AMD CPU shoot-out: Phenom X3 and X4 The birth of AMD’s quad-core Phenom processor was plagued with problems. After a gestation period more akin to an elephant than a CPU, the new silicon popped out puking and bawling but was clearly a bit of a runt. The problem was the now-infamous TLB error that crippled performance and reduced clock speed with the result that …
The Toms article on this seemed to draw different conclusions if I remember. They seemed to find that it offered good performance per Watt and combined with price it made a good choice.
I think their overclocking test was done differently too, with better results.
In general I've found that these kind of chips are great for building low cost servers which run software like Apache or Sybase which thread very well. I'm not sure I'd buy one for my desktop.
OK, so what is the distinct improvement as summarised in the last page if you still enable the crippling TLB workaround as described in page 1?
That should have been fixed in B3 so if you have it on you are actually strangling the CPU.
Can we actually have some results with the workaround disabled and some results on an OS that can make good use of 3 cores and not just 2 of them (hint - it is not Winhoze).
Roger Thomas points out that Microsoft says that its 'power of two' rule for CPU cores is
'... specific to Intel processors and do not affect AMD processors.'
http://support.microsoft.com/default.aspx/kb/950182/en-us
He's right, I missed it. It doesn't change my views on Phenom. Sorry.
How can you possibly call this a fair test when you have left the BIOS fix enabled??? You are effectively running a B2 stepping, and calling it a B3. Keeping this BIOS fix enabled degrades the performance by about 30%. At least disable the L3 cache in the Intel, to keep it even. What a joke.........
As the title says, which version of the Phenom x4 9850 did you use for your testing? i'm typing this comment on a 9850 black edition clocked at 3.1ghz on an Asus M3A32-MVP using a Thermal-Take Big Thunder cooler and 4x1GB Crucial Ballistix, and it gives me no problems what so ever.
With that said, i think you just gave AMD a bum rap, rerun your overclocking tests using AMD's CPU thats made for overclocking!
Mines the one with the rusted Intel logo beneath the shiny AMD logo!
WTF would you kill 30% of your performance for a benchmark? You've enabled a workaround for a problem your part doesn't have for what reason? Why not overclock the overclocking-specific part? Is the Q6600 actually the same price as the Phenoms or several pounds more per unit than many of them, as they are at NewEgg? ($209.99 American for the Q6600, $148.99 for the 8450, $165.99 for the 8650, the 8750 and the 9550 each for $195, $215 for the 9750, and $235 for the 9850 BE today). How much extra do you spend on an Intel-compatible motherboard?
Since the X3 Phenoms start cheaper than the E6550 Core 2 Duo, and the X4 ones start at a little more than the E8400 C2D, where are those comparisons? You also fail to mention that the Athlon x2 and Athlon 64 x2 lines compete just fine, price per performance, with the Core 2 Duo when you knocked the Phenoms for not having a two-core part.
AMD doesn't have the technological lead they did a few years ago, but there's no reason you can't give them a fair showing. They trail a bit in desktop benchmarks, but at street prices (in the US anyway), the performance per investment is worth considering.
Sure, Pov-ray was fancy-pants back when I finished uni in '95, it was da bomb for rendering, but now...
Testing a CPUs performance with Pov-ray shows... how good that CPU is at running Pov-ray. Do you want a prize?
hmph, back to work, drawing & rendering in Revit & Max... NOT Pov-ray! I might play some Crysis later too.
Like everyone have already stated before me. This review is biased and will be discredited until there is a new review with the B3 TLB "fix" disabled so that the system runs full speed since the B3 already fixed the TLB issue in the first place. Thus, there is no reason to leave the software workaround to test it against a processor that has their TLB enabled (Intel).
Until RegHardware redo this review, this review is just a waste of bandwidth.
I rebuilt the test system with the Phenom X4 9850 and re-ran the tests with the TLB fix disabled.
It made no difference to the test results or to overclocking so the only logical conclusion is that the TLB fix does not apply with B3 processors. Clearly it would be unnecessary.
Ref the 'disables L3 cache' point - does anyone know if this is the effect of the TLB fix? I've heard this supposition a number of times but have never had a clear statement from AMD about the problem. It also doesn't explain why AMD pulled the launch of the Phenom 9700 and ploughed ahead with the 9500 and 9600.
'You (won't) be slowing the B3 Phenom’s performance, since the BIOS switching ‘knows’ that B3 does not have the TLB and therefore does not affect performance and bypasses the TLB BIOS settings.'
Apologies for causing confusion in the first place but the figures stand.
Adrian, I quite agree that POV-Ray is a bit old and crusty - does Revit or similar have a built in benchmark that is repeatable and transparent to the world at large?
HI, if you go to http://www.lostcircuits.com and read their orginal Phenom reviews and later B3 reviews. The benchmarks for the Phenom beat any Intel cpu while running Unreal Engine 3 games when you used a Spider platform motherboard. I'm a gamer and I'm been sold, selling AMD cpus since 1998 when they blew away the Pentium 233Mhz.
Thanks for the update Leo. I read elsewhere that turning off the B2 fix at motherboard level may be necesary, but depens on your motherboard and in some cases it may be necesary to flash your BIOS with the latest version.
Out of the three or four other reviews I read they managed between 2.8 and 3GHz with a stable system. Personaly I prefer to leave things as they are.
One of the problems I've seen is finding Windows software that I might use and that properly multi-threads (having said which the previous comment seems to imply UT3 might).
Anyway, I've decided to buy a B3 stepped quad core and will run some tests with Apache, Sybase and so on. So, more server oriented tests.
The whole point of the B3 release is to fix the TLB issue. It is no surprise your results were poor and overclocking barely worked, the TLB fix forces the CPU to read system memory instead of on chip cache! Re-run the test with the "fix" off and then you will see this chip as it was intended!!!!
I have just checked my hardware supplier and see that around the £100 pound mark I can either get:
An Intel Dual Core
or
An AMD Phenom Quad Core
Forget the Mhz speed
Anyone who shells out their £100 for the Dual Core (Intel) when they can get a Quad for the same price (AMD) must be smoking crack!
Gary.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
