TechRadar: All Processors Feeds

AMD Opteron 2384

Jeremy Laird — Thu, 13 Nov 2008 05:00:00 +0000

It's been a long time coming, but AMD finally has a decent quad-core chip. Launched today in Opteron server trim, AMD's new 45nm microarchitecture is far from revolutionary. But it does make for an excellent server CPU and also bodes well for upcoming desktop derivatives.

Known internally at AMD under the Shanghai codename, the new microarchitecture doesn't amount to much more than a die shrink from 65nm production technology to a new 45nm node.

Both the overall quad-core layout and the detailed architecture of the cores have been carried over pretty much unchanged. OK, the memory controller has been massaged. It now supports DDR2 running at 800MHz. The HyperTransport links have also been upgraded to 3.0 spec.

Oh, and AMD has improved some aspects of the chip's virtualisation support. Particular attention has been paid to the time taken for so-called 'world-switching'.

That's the process of shifting a virtual machine between host and guest systems. If that sounds rather arcane, virtualisation is an increasingly important aspect of server CPU performance. This stuff really matters.

Banishing Barcelona

But when you get right down to it, Shanghai is all about 45nm silicon. And the good news is that all the indications are that it's a very healthy process for AMD.

Unlike last year's troubled Barcelona launch, Shanghai is on time and hitting the ground running at decent clockspeeds. At launch, the new Opterons span a range of clockspeeds between 2.3GHz and 2.7GHz.

That might seem a little ordinary compared to the 3.4GHz clock of Intel's top Xeon server CPU. But it's worth remembering that the best quad-core server processor AMD could actually ship a little over a year ago was 1.9GHz.

Arguably even more impressive are the power ratings of the new Opterons. Even the top 2.7GHz 2384 model is rated at just 75 watts. That's a huge step forward compared to the 105 watt rating of the outgoing 2.5GHz 65nm Opteron 2360. Overall, AMD says the new 45nm process is 35 per cent more power efficient.

As you'd expect with any die-shrink, the new 45nm process also allows AMD to pack a few more transistors into the die. For the new Opterons, that takes the form of a boost from 2MB to 6MB of L3 cache memory.

Building on strengths

But what about performance? Given the carry-over architecture, we weren't expecting fireworks. However, what Shanghai does extremely well is build on the existing strengths of its Opteron family of processors.

As before, multi-socket scaling is excellent. In benchmarks that major on data bandwidth and floating point performance, a pair of 2.7GHz Opteron 2384 chips are probably the fastest dual-socket solution currently available.

Certainly they have the measure of Intel's Xeon processors running at 3.2GHz in our computational fluid dynamics benchmark. And this is just the dual-socket version of Opteron. Given AMD's traditional advantage as sockets are added, we'd expect the four-way model to have an even greater advantage.

That said, integer performance is slightly less impressive. But it's significantly closer than before thanks to the boost clocks and cache memory. The overall picture is extremely solid, therefore.

Factor in the big improvements in power efficiency - again, that's something that really matters for high density server installations - and AMD is now in a position to make a pretty compelling overall argument to server customers.

At least, that's true in the context of the current Intel competition. Early next year, Intel will roll out its impressive new Nehalem architecture in servers. We've already seen what a beast it is on the desktop in Core i7 trim. AMD had better make the most of the next few months.

Comparative benchmarks - AMD Opteron 2384 v Intel Xeon X5472

AMD Opteron 2384

Memory bandwidth 20.62 GB/s
Sandra Dhrystone ALU arithmetic 77.72 GIPS
Sandra Whetstone FPU 62.99 GFLOPS
Cinebench 10 multi-core 45secs
Cinebench 10 single-core 4mins 2secs
X264 encode 74fps
Stars CFD 30.32secs

Intel Xeon X5472

Memory bandwidth 10.4 GB/s
Sandra Dhrystone ALU arithmetic 104.57 GIPS
Sandra Whetstone FPU 83.76 GFLOPS
Cinebench 10 multi-core 37secs
Cinebench 10 single-core 3mins 44secs
X264 encode 86fps
Stars CFD 37.34secs

Intel Core i7 processor

Jeremy Laird — Mon, 03 Nov 2008 05:00:00 +0000

Today is Intel Core i7 day. You'll find lots of coverage of the new processor and the X58 chipset that supports it right here. We have everything from technical analysis so hard core it will make your brain boil to benchmarks of a literally record breaking machine based on the fastest Core i7 chip of all.

It's a ridiculously impressive new processor, so it's well worth getting to grips with all the new technology. But if what you really want is a quick guide to the chip itself - what makes it tick and how quick it is - read on.

The big news with Core i7 is improved multi-threaded grunt. To pull that off, Intel has created a much more elegantly architected CPU with added on-die features. Most of what used to be found in the northbridge chip on the motherboard has now been integrated into the Core i7's own silicon die.

That includes the memory controller and much of the system I/O. The latter takes the form of the new uber-wide Quick Path Interconnect (QPI) that can shunt over 25GB in and out of the Core i7 every second. The idea is to provide more bandwidth to feed all those data hungry execution cores.

Unprecedented levels of raw bandwidth

And boy, do those cores get fed. Thanks to the addition of an extra memory channel - for a grand total of three - Core i7 packs an unprecedented 18GB of raw bandwidth. That's literally three times as much as previous Intel processors.

Of course, AMD's CPUs already have on-die memory controllers, but with only two channels and no support for DDR3 they typically only enjoy about half the bandwidth of Core i7. Incidentally, like AMD's quad-core Phenom chip, the Core i7 is a truly monolithic, single-die processor. No longer will Intel bodge up quad-core processors by cramming a pair of dual-core chips into a a single processor package.

Along with the added on-die features, Intel has revved its latest processor with further features designed to give a boost to both old fashioned single-threaded applications and the latest highly-parallelised programs such as media encoding.

HyperThreading returns

The biggest headline grabber is the re-emergence of HyperThreading. It's a feature last seen on the Pentium 4 Netburst CPU family and allows each core to crunch two software threads simultaneously for improved multi-threading and multi-tasking performance.

Since all Core i7 are quad-core chips, that means you'll see no less than eight logical processors when you fire up Windows Task Manager. That's a pretty sexy sight.

Next up is Turbo Mode. The short version of the Turbo Mode story is the ability to overclock individual cores according to work load. The aim is to make the most of the thermal limits of the chip when only one or two cores are under heavy load. Whatever the advertised clock frequency of a gvien Core i7 is, add 266MHz to that and you have the top speed a single core can run at.

Raw performance

Intel has also tweaked the cores themselves in a number of areas, including the addition of a few new SSE instructions for extra multi-media goodness and a new three-level cache memory hierarchy.

And for the record, all those architectural changes also bring a new processor socket, known as LGA 1,366. A new motherboard is therefore a bare minimum requirement to get in on the action.

As for the specifications of the Core i7s available at launch, there are three models. The entry level chip is the 2.66GHz Core i7-920, yours for £200, or perhaps a touch less. Next up is the 2.93GHz 940 processor, likely to sell for £400.

Topping things out is the inevitably expensive 965 Extreme Edition. It costs a cool £1,000, runs at 3.2GHz and adds a few overclocking-friendly features such as an unlocked multiplier. All three chips have 8MB of L3 cache and support up to 24GB of 1,066MHz DDR3 memory.

Exceeded expectations

But what about performance? In truth, our expectations weren't all that high. Mainly, that's because Intel's existing Core 2 processors are so damn good, it was hard to imagine anything significantly bettering them much less blowing them away.

Of course, the Core i7 cares not a jot about our preconceptions. In our multi-threaded benchmarks, it duly proceeds to beat the Core i7 to a bloody, whimpering pulp. It's around 60 per cent quicker in the X264 video encode test than the best Core 2 processor, for example. That's almost silly.

While its superiority isn't always that spectacular, a performance advantage of around 40 to 50 per cent is typical in most multi-threaded applications including image rendering and media encoding of various types.

As for how AMD's Phenom compares, it's an absolute massacre. Intel's finest is now often twice as quick as the best AMD can deliver.

That said, the Core i7's single-threaded performance is a bit less impressive. It seems the individual cores aren't a major advance on the Core 2 architecture. Likewise, it's a little disappointing to note that power consumption has gone up a fraction.

We had been lead to believe that greater feature integration would have the opposite effect. Finally, HyperThreading appears to occasionally have negative effect on performance when only a handful of CPU-intensive threads are being processed.

But those reservations are pretty small beer compared to the performance smörgåsbord the Core i7 cranks out most of the time.

Intel Core 2 Duo E6850

Techradar — Mon, 21 Jul 2008 10:52:00 +0000

AMD might have been ﬁrst to market with dual-core, but there has been a lot of silicon under the bridge since those glory days.

In recent times, Intel's Core 2 Duo has been showing everyone how dual-core should really be done.

Powerful processor

The E6850 is in fact an older 65nm chip, but it is still powerful. Considering the chip only has two cores to work with, it holds up pretty well to the Phenom 9550, and is a considerable improvement over the E2180.

Compared to the Phenom, the 1333MHz bus seems glacial, but it's got 4MB of L2 cache, which certainly helps.

Video encoding, while better than the E2180, is average, and really lags behind AMD's Phenom.

The game scores are strong though and you won't have many problems running most games on this chip.

AMD Phenom 9550 2.2GHz

Techradar — Sun, 20 Jul 2008 10:56:00 +0000

It's a common aphorism that two heads are better than one, but can it be said that four cores are really better than two?

The Phenom suffers from a sloth-like 2.2GHz core clock and has just 512KB of L2 cache per core.

But, by using a lightning-fast 3,600MHz bus, and having the memory controller on the die, the Phenom has a trick or two up its sleeve.

Speedier performance

In the HD video encoding test, the 9550 actually achieved a higher framerate then Intel's top-end quad, the QX9650, and it leaves the E2180 standing.

The 3D rendering is also much better than the faster E2180, although strangely, the game scores are lower, a problem for both this and the 9850.

Still, for just a few pounds more than the E6850, the extra grunt of two more cores is money well spent.

AMD Phenom X4 9850 Black Edition

Sat, 21 Jun 2008 11:28:56 +0000

We'd all feel a little better about AMD's current chips if it wasn't for the name.

You call your processor 'Phenom' and, frankly, people expect a phenomenon. When it isn't one, the sting is all the harsher. It's like finding out a guy called Steve Massive is only five foot six.

Familiar architecture

AMD's latest consumer top-end chip is no phenomenon, but it does see the green party find its feet a little again. Superficially, little has changed since we made disparaging sniffing noises at the 9700 back in January.

The 9850 is another quad-core, 65nm chip, and still a revision of the old K8 design that the aged Athlon 64 used. We'd be fools to expect greatness, especially as the world was due the Phenom 9800 and 9900s back in November, but AMD just couldn't pull it off, plumping instead for the lower-clocked 9500 and 9600.

Still, here's a 2.5GHz model at last, with that confusing '50' suffix denoting Phenom's new B3 stepping, which tidies up a few erratums in the earlier models. Most notably, say goodbye to the TLB erratum 298 in Phenom's L3 cache, which carried a minor risk of system hang-ups, but when fixed with a BIOS update meant a small performance hit.

Speedy performer

Being all too aware of what the 9700 (itself now replaced by the 9750) is capable of, it wasn't much of a surprise to find the 9850 mostly snapping at the heels of Intel's venerable but beloved Core 2 Q6600.

Unlike Intel's low-end four-player, this is natively quad-core, not a dual-die sandwich, plus it's got an extra 100MHz on the clock, though that doesn't count for much in practice. The Q6600 draws slightly ahead in most tests, and significantly in Crysis, but the 9850 can scarcely be called a disastrous performer.

Additionally, it'll overclock on air (with a spot of volt-tweaking) to 3GHz, but realistically bank on a bump of 200-300MHz.

A rival to Intel?

As online stores offer both this and the Q6600 for as low as £150, the 9850 is the fruitiest drop-in upgrade yet, if you're already toting an AM2 platform. If, however, you're building a new system, AMD simply has nothing to compete with the higher-end Core 2s.

The 45nm, socket AM3 chips (which will work in AM2 boards) due later may close the gap, but after many Phenom let-downs, it's hard to bank on that.

So, hardly a silicon superstar and AMD remains conspicuously absent from fashionable high-end parties, but smart pricing makes this the most desirable Phenom to date.

Intel Core 2 Duo E8500

Thu, 01 May 2008 16:37:12 +0000

The new dual-core variants of Intel's 45nm Penryn CPU family look like a no-brainer. For the most part, that's how how it turns out.

The top E8500 model is clocked at an impressive 3.16GHz, which translates into single and dual-threaded performance to match any processor on the planet. It's also impressively efficient compared to the outgoing 65nm generation of Core 2 Duo CPUs from Intel.

Improved performance

Despite running a few hundred MHz faster than the old 2.93GHz X6850 dual-core chip, the E8500 consumes 30 watts less under full load.

It's also significantly quicker in every benchmark than the X6850. Part of that is no doubt thanks to the faster frequency, but the addition of 2MB of cache memory, now a borderline-silly total of 6MB, doesn't hurt. The same goes for the faster 1,333MHz bus. It's all gravy.

Of course, Intel would also claim that the various tweaks made to the Penryn architecture boost the clock-for-clock performance of its latest cores, too. You know, the new Radix 16 divider, enhanced SuperShuffle Engine and all that jazz.

In practice, it's effectively impossible to isolate each of the upgrades and deduce exactly how much benefit they bring.

Power at an attractive price

What we can say, however, is that the overall performance advantage of the E8500 over the X6850 is always above 10 per cent and sometimes as much as 20 per cent.

That's despite the clock speed differential of just eight per cent. And remember, the E8500 is less than £200. At launch a little over 18 months ago, the X6850 was the world's fastest PC chip and cost a cool £600. That's progress for you.

The good news doesn't end there. We can also report that the E8500 is every bit as good an overclocking chip as you would hope; hitting 4GHz on air cooling is a cinch. Up the volts a bit, and even higher speeds are possible.

All of which just leaves one lingering doubt. The full-fat, 12MB cache Q9450 quad-core chip can be had for just £30 more. Sure it's clocked a little lower at 2.6GHz. But for any system that's remotely multi-purpose, the Q9450 is probably a better all-round bet.

AMD Phenom X3 8750

Jeremy Laird — Fri, 25 Apr 2008 12:05:10 +0000

Put yourself in AMD's shoes. Things haven't been going to plan. Your crucial quad-core chip arrived late, ran slow and had a nasty little hardware bug. What's more, you're being beaten silly by Intel's Core 2 processors. What do you do? How about a triple-core chip to distract attention from the mess you're in?

That's the cynical take on the launch this week of the Phenom X3, the new triple-core revision of AMD's Barcelona processor architecture. And it's not entirely fair. It might be true that the X3 would not have happened if the Phenom project as a whole wasn't floundering. But it's still a smart move by AMD and presents punters with an intriguing new alternative.

Processor pitch

For starters, it gives AMD a unique sales pitch and a product that neatly transcends the direct comparisons that have been so unflattering for the initial quad-core Phenom X4 models. Indeed, it's extremely unlikely that Intel will copy it.

Intel's quad-core processors are actually composed of a pair of dual-core dies sandwiched into a single package. Creating a triple-core model would therefore require some fairly hideous hackery. But even if it didn't, such is Intel's crushing dominance, it's doubtful it would bother. Let AMD have a monopoly on triple-core products. See if Intel cares.

More to the point, by pricing the new X3 range to take on dual-core Intel Core 2s, AMD probably reckons it can offset Phenom's core-for-core performance disadvantage. If the X3 delivers more bang for your buck, what do you care whether the whole triple-core thing is a bit of a gimmick? What really matters is the performance it pumps out, not how it's achieved.

Bugged out

So, is the new Phenom X3 actually a chip we can actually recommend? Certainly we're pleasantly surprised to find that retail examples of AMD's triple-ganger sport the all important '50' suffix. That indicates they're based on the latest B3 spin of the Phenom architecture and hence free from the infamous TLB bug. As we've reported earlier, the BIOS work around AMD implemented to patch the bug typically sapped performance to the tune of 10% and sometimes as much as 30%.

It seems the "buggy" B2-based Phenom X3s that AMD announced a few months ago are destined to go only to system builders and won't be sold as boxed processors. Hence in almost regards save core count, the new X3 Phenoms are identical to the latest revision of the X4 line. In fact it's based on precisely the same quad-core processor die but with one core disabled.

With that in mind you might think the X3 is an effort to increase yields and profitability by rescuing dies with one dud core from the scrap heap. But apparently not. When we put that very question to AMD it was flatly denied.

Whatever, our first experience of the X3 comes in the form of the 8750 model. At 2.4GHz its the quickest option at launch and like other Phenom X3s packs 512K of L2 cache memory per core and a shared pool of 1.5MB of L3 cache. It's also a socket AM2+ chip and in theory is compatible with the large installed base of AM2 motherboards. In practice, we strongly recommend you consult you board's maker before assuming it's good to go for X3s.

Triple whammy

But how does it perform? For the most part, just as you would expect for a Phenom chip short on cores to the tune of one. In heavily multi-threaded applications such as video encoding or professional rendering, the benefit over competing dual-core Intel processors is clear to see. The X3 8750 is quicker than the price-parity Core 2 E6750 in such workloads.

That said, it's worth noting that some multi-threaded apps like Window Media Encoder are currently foxed by the odd-numbered core count and will only make use of a pair of the cores. It's a minor glitch that's likely to be addressed in the near future. But there are two more serious counts on which the X3 fumbles.

Firstly, in single and dual-threaded software, including the vast majority of games, it's much, much slower than the Intel competition. But even more troublesome is the close proximity of the X3's quad-core sibling in AMD price list.

Precise UK retail prices will emerge over the next week or so. But it certainly looks like the premium for an identically-clocked X4 chip will only be around £20. Add another £10 and you're in Intel quad-core territory. All of which means the X3 is an interesting new option that just needs a little shaved off the asking price to really make sense.

AMD Phenom X4 9850 BE

Jeremy Laird — Tue, 15 Apr 2008 15:57:51 +0000

The launch of the original Phenom X4 quad-core processor was blighted by a hardware bug and disappointing clockspeeds. Can the new Phenom X4 9850 Black Edition and its revised B3 stepping salvage the situation for AMD?

It's hard to know just how much AMD has been hurting following the frankly disastrous introduction of its quad-core CPU architecture. But here's one metric of its current predicament that is rather revealing. Back in 2005 AMD's entry-level dual-core chip, the relatively feeble Athlon 64 X2 3800+, commanded a £275 sticker. Today, the brand new Phenom X4 9850 Black Edition flagship quad-core model is priced to sell at £155.

Phenom's weakness

And lest you have forgotten, Phenom is a big old beast that must surely be costly to manufacture. All four cores, plus a memory controller, HyperTransport I/O and 3MB of cache memory are packed into a single 65nm processor die. Making matters worse, AMD has been having difficulty clocking the chip up to competitive speeds. Despite claims to the contrary, therefore, we suspect the upcoming triple-core Phenom X3 model is an attempt to lift production yields out of the trash can.

Intel's latest quads, by contrast, are composed of a pair of dual-core dies and are built using finer 45nm production technology. Thanks to the great per-cycle performance of the Core 2 architecture, Intel also has the luxury of not needing to push its chips to limit in terms of clockspeeds. All of which takes the pressure off yields as a whole for the stupidly successful Core 2 family.

But back to the matter at hand, namely the new Phenom X4 9850 Black Edition. Arguably the most significant aspect of this new chip, which benefits from the new B3 stepping of AMD's quad-core architecture, is the removal of the nasty little TLB bug.

TLB bug is gone...

As we've reported on numerous occasions, the TLB bug is a hardware erratum effecting the chip's translation lookaside buffer. In certain scenarios, it can cause systems to fully lock up. In reality, that's likely to be an extremely rare occurrence. But the mere possibility forced AMD to create a BIOS-based work around that disabled the faulty circuitry.

You might think that was problem solved. But in testing the BIOS fix compromised performance by around 10 per cent on average and as much as 30 per cent in isolated benchmarks. With Phenom struggling to reach competitive and the chip's rather conservative architecture not delivering competitive clock-for-clock performance, the bug added to a rather inauspicious compendium of shortcomings.

Nevertheless, the TLB bug can now be crossed off that list. With the B3 stepping it has been banished – at least for all Phenom processors with the '50' suffix in the product number. The X3 triple-core chips will keep the '00' suffix and remain TLB bugged.

But what about clockspeeds? Here the news is less gratifying. Though the new 9850 Black Edition model does raise Phenom's game, the move is from a rather pitiful 2.3GHz to merely mediocre 2.5GHz. To put that into perspective, the slowest model in Intel's latest 45nm quad-core range is clocked at the very same speed. The fastest Intel quad-core CPU now weighs in at 3.2GHz.

Excellent multi-core scaling

As for the 9850's actual performance, it's a tale of two halves. In isolation it delivers a pretty solid experience. The 2.5GHz clockspeed is just enough for decent single-threaded performance. Meanwhile, AMD's integrated memory controller ensures low latency and extremely data high bandwidth to all for cores which makes for excellent multi-core scaling.

Unfortunately, however, the 9850 does not exist in isolation. Currently priced around £155 it must compete with the entry level quad-core chip from Intel's 65nm range, the Q6600. It may only run at 2.4GHz, but the Q6600 still has the measure of the fastest Phenom in most benchmarks. Admittedly, the 9850 does match the Intel opposition in our H.264 video encoding test. But it's squarely beaten in the intensive Crysis CPU gaming gauntlet and Cinebench rendering benchmark, for instance.

Of course, all of that applies to operation at stock frequencies. As a "Black Edition" chip, the Phenom X4 9850 has an unlocked multiplier designed for simple overclocking. However, although AMD appears to have released a little more headroom on this latest quad-core model, the 2.8GHz we achieved remains a modest overclock in the context of what Intel's quad's are typically capable of. Any Core 2 Q6600 you buy today will hit speeds well in excess of 3GHz.

So, AMD's revised Phenom CPU in the form of the new 9850 model is certainly a step in the right direction. But it's simply too incremental to significantly close the gap to Intel. For that we'll have to wait until AMD releases its upcoming 45nm shrink of the Phenom later this year. By then, of course, Intel is expected to be shipping its brand new quad-core chips based on the ominous looking Nehalem architecture. It's not going to get any easier for AMD.

Intel Pentium Dual-Core E2180 2GHz

Sun, 10 Feb 2008 16:34:36 +0000

Spare a thought for this feeble, emaciated little chip. The E2180 potters along at just 2GHz, packs just 1MB of cache and suffers a comparatively glacial 800MHz bus. Glacial, that is, compared to the 3.2GHz of Intel's own Core 2 Extreme QX9770.

Surprisingly capable

Funny thing is, however, even at stock clocks, this processor delivers a decent experience. Thank the fact that it's based on Intel's Core 2 architecture for that, despite the 'Pentium' branding.

Consequently, it has just enough juice for gaming and hi-def video. But the way this beastie overclocks is something else. How about over 3GHz using an air cooler and stock voltages? This chip cranks out performance comparable to any of Intel's pricier dual-core models. And all for a whiff over £50. Impressive.

Intel Core 2 Extreme QX9770 3GHz

Sun, 10 Feb 2008 10:02:44 +0000

If you want the finest PC processor money can buy, look no further than Intel's Core 2 Extreme QX9770. Or at least that would be true if you could actually buy it, which is still not possible as we write. Intel has been wowing the press for many months with samples. A full retail launch is long overdue.

Pricey but worth it?

Assuming it is available by the time you read these words, can it possibly be worth over £800? Thanks to a higher 3,2GHz core clockspeed and a uniquely quick 1,600MHz bus frequency, it's clearly the fastest stock-clocked processor on the planet.

The boosted core voltage also makes it easier to hit really high overclocked numbers. But overall, the end user experience is essentially the same as the Q9550 - a model that costs one third the price. Nuff said.