bunch of R290X cards, the non X retail card they tested was 23% slower than the press card, and they weren't the only ones to mention a problem they were the first, it was replicated by others including AMD themselves, however in both cases the issue has been fixed by the new driver AMD released in response to the problem, so all is now well.

Secondly, they have actually increased the fan speed from what the press samples had, it's now standardised ~150RPM above press card speed, as a consequence performance on R290/290X is now up slightly over the original reviews so it's a win all round, Toms got a great scoop and kudos for pinpointing the issue, AMD gets kudos for fixing it super fast and the consumer gets more performance

It wasn't fixed because retail samples still ran 40-80mhz slower than the press sample with the 'fixed' drivers, all AMD achieved was to reduce the deficit. I don't know where the "it's all fixed now" thing is coming from because if you actually read the Toms article the retail samples still underperform compared to their press sample. Toms even spells it out:

By default, 190 MHz separated the average core frequencies of our fastest and slowest samples. After AMD's newest driver build, that number shrinks to 81 MHz.

Their press sample is still BETTER than their retail samples, just less so.

 

[Suarez7]

You got the mod who sent Suarez7 on holiday to thank for that

He's very good at baiting every topic into a warzone

lol

 

[CAT-THE-FIFTH]

It wasn't fixed because retail samples still ran 40-80mhz slower than the press sample with the 'fixed' drivers, all AMD achieved was to reduce the deficit. I don't know where the "it's all fixed now" thing is coming from because if you actually read the Toms article the retail samples still underperform compared to their press sample. Toms even spells it out:



Their press sample is still BETTER than their retail samples, just less so.

You mean like when Hardware.fr found that press GTX680 samples boosted 30MHZ to 60MHZ higher than production models??

Nvidia used non-deterministic boosting which had no upper limit. Dependent on the quality of the chip you got and the conditions,there was big variance in performance,as it was a lottery on what clockspeeds you got. I remember talking to some of the review and mod crew on Hexus about it.

It also meant Nvidia press edition cards tested in open air test benches had artificially inflated scores in reviews.

Nvidia cards use non-deterministic boots,which had a minimum limit of boost frequency and no upper limit in practice. Looking at the last three generations of AMD GPU boost(it was first done on their IGPs),it has a defined upper limit.

People really need to look at the articles from Hardware Canucks,hardware.fr,pcgameshardware and other websites,where they showed the GTX660TI,GTX760 and Geforce Titan having throttling issues with the stock cooling,especially after longer runs,where there were clockspeed drops over time.

This lead to many websites,making pre-warming runs before testing current cards,so they would heat up and is also why some websites attempted to lock the clockspeed at fixed values.

Here are some of the articles:
http://www.hardwarecanucks.com/foru...roundup-asus-evga-gigabyte-galaxy-msi-21.html
http://translate.googleusercontent....s.html&usg=ALkJrhj8G8JD8-Rfoq7gPcX-0XdORhBOyQ
http://translate.google.com/transla...97980/Tests/Test-Geforce-GTX-Titan-1056659/4/
http://translate.google.es/translat...013/nvidia_geforce_gtx_760_im_test/index8.php

Some of the drops over time are quite large.

This brings whole exercise could bring up some worrying points about benchmarking NVIDIA’s Kepler-based cards in reviews (and charts) where every single FPS counts. Sites benchmarking with a single run or shorter sequences will likely achieve the “best” results rather than realistic performance. Luckily, we have been able to avoid this issue by using four run-throughs of every benchmark, each with somewhat long testing times. We’ll have a full article looking at GeForce Boost and AMD’s equivalent in the coming weeks but for the time being, this is certainly food for thought.

As mentioned in the introduction, with Anno 2070 as extreme an example, the GeForce GTX Titan is able to reach its maximum turbo frequency as the GPU does not reach 80 ° C. And it reaches this temperature in all games we tested on a bench bench, with and without additional cooling around the map. If we measure the performance of a traditional way, we would get results more pupils and non-representative of the playing conditions in all our tests.

We had to take the time to observe in detail the behavior of the GTX Titan in each game and on each resolution to make sure we do performance measures in representative conditions.

Here are two examples with Anno 2070 and Battlefield 3 with a rapid test, a test temperature stabilized after 5 minutes and the same test as the latter but with two 120mm fans positioned around the map:

Anno 2070: 75 fps -> 63 fps -> 68 fps
Battlefield 3: 115 fps -> 107 fps -> 114 fps

The drop in performance once the temperature reaches cruising can be considerable. An efficient cooling may partially offset this decline, but are there not a contradiction in having to add noise to compensate for a graphics card is trying to remain discreet at all costs?

It also raises the issue of reliability of performance comparisons that you can read here and there, since this may be the big difference based on test conditions (ventilation card but wait or not temperature rise of the GPU for measurements) between extreme cases the gain is 19% in Anno 2070 and 7.5% in Battlefield 3!

Here are the frequencies that we obtained in practice for two selected scenarios: limited to 889 MHz without additional / card capable of up to 1006 MHz with additional cooling cooling card. 3DMark is only able to maintain the maximum frequency due to loading time between scenes that allow the GPU never have time to reach 80 ° C.

It is worth emphasizing that a typical benchmark behavior, immediately after starting the game in all situations almost always has a GPU clock speed of 1137 MHz result. After 15 minutes in a static scene is an entirely different value is emerging, which is on average only in the range of about 1000 MHz and this at 21 ° C room temperature. Thus we see an agent that acts still below the NVIDIA naming of 1032 MHz.

The speed of the fan behavior is interpreted here in a maximum noise level of about 2190 revolutions per minute. After that, then usually provide a further clock reductions, which are also observed in the course of the game that way. The regular course can cause a very different behavior than a static benchmark scene, as used in the present case. The example we have here demonstrated.

Geforce GTX Titan review: Benchmark Boost

The following is an excerpt from our book-article GTX titanium in PCGH 04/2013 (6 March in the trade):

For our benchmarks, however, GPU Boost 2.0 is a major limitation. We usually test on an open test bench, in which the graphics card always has enough cooling air to room temperature available - Optimal conditions if you will. But inside of housings often prevail considerably different conditions, particularly in the summer, even in well-ventilated enclosures significantly higher values ​​than our reach around 22 degrees Celsius.

Since the GTX titanium the temperature falls to a central role, we have taken the test a lot of effort and recorded the achieved clock rates at 28 ° C inlet air is warm for each benchmark game in each resolution separately and enforced consistently for the benchmark runs via Nvidia Inspector . For an additional point is added:
Usual benchmark sequences are 30 to 60 seconds long gameplay snippets, preceded by a loading process usually. Here, a GPU Boost Technology 2.0 "gain momentum" so to speak for the benchmark and drive a part of the test with higher clock speeds thanks to the cooler through the idle GPU loading phase. However, this does not correspond to what the player will see in everyday life, since it incurred longer playing periods when the temperature rises higher and the clock continues to decrease accordingly. Therefore, would such a "standard test" hardly does justice to our claim to provide meaningful game benchmarks.

In summary, we have the GeForce GTX Titanium why shooed in four settings through our course to cover every possible scenario useful:

• Standard method with artificially limited-on the Nvidia "guaranteed" Boost clock rate of 876 MHz. Similarly, we handle it since the GTX 670 These values ​​are also the basis of our tests represent ("@ 876 MHz")
• Free boost development on our open test stand with enough cooler air ('dyn. Boost ")
• Individually applied to the minimum clock rate at 28 ° C inlet air is warm, and after 30 minutes of warm-up phase at constant load in-game map set. This corresponds to the housing operation in summer temperatures ("28 ° C")

And last but not least, we use the OC potential of the map something out, by increasing the offset clock to 100 MHz via EVGA Precision, the Power Target to 105 percent and the target temperature is raised to 85 ° C - all relatively safe levels .

That is three different SKUs alone,ie,the GTX660TI,GTX760 and Geforce Titan with reference coolers. Multiple websites spanning Europe and the Americas saying that Nvidia GPU Boost was artificially boosting benchmark scores.

All these false crocodile tears over the R9 290 and R9 290X were not made into such a big thing over multiple Kepler based cards. Did we see threads on this forum specifically about Nvidia stock coolers not doing the job?? Unless I missed them that would be a no!

I also remember that I linked to one of the articles on Titan a few months ago,ie,the Hardware.fr one, in the body of another thread. People said why would anyone use a reference cooler for such a high end card anyway and you would get better cooling anyway?? It was of no importance back then. With that logic accepted I would say its the same in this case.

The simple way to solve the problems is to raise the fanspeed for any Kepler or GCN based card using a reference cooler,or just buy a card with a better cooler or install one yourself. Simples and melodrama is over.

Another thing:

http://hardforum.com/showpost.php?p=1040349752&postcount=212

Cooling with the Jet Engine stock cooler is a non-starter for me, will reconsider when QUIET alternatives come out. I already have a box with 4 previous ATI/AMD stock coolers in it ... and some NVidia coolers too of course.

This sounds exactly like an NDA meeting I was in last week. The fan was referred to multiple times as "jet engine" etc., but all performance comparisons shown were with the card in Quiet Mode. Funny stuff when you are reaching for straws.

 

[Ian Evey]

If amd r290's had a decent cooler they would have cleaned up.

The fact that it's as fast as a 780 means nothing if the card amd
Released has a flaw, people will wait for non ref cards, hitting release sales.

And WTf is with this switch for quiet or über mode ?

Firstly I don't use stupid words to describe things,
And second that is what fan profile software is for.

Word