Wolf Akela
Member
Well ouch. Beaten on all benchmarks? I don't think we'll see a price drop on SB when BD releases. I guess I can safely get an SB before launch.
Trinity is the Fusion successor to the recently launched Llano APUs. AMD has publicly acknowledged Trinity and said the new hardware will be available in 2012. They've also said that the performance is roughly 50% faster than Llano. The Trinity results I've seen under Linux have been impressive, but tough to compare to my current Llano hardware due to other hardware differences. The Catalyst graphics support seems a bit premature, but we're still months away from the actual Trinity launch. The compute performance is at least in order.
One of the AMD Linux engineering systems for Trinity is running nicely even on Ubuntu 11.04 with the Linux 2.6.38 kernel. The CPU string is AMD Eng Sample 2M252057C4450_32/25/16_9900_609 and its graphics are the Trinity Devastator Mobile with 512MB of video memory and an AMD Pumori motherboard. The PCI ID on the Trinity Devastator appears to be 0x9900. This Trinity APU is quad-core and running at 2.50GHz. The current quad-core Llano offerings are clocked at 2.6GHz (A6-3650) and 2.9GHz (A8-3850), while this Trinity part is clocked slower, it's numbers are nice compared to my A8-3850 Linux system.
To end for now, AMD Trinity should be very nice and I'm glad to see it already running on Linux -- it's a much better situation than when already released AMD hardware would fail with Linux. (When the first AMD Phenoms were released there would be kernel panics and other problems with the latest Linux code at the time.) Stay tuned for more.
That would be pretty ridiculous, Bulldozer has been part of AMDs roadmap for years. It was heavily delayed during development, and delayed again before launch (was planned for June), and now they still have issues and technically need to delay it again?McHuj said:It's starting too look like AMD has some sort of bug with Bulldozer and this release is a "oh shit we have to ship something this holiday season" and a fix won't come until next spring.
If the leaks are true, this isn't good on the low end either.jonremedy said:AMD's getting really good at the low-end, APU-style processors, but are still struggling hard to reach Intel levels on the mid-to-high-end. Let's hope Piledriver will be more than just catching up to Sandy Bridge, though this is exactly what it looks like now.
chaosblade said:That would be pretty ridiculous, Bulldozer has been part of AMDs roadmap for years. .
Esperado said:I feel like there must be some Windows update or AMD driver that hasn't been released to the public yet. If Bulldozer is really this bad, then it would have made more sense for them to just switch Phenoms over to the 32nm process, and then delay or cancel the chip altogether. Who knows... maybe it's going to end up being primarily a server chip.
Should be a winner in the professional / server space.Esperado said:I feel like there must be some Windows update or AMD driver that hasn't been released to the public yet. If Bulldozer is really this bad, then it would have made more sense for them to just switch Phenoms over to the 32nm process, and then delay or cancel the chip altogether. Who knows... maybe it's going to end up being primarily a server chip.
With all of the negative news we've heard over the past few months, I don't think that it's really all that "ignorant" to expect the worst at this point.belvedere said:Until the major sites release their in-depth impressions, it's pretty ignorant to make a final judgement.
http://www.amdzone.com/phpbb3/viewtopic.php?f=532&t=138852&p=212152#p212152Actually, we already have such an issue known for Bulldozer, and NO bench-marked system has the patch installed!
The shared L1 cache is causing cross invalidations across threads so that the prefetch data is incorrect in too many cases and data must be fetched again. The fix is a "simple" memory alignment and (possible)tagging system in the kernel of Windows/Linux.
I reviewed the code for the Linux patch and was astonished by just how little I know of the Linux kernel... lol! In any event, it could easily cost 10% in terms of single threaded performance, possibly more than double that in multi-threaded loads on the same module due to the increased contention and randomness of accesses.
Not sure if ordained reviewers have been given access to the MS patch, but I'd imagine (and hope) so! Last I saw, the Linux kernel patch was still being worked on by AMD (publicly) and Linus was showing some distaste for the method used to address the issue. One comment questioned the performance cost but had received no replies... but you don't go re-working kernel memory mapping for anything less than 5-10%... just not worth it!
[PATCH] x86, AMD: Correct F15h IC aliasing issuedusty1974 said:There is a problem with the two-way 64 kB L1 shared instruction cache. This IC aliasing issue was discussed in the Linux Kernel list back in July, and a patch by AMD was submitted. Linus Torvalds was not happy with the patch, and I do not know whether it has been patched into the kernel or not.Smartidiot89 said:So this is a patch that needs to be applied to the operating system? I knew about the scheduler issue but not this...
The following Linux Kernel list archive is one of the more readable ones, and you could find the whole discussion at the bottom of the page:
http://www.gossamer-threads.com/lists/linux/kernel/1409305?do=post_view_threaded#1409305
I do not know if this IC aliasing issue manifests itself in Microsoft Windows as well. The performance impact, according to AMD, is as follows:
"Up to 3% for a CPU-intensive style benchmark, and it can vary highly in
a microbenchmark depending on workload and compiler."
As such, there does not appear to be a magic OS patch here that would suddenly increase the performance by 20% or more.
I am waiting for the official Opteron Valencia and Interlagos launch, but I have lowered my expectations.
Cheers.
[July 2011]
From: Borislav Petkov <borislav.petkov [at] amd>
This patch provides performance tuning for the "Bulldozer" CPU. With its
shared instruction cache there is a chance of generating an excessive
number of cache cross-invalidates when running specific workloads on the
cores of a compute module.
This excessive amount of cross-invalidations can be observed if cache
lines backed by shared physical memory alias in bits [14:12] of their
virtual addresses, as those bits are used for the index generation.
This patch addresses the issue by zeroing out the slice [14:12] of
the file mapping's virtual address at generation time, thus forcing
those bits the same for all mappings of a single shared library across
processes and, in doing so, avoids instruction cache aliases.
It also adds the kernel command line option
"unalias_va_addr=(32|64|off)" with which virtual address unaliasing
can be enabled for 32-bit or 64-bit x86 individually, or be completely
disabled.
This change leaves virtual region address allocation on other families
and/or vendors unaffected.
Signed-off-by: Andre Przywara <andre.przywara [at] amd>
Signed-off-by: Martin Pohlack <martin.pohlack [at] amd>
Signed-off-by: Borislav Petkov <borislav.petkov [at] amd>
more...
Monstru (lab501 reviewer)'s Q&A:
Q: Why does Lab501 get to break NDA?!
A: In order to break an NDA you have to sign one first...
Q: I suppose there is also a small chance that all the final BIOS updates for the BD won't be released until the day of the launch
A: Actually no, there is no chance, because you need reviewers to have time to test this stuff before publishing a review. Nobody makes reviews in one day...
Trust me, we retested many times, because we found this hard to believe even if the results unveiled under our eyes. We checked everything countless of times, and still get the same results. Unfortunately, this really is how Bulldozer is performing, and that makes me as sad as any other enthusiast in this industry!
Q: Monstru on AMD's "damage control":
A: Now the saddest thing is not only the performance, but also the way in which AMD tries to manage this. I heard from a little birdie that some folks at AMD will start calling press tomorrow morning to ask them how reviews are going and try to do some damage control (this reminds me of Nvidia calling press before GTX480 launch). Actually many of the press guys I talked to are a little bit puzzled and don't really know how to approach this situation. From my point of view it is pretty clear, the truth (no matter how much it hurts) is the only way.
Bulldozer is AM3+, that board is only AM3.Fëanor said:It probably has been asked somewhere in the thread, but I have this mobo from Gigabyte GA-MA770T-UD3P, will it support AMDs new line of CPUs?
Unknown Soldier said:Clock-for-clock, Sandy Bridge is undeniably faster than Bulldozer, but if AMD prices it cheaply enough and it's a decent overclocker, it will find the same niche that Phenom II X6 found. People who only care about pure performance will still buy Intel. In other words, business as usual.
Got to play with a review sample. Not mine. I'm not NDAed.
The figures above posed by w00key do seem legit, but they're worst cases. In most games, BD trades places with a 2500K, in my own Fallout New Vegas testing, it beat my Phenom II X4 (3.7 GHz) by around 20%. The 2500K beats my Phenom II by the same 20%.
It's hilariously overclockable. Jury-rigging the biggest heatsink I could find, a ~7 year old Coolermaster Hyper6+ (which doesn't fit AM2/3) hacked into a pretty cheap Gigabyte GA990XA-UD3, I got 4.85 GHz out of it. At that kind of clock, the 2500K was looking for its parents with tears in its eyes, losing out on single threaded benchmarks by 10% and multithreads.... well, it wasn't really funny anymore. I'll share with you this:
Code:Code: x264 encoding 1080p BD 4.85 12:33 2500K 3.8 18:56 PII X4 3.7 21:30 720p BD 4.85 8:40 2500K 3.8 11:01 PII X4 3.7 15:15
* I'd like to have overclocked the 2500K more, but I was limited to a pretty shitty motherboard which looked to be an Asus but had no identifying marks and Intel's stock cooler.
** x264 June 2011 used. VirtualDub passing through AVISynth to the CLI x264 binary. VD was using a MoComp deinterlace filter which scales linearly to, according to the author, at least 24 cores.
It's competitive. For some workloads, you're going to want Intel. For others, AMD at last can out-perform Intel. I didn't have a 2600K to test, but if I had, I'd imagine BD would have beat it.
Of course these are all overclocked figures. Blame contractual loopholes.
Edit: Oh yeah, not allowed to say which model BD I was playing with. Also, to address the below, I couldn't boot the 2500K box with a multiplier any higher than 38. Even with scary voltages. I blame the Hybrid-EFI BIOS or some quirk of the board.
Especially since it seems to support sse4.1 and similar shit FINALLYZoddGutts said:Hmm might get this if overclocking is crazy high. Will work well with game emulators especially with Dolphin.
I really, really don't think price means anything. Phenom II competed with socket 775 parts and the 955 debuted at $250. That wasn't a competitive price at the time, which is why the price dropped pretty quickly.eastmen said:clock for clock doesn't matter , what matters is what the clocks are.
the i5 2500k is 3.3ghz and the i7 2600k is 3.4 ghz . Bulldozer is 3.6ghz for the FX 1850 .
Amd is trying to price this between the two. I don't think they would do that if performance wasn't between the two. So i'm waiting to see what goes down.
herod said:Oof, that's a lot of heat for no clear benefits. When is Ivy Bridge due?
herod said:Oof, that's a lot of heat for no clear benefits. When is Ivy Bridge due?
chaosblade said:Leads me to wonder why they didn't just clock the things higher out the gate if they have so much potential.
.
chaosblade said:I really, really don't think price means anything. Phenom II competed with socket 775 parts and the 955 debuted at $250. That wasn't a competitive price at the time, which is why the price dropped pretty quickly.
And that OC'd performance is pretty good, but comparing it to a 2500k that is over 1GHz slower doesn't tell you anything. Leads me to wonder why they didn't just clock the things higher out the gate if they have so much potential..