Higher Clock Speeds, No TLB Issues and Better Pricing: The New Phenom
by Anand Lal Shimpi on March 27, 2008 12:00 AM EST- Posted in
- CPUs
When Phenom launched it was slow, later plagued by a performance-hindering TLB bug and priced entirely out of the realm of rational thought. It's a new year and while I'd like to say that AMD has learned from all of its mistakes and will be back to a fully competitive state, that's only mostly true. There are still significant struggles within the company but at least this launch is a step in the right direction.
Today's launch is actually much bigger than Phenom's original debut, encompassing a total of seven new processors:
| Cores | Stepping | Clock Speed | TDP | L2 Cache | L3 Cache | 1 Ku Price | |
| AMD Phenom X4 9850 | 4 | B3 | 2.5GHz | 125W | 2MB | 2MB | $235 |
| AMD Phenom X4 9750 | 4 | B3 | 2.4GHz | 125W | 2MB | 2MB | $215 |
| AMD Phenom X4 9750* | 4 | B3 | 2.4GHz | 95W | 2MB | 2MB | $??? |
| AMD Phenom X4 9650* | 4 | B3 | 2.3GHz | 95W | 2MB | 2MB | $??? |
| AMD Phenom X4 9550 | 4 | B3 | 2.2GHz | 95W | 2MB | 2MB | $195 |
| AMD Phenom X4 9100e* | 4 | B2 | 1.8GHz | 65W | 2MB | 2MB | ~$200 |
| AMD Phenom X3 8600* | 3 | B2 | 2.3GHz | 95W | 1.5MB | 2MB | ~$175 |
| AMD Phenom X3 8400* | 3 | B2 | 2.1GHz | 95W | 1.5MB | 2MB | ~$150 |
Ok, let's see if we can make sense of this. There are three new CPUs that you'll see at places like Newegg: the Phenom X4 9850, Phenom X4 9750 and Phenom X4 9550 running at 2.5GHz, 2.4GHz and 2.2GHz respectively. The 50 at the end of the model number means that these CPUs are based on the new B3 stepping, which includes the fix for the TLB erratum - in other words, these are the CPUs you want. The prices are also pretty reasonable, they are all finally cheaper than Intel's Core 2 Quad Q6600.
The Phenom X4 9850 is a "Black Edition" part, meaning it ships with its clock multiplier unlocked. It is also the first Phenom to run its L3 cache/North Bridge/memory controller at 2.0GHz and not 1.8GHz like the rest of the Phenom lineup.
Here's where it gets complicated. AMD needed something to do with all of its B2 stepping Phenoms, so it's selling those to OEMs who don't really seem to care about the TLB bug. The CPUs with a * next to them are OEM only; although that doesn't mean that they won't appear in retail, they aren't intended for end user purchase.
The Phenom X4 9100e is AMD's first 65W TDP quad core CPU thanks to a relatively low operating frequency of 1.8GHz. Unfortunately it's plagued by the TLB bug since it's a B2 stepping core, so you have to take into account that its performance may suffer because of it. The same applies to the two new triple-core parts; the Phenom X3 8600 and 8400 are both quad core B2 stepping CPUs with one of the cores disabled. AMD doesn't have any plans to introduce a new, smaller triple-core die because the costs would be too great. Instead AMD wants to focus on getting its 45nm transition started before the end of the year.
AMD's plan is quite ingenious, keep the TLB bug CPUs out of the hands of the enthusiasts who will complain and use them to keep OEMs happy as well as use them for the first triple-core CPUs.
The OEMs don't completely get the shaft as there are two 50-series CPUs they get that we don't. First off there's a 95W TDP Phenom X4 9750 (most likely a lower yielding 9750 that just runs a bit cooler, hence the lower TDP) and next there's a Phenom X4 9650.
Within the next month or so, AMD will be releasing B3 versions of all of the remaining CPUs, so you'll see a Phenom X4 9150e, Phenom X3 8650 and Phenom X3 8400. In general, AMD told us to expect around a $50 price difference at the same clock speed between triple and quad core. Given that most applications still don't benefit tremendously from four cores, tri-core may be a nice way for AMD to compete with Intel's higher performing dual core options. As soon as there are B3 Phenom X3s available for review, we'll let you know how they stack up.
Facebook
Twitter
RSS
65 Comments
View All Comments
fireedo - Saturday, May 24, 2008 - link
can we call this fair comparison since AMD side using only DDR2 800 Mhz and Intel side using DDR3?can that things makes any differences in benchmarking?
if so then this comparison is not valid
sorry no offense :)
magnusr - Saturday, April 12, 2008 - link
Consider the power usage of the 9850 compared to the Q9300.If you get the 9850 for free and pay full price for a q9300, then you leave both systems running for an x amount of time, the 9850 will then after a x period of time be more expensive (considering expensive power bills).
Common AMD you can be better than this. When Athlon 500MHz came out they rocked the boat until Intel came out with core 2, amd just got lazy....
Currently running Q6600 @ 3GHz, had it for almost a year now.
js01 - Friday, April 11, 2008 - link
When i saw the crysis bench and saw the 9500 only getting 27 fps I ran the same bench with my system which consists of Phenom 9500, hd 3850, k9a2 cf, and 2gb ddr 800. My results were 53.095 fps average, 70.70 max, and 34.24 min. It just goes to show you the only benchmarks you can trust are your own.ntavlas - Saturday, March 29, 2008 - link
AMD can`t compete in the high end, there is no doubt about that. But I think their real purpose with this launch is improving their presence in the mainsteam. Since they can`t reach the clock speeds or instructions per clock of intel they use more cores. This approach does have it`s merit: in 4 threaded apps they are faster than intel`s dual cores in the price range. Of course they are much slower in 2 thread applications but this doesn`t tell the whole story: a core 2 8200 might be faster than a phenom 2,2 when running a 2 threaded benchmark, but don`t forget that the later still has 2 cores in reserve that can be used for other tasks. It`s something that can improve the users computing experience in real word situations.Of course if you can afford an intel quad core, by all means go for it, it is the better chip, but I think that up to the $200 price point the phenoms make good sense.
Thorsson - Friday, March 28, 2008 - link
Despite the lower power consumption the word I'v heard is that Intel's new 45nm Quads can't take extra voltage and consequently don't OC as well as the Q6600.It would have been nice to see Anandtech address this issue.
JustAnObserver - Friday, March 28, 2008 - link
I love seeing you morons arguing about who has THE best, clock per clock, top extreme black super duper alpha and omega overclocking part. When the enthusiast processor is more like a ferrari. Nice to show, but who really mekes the big money is Toyota and their common cars. You see MOST people won't overclock, MOST people won't CrossFire nor SLI, MOST people just want an average system for a day in day out workload. So the only thing that really mather is this:Keep struggling for the Extremely Over priced parts, so that the average parts (AMD or Intel, whatever) keep coming cheaper and cheaper every day. After all, I'll get, let's say, 75% of your performance, but pay less than 50% what you paid. Sounds nice to me.
But hey, thats just me.
Ps.: You americans are funny, you can't even realize that the US only answers for 7% of the world processor market. And the rest of the world is a bit smarter about how to spend their money.
gochichi - Friday, March 28, 2008 - link
Intel outdid itself, kind of shot too high I guess. Look at how Apple for instance hasn't released an iMac with the Q6600... if they did, it would completely mess up their line up and price points.Anyhow, AMD's "competition" is pretty lame, was I the only one to notice how far inferior the Phenom line is to the tried and true Q6600. Everybody knows it's basically a downclocked chip... can easily do 3.0ghz. The only reason Intel doesn't do that is that it doesn't fit in the market.
There is currently nothing in the horizon to handily beat a 3.0-3.2Ghz Intel Quad so they are showing very little interest in maxing themselves out.
Dell should absolutely feel ripped off, so many years trying to carry both intel and AMD only to achieve that when AMD makes relatively junky chips.
Anyone can plainly see (hindsight is 20/20) that purchasing a Q6600 the day it came out was the best value to be had in years (both before and after).
I did not, but I finally have one (whole computer w/ Q6600 was $400) and I feel very confident that its performance will be relevant for a couple of years to come still.
We are pampered to death these days... technically the Phenom is plenty but it's just not Intel-grade. Pentium 200 MMX for $380 anyone?
Furious computing power is now more affordable than ever and with power envelopes (Intel-wise anyway) that will hopefully bring the form factors down even further. Even graphics cards are being re-corrected in terms of efficiency... they were runaway for a while there.
I think that's what the 2000's are all about, nothing new, just delivering on old promises. Finally computers that approach "fast enough", worlds more reliable and affordable.
papounet - Friday, March 28, 2008 - link
The article is lame, it is utterly biased in its conclusion.The logic it tries to build around the fact that here is no speed improvment (" We get the impression that there are some speed paths that could be optimized on the current B2 and B3 Phenoms that simply aren't because of a very sensible thought process. AMD is still on track to begin shipping its first 45nm Phenom processors (Deneb core) by the end of this year and it doesn't make sense to waste time and resources respinning a 65nm Phenom, when presumably these clock speed issues are addressed at 45nm.") displays an absolute ignorance of the CPU design business. This stepping was an attempt to speed up the whole Phenom and improve yelds (and not only to correct the TLB bug). It does not at all deliver the speed bump (2.5% improvment ?). so let's claim that AMD did not try ... Ahahah.
With the pressure from Intel and Wall Street and OEM, AMD can not afford to do yet another stepping to try to fix their technology and/or their design and/or their process, so AMD wil bet on 45nm to fix everything.
Some other sentences just made me laugh "While we only had a few hours left over to test the overclocking stability of the 9850 it does look like it may be the first Phenom capable of easily breaking the 2.6GHz barrier." how difficult is this when starting from 2500Mhz ??
"In our opinion it's highly unlikely we'll see AMD release a 3.0GHz Phenom on 65nm this year. " Indeed
The Phenom with this performance level is one year too late to compete. It is useful for people who have bet on the m2+ plateform. but how many did ?
The model released at 2500 Mhz is very close to the design limits of the stepping (little overclocking room).
The Q6600 is not the best chip ever (celeron was, then the XP-mobile, then core2duo 4500), but is is rather good.
With some research it was possible to buy the Q6600 B0 just when it came out at a decent price: it was cooler , cheaper, faster than s3 stepping and most motherboard bugs around quads had been solved.
The higher fsb of the Q9300 as annouced by Intel could be already seen as limiting the potential easy overclock.
Did I need today that my q6600 bench at 3.6 Ghz on air on a GA-p35ds3p with 4 gig of crucial ballistix ? i run it on a daily basis at 3.2 Ghz at nearly stock voltage.
:-) OK not all my buys are as smart
(I am no fan of Intel or or AMD. I have used parts from both to build computers professionally and I still dabble into the hobby).
eye smite - Thursday, March 27, 2008 - link
Why am I seeing an Intel review when the article is supposed to be about Phenom? Is this the same cry baby ranter that did the original phenom review for AAT back in November? I see sticking to the facts and being professional isn't a high priority at this site, but hell I knew that already just from reading the articles on daily tech.rodh - Thursday, March 27, 2008 - link
The TLB error DOES NOT CAUSE PERFORMANCE LOSS!!!!!!!the PATCH for the TLB error causes a performance hit, but all the TLB error does is cause instability in extremely rare situations involving hardware virtualization (when did you last use that) on multiple cores at the same time.
Cant stand it when people overblow this error. Not that it matters anymore, its fixed.
Rod
http://roddotnet.blogspot.com">http://roddotnet.blogspot.com