CPU's / Processors

Re: AMD Phenom reviews.

Baer — Fri, 23 Nov 2007 02:10:06 GMT

cman:

I am curious as to what you think AMD has plans for and what they will deliver before they change sockets for the "Fusion" series in late 2008/early 2009???

18 months ago I was planning on only building and supplying to my clients AMD. For the last year it has been Intel. I only plan on using what is out there and AMD is not in the hunt right now. A few dollars more or less is so low in the decision tree that it is not a factor. I cannot see using AMD for any performance or enthusiast rigs for some time. Making the best of it press releases do not a CPU make. From what I see if the Phenom were boosted to compete with the upper 60% of the present Intel line the power consumption would be through the roof. I would like to see AMD come back and really compete but alas, it looks like it will not be in 08.

AMD and Nvidia do the double-green shuffle

ThePredator31 — Tue, 20 Nov 2007 20:24:51 GMT

http://www.theinquirer.net/gb/inquirer/news/2007/11/21/amd-nv-green-double-shows

For all you people like me looking for news about 780a and phenom stuff.

Re: AMD Phenom reviews.

digitalpunk — Tue, 20 Nov 2007 00:09:08 GMT

In the case of the Q6600 the cache is 4mb x2. Essentially 4mb of cache for each dual core processor. The two processors do not share cache between each other in the same way that Phenom does. That said it'd be nice if Phenom didn't suck the big one so badly. I was honestly expecting the same thing that happened when K8 was released. Does anyone else remember how badly intel was behind with netburst? Hell even if I do have the chance to upgrade to Phenom FX's I'm not sure it's worth it. I may just sit on this platform for a few more months and then move on.

Re: AMD Phenom reviews.

Tallon41 — Mon, 19 Nov 2007 23:29:10 GMT

prtuc2:
The price of the phenom kind already gave hints on the performance. I don't know how much longer AMD can use that same socket if they want to boost the performance.

agreed, although Intel likes to change sockets more often than the Grinch changes his socks.....I think this one has run it's course.

VERY lackluster performance.....and 1400MHz memory.......? it's server memory so need Fully Buffered......667 is the fastest right now on newegg.

Tallon41

Re: AMD Phenom reviews.

RjBass — Mon, 19 Nov 2007 23:19:50 GMT

This just really f***ing sucks.

I really thought AMD would come out swinging with the Phenom release. I didn't expect them to flat out beat Intel, but I did expect a good fight.

The only advantage we have with the Phenom is backwards compatibility with the AM2 socket, but even that is debatable as most AM2 boards most likely won't have proper bios updates for the Phenoms for several months at best.

I have kept myself in AMDs corner for to long now I think. It's time to move on.

Re: AMD Phenom reviews.

prtuc2 — Mon, 19 Nov 2007 22:00:55 GMT

The price of the phenom kind already gave hints on the performance. I don't know how much longer AMD can use that same socket if they want to boost the performance.

Re: AMD Phenom reviews.

cman — Mon, 19 Nov 2007 20:53:42 GMT

I am curious as to what you think AMD has plans for and what they will deliver before they change sockets for the "Fusion" series in late 2008/early 2009???

Re: AMD Phenom reviews.

a59cheffy — Mon, 19 Nov 2007 20:43:37 GMT

As a long term minded consumer, the AMD Phenom makes more sense than Intel right now. If I wanted to upgrade with Intel today I would have to buy a processor that would be socket obsolete in a month. AMD's Phenom makes more sense to me in the long run with the AM2+/AM2 socket.

Re: AMD Phenom reviews.

cman — Mon, 19 Nov 2007 20:24:04 GMT

Ars Technica

AMD's Phenom staggers out of the gate to a lukewarm reception

By Joel Hruska | Published: November 19, 2007 - 01:07PM CT

AMD officially took the wraps off its new Phenom brand today, launching the new desktop quad-core at speeds of 2.2GHz and 2.3GHz (Phenom 9500 and 9600, respectively). Contrary to earlier rumors, the processors will reportedly be available in quantity, though it may take a few weeks to see whether or not that availability stretches down to the retail channel.
The Phenoms were launched as part of AMD's "Spider" enthusiast platform, a platform that also includes the ATI-designed Radeon HD 3800 series GPUs (aka the RV670, which we'll review shortly) and the new CrossfireX-capable Series 7 chipset based on Socket AM2+.

Pricing on the two Phenom chips is set at $251 for the Phenom 9500, and $283 for Phenom 9600, both in 1,000-unit quantities. Intel's Q6600, by comparison, is currently $260 OEM ($279 retail) at NewEgg. Current benchmark results from Anandtech, HardOCP, and Hexus indicate that the Phenom, while notably more efficient than Athlon 64 X2 in certain scenarios, still lags the Q6600 clock-for-clock. Phenom may have finally given AMD the ability to offer a quad-core processor to compete with Q6600, but Intel's chip still holds a better price/performance ratio.

Before I make any other comments regarding Phenom, I want to make it clear that getting Phenom out the door is a net positive for AMD, and does improve the company's competitive position vis-à-vis Intel. With that said, however, there are significant reasons to be concerned about Phenom, AMD, and the company's future.

Launch troubles

The run-up to Phenom's actual launch today has been an enormous mess. Under normal circumstances, a company with a major product on the way is able to give concrete information regarding that product, its availability for review, and its specifications two to four weeks before the actual launch. In contrast, the information coming out of AMD the past four to six weeks regarding Phenom's launch has been simultaneously vague and subject to change. Originally, Phenom was supposed to debut at 2.6GHz with a 2.8GHz part in December. On November 1, AMD "confirmed" that it would launch Phenom at 2.2, 2.3, and 2.4 GHz, with a 2.6GHz part available before the end of 2007. In the 18 days since that presentation was given, AMD did an about face and decided that 2.3 GHz would be the fastest chip available at launch, with a 2.4GHz and a 2.6GHz part available in Q1. Accurate numbers are always better than paper launches, but this sort of thing should've been settled weeks ago.

Discussions on sampling were similarly vague. Yes, chips would be available for testing, at some point in the future, but no clear guidance was given on when, exactly, that might be. Then AMD came up with a brilliant idea, and offered would-be Phenom reviewers the chance to fly to Tahoe on an all expenses paid trip where they'd be given the opportunity to review Phenom in a closed environment on a system prebuilt by AMD. In order to help things along, AMD also offered a recommended set of benchmarks at hand. Not only is this an extremely poor use of money for a company bleeding red ink, it's precisely the wrong move for a company trying to build any sort of confidence in a new product.

Ars was offered a chance to fly to Tahoe and refused it, as did Anandtech. This turns out to have been a very good move. According to reports on the event, at least some of the systems AMD offered up for testing were clocked at 2.4 and 2.6GHz, rather than the actual launch speeds of 2.2 and 2.3GHz. We urge readers to check the "fine print" on individual site reviews for information on whether or not the reviewer went to Tahoe, and to keep it in mind when comparing results. Any website demonstrating 2.4-2.6GHz results without specifically stating that they didn't attend the AMD event almost certainly did—and while that's not a crime, it casts a shadow on the validity of benchmark tests.
The verdict on "native" quad-core: an "own goal" for AMD
There's more to be concerned about than just a badly managed launch. Although the recent discovery of an errata in Phenom's TLB (Translation Lookaside Buffer) when running above 2.4GHz may explain part of Phenom's seeming inability to scale, AMD doesn't appear to be capable of shipping higher-speed Barcelonas in significant quantity, either. As of this writing, NewEgg isn't carrying any Barcelona 1.7-1.9GHz processors and they've been out of stock more often than not over the past month. There's also the issue of clockspeed—if AMD sticks to its current roadmap, we won't see a 3GHz Phenom until sometime in the second quarter of 2008. That's pretty grim news, considering that Intel's Penryn shows every sign of being able to scale upwards whenever Intel wants it to. Sunnyvale, on the other hand, apparently needs to move heaven and earth to squeeze another 100-200MHz out of its next-generation CPU.
Finally, and perhaps most significantly, there's the question of whether or not AMD's decision to push quad-core across its entire product line was actually a good one. While CPUs like the Q6600 (and the new Phenom series with a price cut) can be a great value for end-users who run software capable of using all four cores, the vast majority of software simply doesn't do this and probably won't for some time to come. Furthermore, AMD's decision forego the easier multichip module-based approach to quad core (a la Intel) in favor of waiting to produce a "native" quad-core design seems to have cost the company a great deal of money and time without gaining AMD any performance in return, as demonstrated in Anandtech's cache coherency tests. While the "native" approach didn't seem to buy AMD a performance advantage vs. Intel's MCMs, it did manage to delay AMD's introduction of a competitive quad-core part while apparently depressing yields at launch. (Sunnyvale's upcoming tri-cores are a novel way of making good product out of bad, but they also represent just how much it costs for AMD to throw away four cores when one of them is bad.) So "native quad-core" is truly an own-goal for AMD, and its delayed arrival provided just the opening that Intel needed to regain valuable momentum in the server space.

As I said earlier, Phenom's launch does improve AMD's overall competitive position, but not nearly as much as the company needs. With tri-cores and a native dual-core design still reportedly coming in early 2008, Sunnyvale still has a chance to demonstrate a higher-scaling, more competitive product. It's also possible that we might see dramatically better scaling from Phenom than currently forecast—such things have happened before in the CPU industry. One way or the other, Phenom's competitive position must improve, and improve soon. If it doesn't, the ~$700 million dollar cash infusion the company received last week will do little more than buy the company some additional time.

Re: AMD Phenom reviews.

ish718 — Mon, 19 Nov 2007 18:38:25 GMT

prtuc2:
Perhaps jamming the L3 cache wasn't a good idea that why none of the intel desktop cpu use L3 except the server cpu has L3. Too much going on and yet nothing come out right. Native quad sounds promising at first, now the fact been proof.

A native quad core on 65nm with 8mb shared L2 cache would have been great which is basically a native core2quad but AMD isn't good with 65nm and L2 cache thats why 65nm Brisbanes have slower and smaller L2 cache.

Re: AMD Phenom reviews.

prtuc2 — Mon, 19 Nov 2007 18:00:43 GMT

Perhaps jamming the L3 cache wasn't a good idea that why none of the intel desktop cpu use L3 except the server cpu has L3. Too much going on and yet nothing come out right. Native quad sounds promising at first, now the fact been proof.

Re: AMD Phenom reviews.

cman — Mon, 19 Nov 2007 16:09:31 GMT

You are correct ish718. Generally speaking as the cache moves farther away, the latency increases. Registers --> L1 --> L2 --> L3 --> RAM --> Hard Drive. Each stage progressively takes longer to access (latency) than the previous stage.

The difference between the last stage of cache and accessing the RAM is still nowhere near as small as the difference in going from L1 to L2 or even L2 to L3 but it is enough to make a difference in performance as we have seen.

Re: AMD Phenom reviews.

ish718 — Mon, 19 Nov 2007 15:13:53 GMT

WOW, this is a huge dissappointment!, so much for NATIVE QUAD CORES

Amd needs to cut the c**p and hire some new engineers, sh** do something.

Hmm, I believe this performance difference to be cache related.

Core2Quads have a total of 8mb L2 shared cache while Phenom x4s only have 512kb of L2 cache for each core and the 2mb L3 shared cache.
L3 cache has higher latency than L2 cache from what I understand.

Re: AMD Phenom reviews.

DeadCheckR — Mon, 19 Nov 2007 15:08:00 GMT

I love both AMD for their past history and helping make Intel what is today as well as Intel for stepping up and providing performance product because of AMD.

So, I've read all the reviews and I would state the following objectively:

1. Going the "true" quad core route and chip design has definitely hampered AMD in the short term. Had they gone the same route as Intel in design and manufacturing, things may be a little different in the short term. I believe that next year will truly tell if AMD's path will pay off. Tortoise and Hare situation.

2. Because of going this route AMD has "positioned" these chips in the value segment. Makes sense as the performance/price ratios do match up. This however, cannot be the bastion of their existence for long. They need to have something in the upper echelon to truly compete with Intel. More to come on this in Q1 2008 with higher speed Phenoms.

3. I would like to see these chips perform in an environment with 3.0. Seeing as their limited to 2.0 by their testing enivronment and availability, their true performance cannot be shown.

4. Socket compatability. While this is a very nice bonus it would mean a heck of a lot more if there were more AM2 boards out there. The success of intel in the past year has seen to it that there aren't as many out there as could have been.

Just some observations.

Re: AMD Phenom reviews.

prtuc2 — Mon, 19 Nov 2007 12:47:10 GMT

AMD claims Phenom to be so phenomial and yet this benchmark tells me they are really far behind from Intel.