Original Link: http://www.anandtech.com/show/2651
Retesting the Radeon HD 4830: Too Few SIMDs Enabledby Derek Wilson on October 25, 2008 12:00 AM EST
- Posted in
UPDATE: As certain as it seemed AMD was that no 560 SP 4830 boards had made it into the retail channel, it appears that they were, in fact, wrong. Fewer than 400 HIS Radeon HD 4830 boards made it into the channel. Here is the statement we recieved from AMD's Jay Marsden today:
AMD has identified that, in addition to reference samples of the ATI Radeon™ HD 4830 boards sent to media with a pre-production BIOS potentially impacting the card’s performance, a very limited number of ATI Radeon™ HD 4830 boards were released to market with the same pre-production BIOS. This is in no way hardware related, and an updated BIOS fully resolves the performance limitation.
Through consultations with AMD board partners, it has been determined with a high degree of certainty that fewer than 400 ATI Radeon™ HD 4830 boards from one AMD board partner, HIS, have reached the market with the pre-production BIOS incorrectly provided by AMD. As only a small number of HIS-branded ATI Radeon™ HD 4830 cards are impacted, we ask any customers that purchased an HIS-branded ATI Radeon™ HD 4830 to test the board using the GPU-Z utility (available at http://www.techpowerup.com/gpuz). If the GPU-Z utility reports fewer than 640 shaders, please visit the HIS website for information on how to update the card BIOS via a downloadable install utility.
We had thought that retail parts would be fused to a certain number of active SIMDs to avoid soft mods and that review samples were probably based on test/engineering hardware AMD was using when developing the final specs for the Radeon HD 4830. It seems that theory is out the window, but we will continue to provide updates as we learn more.
Thursday, AMD launched their newest RV7xx variant completing their line up in time for the holiday season. As is usual with a new graphics card launch, we published a review. The hardware looked good and offers gamers that really want to play games with good quality but don't need high resolution performance a solid money saving option.
But a number of curious things happened throughout the course of the day yesterday. We first noticed an interesting article from the makers of GPU-z. Techpowerup.com posted a story on the fact that their reference sample from AMD only enabled 560 SPs rather than the full 640 we were all promised.
We were certainly intrigued by this, so we picked up the updated GPU-z and it told us we also only had 560 SPs. Now, this isn't a common occurrence at all, and multiple other issues could have been at work. We wanted to hold off on commenting until we could learn what was really goning on and bring you the whole story. It took us a while to sort everything out, and with this issue out there it's understandable why AMD needed some time to track down the causes and potential effects of this problem.
All the details we have point to the missing SP problem being limited to review samples only. AMD's partners should not have this issue pop up in the wild. This is good news for consumers, and good news for AMD as well. We saw that the 4830 wasn't a bad part, but because our tests were run with 87.5% of the full compliment of SIMDs our numbers don't reflect the full performance potential the hardware has.
AMD got us an updated BIOS for the card and we have new numbers showing the performance improvement. We have updated our tests from the other day to better reflect the relative performance end users can expect.
It is also worth noting that AMD has supplied all their partners with an updated BIOS as well to cover the just-in-case scenario. They believe they know why this happened and so believe that there is no cause for concern over retail parts exhibiting this problem. But giving their partners a new BIOS is a good idea anyway as it covers all the bases.
So Why DID This Happen?
We asked, and AMD answered. They did not test the review samples before they sent them out to reviewers. We can usually expect to recieve boards that have not been fully QA'd, as that can take a while, but when ever we get new boards or software companies tend to make sure that what we are getting works right. Apparently AMD was in such a rush to get reviewers parts for launch that they didn't have time to run even a basic check after the cards came back from the factory (which ever company they use to build their reference boards).
So the answer is that they were in a hurry and assumed that the correct BIOS would be installed. It wasn't and they didn't catch it.
Now, apparently there were some review samples that had the correct BIOS on them. We aren't sure who received those samples, but AMD indicated that it was based on how early the sample was sent out. Those who got later batches were more likely to have boards with the correct BIOS.
So why won't this happen in the wild? Because AMD's board partners all QA the boards they sell and because they all had a different version of the BIOS (one that functioned correctly) from the beginning.
We've been doing this for a long time, and there are times where an engineering sample or some pre-alpha something or other will have major problems. The closer it is to ready the less disastrous the testing experience tends to be. Sometimes review hardware has big issues too, perhaps with power saving optimizations or fan control. We've have stability issues on plenty of review samples.
These type of problems tend to be easily noticable, and can usually be fixed or worked around. But we tend to know what's wrong (or even that there is something wrong). In most cases these issues are taken care of before hardware makes it into the hands of end users. In this case, there shouldn't be anything for end users to worry about either.
But this is still sort of a big deal. Not because it impacts the hardware people will buy, but because it invalidates the evaluations of many of the reviews that went live on launch day. You only get one chance to make a first impression, and the diminished performance from these errant review samples could leave a false impression on those who are most likely to keep up, read, and recommend graphics hardware.
We were softer on the Radeon 4830 than on the other RV7xx hardware we've reviewed. Yes, we still saw it as a part that offers a lot of value, but the price competition from NVIDIA has been heavy and there are more powerful cards that can be had for maybe $30 more with all the mail in rebates going around.
So what do we think now that we've seen the newly flashed Radeon HD 4830 in action? Well let's take a look at the numbers first.
Performance Improvement with 8 SIMDs
The maximum theoretical performance difference between the two configurations is 14.3%. We re-ran our tests at 1680x1050 for all games but Crysis and Age of Conan which we re-ran at 1280x1024. We computed the percent increase between our previous data and the new numbers and plotted them on the following chart. To get a better sense of the significance of the performance increase, we set the maximum on the x-axis to 15 (to reflect the maximum performance increase we could possibly see).
We did re-run all of our numbers, but rather than reporting them all here, we have just updated our previous article with the corrected data. This shows that there was indeed an impact, but that it wasn't quite as large as the theoretical maximum. Because the only difference between this and the 4850 is clock speed an the number of SIMDs, this would indicate that there while more compute and texture hardware does improve performance, the performance of AMD hardware doesn't scale linearly with SPs. NVIDIA hardware does come closer to scaling linearly with SPs as we've seen in past tests (and as evidenced by the GTX 260 core 216), but the architecture is a bit different. It is possible that the bottleneck in the games we tested is just elsewhere and we would see more linear scaling on more shader intensive code.
In any case, while the increases are significant (in most cases), the impact on our conclusions isn't huge. It doesn't fundamentally change the class of the hardware.
Well this has been a first. We never see review hardware that shows worse performance than we can expect: companies are always trying to pump things up when they send them to us. For instance, we very often get overclocked versions of graphics cards rather than the stock versions. I can't count the number of times I've had to re-run tests after discovering that I needed to underclock the review sample.
While the discrepancy isn't as large as it could have been, it does seem to have affected DX10 or shader intensive DX9 more heavily.
After re-running all of our test, we updated the original article with the new charts and data. The final conclusion really doesn't change that much, but we did tweak things a little bit. There will still be a heavy emphasis on price and finding the lowest one possible when deciding between the Radeon HD 4830 and the GeForce 9800 GT. But the Radeon 4830 does look a little better in this new light and there is less of a threat from the myriad overclocked 9800 GT versions.
The performance of the Radeon HD 4830 pushes up a little closer to the Radeon HD 4850, and that's a good thing as well. The 4850 is a great option, and for those who can spend the money to pick one up we recommend it. But the point of introducing lower priced models with good performance is to provide for those who can't afford the higher end hardware. For those who can't afford more than a Radeon HD 4830 (which can be had for as low as $110 after mail in rebate if you shop for it), AMD has provided hardware that is much more capable under current games than what we are used to seeing at this price point.
Today, for gamers with 1280x1024 and 1680x1050 sized monitors, the 4830 will provide a quality experience with maximum settings in most current games. The advantage of the Radeon 4850 will be that it could provide a bit more longevity with the capability to run higher settings in future games. While that is desirable it is very hard to predict what future games will perform like on any hardware.
We can sort of look at it like a game console. The hardware is capable of a certain quality level and resolution and while quality may improve a bit with experience over time, it is limited at a certain threshold. Your current and future games will always look as good as they look today, but the advancement in other games will make the cool new effects that are around now look much dimmer by comparison. What you know you are getting when you purchase a Radeon HD 4830 is hardware that will be capable of delivering at least the visual fidelity you see on these current games with resolutions and settings we tested. And that's pretty darn good if you ask us.