AMD backs USB 3.0 on desktop and laptop chipsets AMD has become the first major microprocessor baker to have USB 3.0–capable chipsets certified by the USB Implementers Forum. This announcement was made on Tuesday by the USB-IF, although the impending certification of AMD's A75 and A70M Fusion Controller Hub (FCH) chipsets had been outed by SemiAccurate three weeks ago from …
In addition, to the above, eSata and firewire connections for storage devices are also likely to be swept away.
ie1394 may live on for a while in devices like video capture cards and camcorders, though it may eventually be replaced by thunderbolt, which is also headed for a similar niche position to what firewire had all these years. Granted its a lot faster but right now that just makes it faster than what anyone can use...
Firewire has been a wee bit over the hill for a little while.
Esata on the other hand... I'm not so sure that USB3 will just blow it away. Now we have useful stuff like power-over-esata the benefits of USB3 aren't quite so cut and dried. I'll bet eSata gets a speedbump before USB3, too. I guess it will just come down to the cost of the supporting hardware in the end.
FW has protocols neither SATA or USB have, including ones critical to high end camcorder operation and HiFi audio systems. It;s a niche port, most people do not need it, but we can not eliminate it. For these people, including a TB port is just as good as a native FW port since it can be easily used with an adapter, not possible on notebooks that do not already include FW ports natively and/or an ExpressCard slot.
eSATA2 in practice is faster than USB3. USB is not full duplex and that alone is enough, not to mention the CPu latency inherent to USB that SATA does not apply. Also, lacking NCQ, Comitted disk writes, and more, means USB3 devices can still not be used in RAID, can;t host VMs reliably (and in some emulators at all), and will be very poor for heavy random read/write operations like databases. USB3 is good for little more than bulk transfer and/or backup, but if we need eSATA foer everything else, we might as well just USE eSATA (since the drive controller supprots that natively without ADDING a USB host on top, making it both faster and cheaper and simpler architecture). eSATA6, now present in every current generation chip, (it already GOT it;s speed bump, a few years ago, and is now COMMONLY deployed)is more than twice USB3 speed in practice, and 6gbit (duplexed) vs 5 gbit (not duplexed) theoretical.
USB3 coasts more (for storage), is more complex, imparts more CPU load, has higher read/write latency, lacks critical protocols, and in every other case if fully redundant to all other EXISTING technologies. The ONLY thing USb3 brings is a combination of bulk transfer speed with legacy device support in the same port. However, TB brings that and SO much more also without adding any new ports (or cables for that matter, using common DP-mini to DP cables).
Actually USB3 _is_ full duplex, unlike USB 1 and 2. This is also how USB3 gained the ability to do interrupts rather than always having to be polled. The dedicated lanes for transfers in both directions allow devices to issue messages when they want to rather than only when asked to. I believe it can even do DMA, which should reduce CPU needs a lot too.
Not that I think eSata is going away (I sure hope not), but USB3 doesn't actually look too bad.
the drive is already SATA. Why go SATA (ion the bus) to USB host, to USB node, and BAXCk to SATA, wasting all that resource and overhead, and loosing duplexed communication and committed writes in the process, when you could just have used the native SATA connect already present in the drive?
eSATA drives were more expensive when they were IDE inside the caddy, but now that all drives are SATA native, USB is actually more complex and more expensive, and slower and more limited...
Better still, TB carries SATA over PCIe, so an external TB device is seen as an internal hot-swappable disk by default, and another drive or a display can be chained off that without requiring a hub, and at 20gbit (duplexed speed) instead of 5, and at 10w instead of 5w power.
USB3 came WAY too late to catch on. it got replaced before it was ever available.
AMD has had a history of not supporting OSS until fairly recently. Users the world over built their computers with AMD as a base but I'm not sure AMD recognised it for a long time. If they continue to keep supporting us then they have a real chance here at building a competitive business...
Well, I've had no problems with USB 3.0 on my AMD64 desktop. Admittedly I've only got one USB 3.0 device, a disk enclosure with a 5400RPM HDD at present.
hdparm suggests its capable of speeds I would not be doing over USB 2.0, but I should do some more concrete testing. The drive in the enclosure happily did peaks of 30MB/sec (average well over 20MB/sec) when rsyncing data from there to a 750GB HDD that I had installed in an Apple MacBook computer (the HDD in the case is the 160GB Toshiba originally installed in the MacBook).
Being Thunderbolt mostly an external PCIe bus thing, it will have to rely on bridge chips to be usable, so quite possibly the cheapest ones will be TB-eSATA and TB-USB 2.0/3.0.
(Does eSATA support USB-style drive mount-unmount without rebooting the host? If not so, the winner would be USB 3.0 for cheap external drives)
They made it faster, but its still not full duplexed, still lacks core protocols needed for high performance disk use (other than bulk transfer), can't be used in a RAID set, and more. If you already have SATA onboard, and in use cases it;s faster and has a superior protocol set, why not use it and forget USB3, especially when every current chipset ships with SATA6 which is even faster, and when eSATAp carries 5v on laptops and 12v on desktops at 2A (more than double USB3). TB makes it even less relevent.
USB3 is too little too late. It has no single use case not already met by eSATA + USB2. going to USB3 requires new cables, and more complicated devices and more cost. Using SATA for all storage universally makes everything simpler, and there are no other "high speed" peripherals out there already that can be used over USB (even USB3) due to latency and CPU load issues. (they require FW specific protocols, or low PCI latency USB can not meet).
This is mainly due to that market cap stuff... and now some technicals...
AMD chips are better optimized per clock than intel chips. intel has a raw speed advantage due to the die shrink advantage, but still lag in their floating point unit designs. This is why games typically run better on AMD chips than intel chips. that, plus there's no hyperthreading on AMD chips that crates logical CPUs which can overburden a single physical CPU and cause a program to slow down when branch prediction is not optimized for that actual family of processors. AMD chips don't have that overhead to deal with. Hyperthreading CAN provide a 30% increase... BUT it can also cause a halving, or worse, of speed when running older/unoptimized software. There are tradeoffs to each one for sure. If you need raw integer throughput, you'll probably be going with whatever bleeding edge top end that intel has up to offer. if you're wading through scientific data that requires massive floating point maths, you'll probably be going with some combination of the two. If you're a gamer that doesn't have endless cash... AMD is typically the best bang for the buck processor wise.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
