It seems to me...
AMD have spent more time on acronym development than doing anything uniquely useful. Perhaps it's a sign of desperation for the one-time Intel challenger. They seem to be playing catch-up with system-on-a-chip.
AMD has unveiled the mobile version of its "Kaveri" desktop processor, topping off its series of processors in which CPUs and GPUs not only reside on the same die, but also work together in shared-memory harmony – or heterogeneity, to be more precise. AMD Kaveri for Mobile: overview The top end of AMD's new mobile Kaveri …
AMD have spent more time on acronym development than doing anything uniquely useful. Perhaps it's a sign of desperation for the one-time Intel challenger. They seem to be playing catch-up with system-on-a-chip.
Dude, you are so full of it, it is not even funny.
I do my work on an FX nowdays - it has the same (or higher) performance as a Xeon workstation at 20% of the price. My standard benchmark is VM boot (I work on virtualization and virtual networking). A top of the line Datacenter Xeon boots one of my "pets" without load in 32-42s. The FX boots it in 8s. That is 4+ times difference. Under load the Xeon crawls at 128-192s. When I load the FX, it slows down to ~ 24s.
I have retired all intel notebooks too - none of them even compares to any of the recent A4 and A8 based systems. I can now work on the road (finally) as even a lowly 270£ HP Touchsmart can do my virt stuff and re-build it too.
This is now (before hUMA). With hUMA they can run circles around Intel any time as it now becomes realistic to use GPU for processing which has been traditionally on the CPU.
AMD is not just challenging Intel, it is actually beating it silly (at least for what I do). The reason why Intel looks better is that the idiots in PC manufacturers continue to artificially position Intel as premium - give it an SSD and more memory. At the same time AMD systems continue to ship by default with stupid amounts of RAM (2G) and a hideously slow "thin" drives. The first thing I do after buying one is to rip the covers off, stick 16G RAM in (AMD APU based stuff is not Intel, all vendor spec "limits" are fake, you can ignore them) as well as the same drive that goes into Intel systems. Once you do it you can actually get a real comparison and based on that comparison Intel royally stinks.
Have to agree actually. I've recently been quoted for an low grade intel server with a dual core processor.
Thinking that this was a bit puny in the age of virtualised workloads and quad core processors I googled AMD's server offerings and discovered their flagship 16 core chip processor. I got a quote for it, and this astronomically more powerful machine is only marginally more expensive. Intel's nearest competition to this is eye wateringly expensive.
No prizes for guessing which i'm going to buy...
Well, Mr Hand, your own empirical server-based evidence aside, I'm an ex-AMD fanboi who regularly chucked Intel's finest in the bin and replaced them with shiny new K5 & K6 parts. It was AMD or nothing. Simple as. But when the first Core pieces rolled out from Intel, around eight or so years back I guess, I began to notice disappointing desktop performance against the various comparable Athlons of the day. I put this down to bad chipset selection, as in cheapie laptop designs (Yes Packard-Bell, hang your head).
Over time my observations and suspicions crystallised into a definite recognition that these new Core thingies were blinding. In the many years since crappy Prescott bit the dust, I have not had a single AMD-powered desktop or laptop that can match Intel's newer chippery, in comparable situations (by price point or CPU speed). And please, remember I came to this position from one of disbelief that my favoured champion had seemingly slipped so far.
This still holds true today; when asked to come on-site to assist with sluggish desktops or laptops, I almost groan aloud when I see the now-dreaded AMD sticker on the front. I know before starting that I can get the performance so good and no better. I have no idea why this is. It is simply the experience of many hundreds of different machines over about an eight-year period.
I am, in truth, saddened by this. I like the underdog to succeed. But I do hope I'm not called out as also being 'full of it'!
Funnily enough in the rare cases when I am called as a "witch doctor" to tend to the sick and wounded PCs somewhere I dread the AMD sign as much as you do. I know what I am going to find:
1. Half of the memory spec required for the OS
2. The slowest disk possible money can buy (I did not believe people still do SATA with 8MB cache until I pulled one out of an AMD machine 2 months back).
3. Add to that a crap realtek or ralink network and you get a complete picture.
As I said - there is nothing wrong with the CPU. In fact the current crop is superior to Intel at all levels (that was not the case with the early E series Fusion vs Core). It is the cretins in PC vendor marketing that continue to insist on downmarketing and crippling the kit on purpose which are the problem. I am typing this on an AMD notebook by the way which is spec-ed by HP as 4G RAM max, sold with 2G, Windows 7 and Seagate Thin (should be called Seagate Slow). It presently has 16G and a hybrid drive and it goes like the clappers (as it should).
This downvote business is annoying. Just cos someone says something you don't like it isn't necessary to slam it. I like AMD and will likely buy theirs next time, and dislike intel but the benchmarks I've looked at when speccing a server suggest AMD, per core, is significantly slower. Plenty cheaper per core but slower single threaded.
A workstation/server builder, Armari, told me they no longer use AMD chips because of the performance (which is why I'm not buying from them next time). Leaving aside Voland's right hand's post, which I don't doubt the sincerity of but which I can't explain, AMD are just slower per core by a fair whack.
@Voland - you need to dig a bit dude, that level of discrepancy can't be right.
@ Mr Hand,
Well, tbh I did say 'I don't know why' - I simply pointed out that - for whatever reason - modern AMD consumer PCs perform atrociously. I'm gladdened to hear your assessment that you believe they are hamstrung from the outset - this was a partial conclusion of mine too, as it seems AMD is seen as the 'budget' choice - so let's bung loads of shite hardware in there.
As I say, I am an AMD fan who is sad that AMD-centred systems are outperformed so readily. I'm perfectly willing to believe they are being unfairly knobbled.
PS - interesting to get so many DVs. I kind of expected them, but surprising anyhow.
There is disparity in the industry.
Look on Ebuyer and the most expensive AMD A8 laptop is a £620 HP with 4GB of ram and a 500GB 5400rpm HDD.
Thats it! Some with 8GB but nothing special to tempt you. All with mundane HDD tech.
As for Intel? £2200 and every price and option downwards.
Pat, I'd like to buy some vowels...erm, no, I'll solve...
Hewlett-Packard Bell...... (righto...)
Could it be that some marketing, um, funds are being used to discourage competitive configurations on AMD-based PCs? or could it be that AMD buyers/users are (falsely) viewed as "bargain basement buyers" when in fact they're simply looking to get the most for their budgets, which could be substantially more that the marketing geeks think?
I think there's something foul, er, expensive in the entire Intel foodchain, to wit: I have a Q-9450-based system with a Gigabyte mainboard which has shuffled off it's mortal coil (or capacitors, resistors, chips, whatever)... OK, 775, nothing "new" out there to replace the mainboard, ebay's USED equivalents are all in the 80-euro and above range...oh wait, I can get a NEW mainboard, faster CPU, etc., for between 80-90 euros... provided it's AMD....Intel, erm, solutions are rather north of 100 euros just for the mainboard...sigh...
Show me the benchmarks demonstrating the FX processors destroying Intel's chips.
Believe me, I'd quite like to run an FX processor, especially as it supports ECC RAM. I've looked at a lot of benchmarks and AMD almost always loses. It doesn't really have 8 decent cores and is embarrassingly slower at compilation compared to the i7 whilst running hotter.
That's not to mention that the support chipsets are substandard too. Very few motherboards support PCI-e 3.0 now and as far as I can make out, all the exciting virtualization progress is being made by Intel (don't argue 'PCI-e 3.0 isn't needed', systems should last for years and future proofing is important as is bandwidth on <= 4x physical cards)
AMD hasn't been competitive for years, except at the low end, where their built in GPU is much better than the Intel offering.
"I know what I am going to find:"
People equate "AMD" with "cheap" and spec accordingly.
I have some servers with 4 16-core AMDs sitting in them humming along nicely. They have 6 "slow" 2Tb drives in them (local scratch disk) because that's what will fit but that disk is only used for intermediate product files (the real storage is on the networks) - 1TB ram ensures they don't go near swap and yes, each core is a little slower than Inter but the aggregate is a lot faster for the money paid.
WRT the notebook: how did you manage to find out how much memory it could actually address, or did you just wing it?
"Believe me, I'd quite like to run an FX processor, especially as it supports ECC RAM"
Equivalent Xeons to Intel consumer CPUs support ECC and are usually within 2-3% of the same price (often cheaper).
That said: at every comparable performance point, AMD are significantly cheaper. Yes, intel go faster but speed isn't everything, especially when the prime throttle is memory speed and latency, not internal multipliers.
Can't wait to see how this performs on a laptop.
>Can't wait to see how this performs on a laptop.
The benchmarks I've seen suggest the 7600 Kaveri is comparable to an i5, with better graphics, for less money.
But it's only comparable. It's not awesomely better, and it's slower for some applications.
We all know how well the Marxian bit worked for economies
">>We all know how well the Marxian bit worked for economies"
Do we? Are you one of those people who think the USSR was a good example of Marxism? Or Cuba? I'm actually not very socialist at all, I'm quite Right Wing and definitely a capitalist. But I dislike off-hand weak arguments.
For many problems, getting things to work efficiently on a GPU is limited by the small working memory of the stream processors, and the costs of transferring data to and from this working memory. Besides, certain tasks have parts best performed on the CPU, and other parts better suited to the GPU, be removing the need to pump data from one memory partition to another, speeds could increase dramatically.
Isnt it funny when all you've got is a graphics core everything looks like a graphics problem? (ok a bit harsh but you get my point)
Sounds like AMD's boffins have hit the spot on their design briefs - I just question how much of that design brief covers real world usage.
Will my outlook searches, excel and word docs run faster on AMD or Intel? All this visual stuff is a red herring for 90% of use cases. The only grudging point I will give them is more video optimisation in this realm of skype calls is always good, although given the low-ish res of most non-dedicated VC feeds I suspect the peak of tuning for them was reached years ago.
Will they do "instant on" as well as haswell? Do I have week long standby power?
Dear AMD less TLA and FLA nerdgasms more real world use cases please.
Oooo lots of downvotes but no-ones arguing with the basic point.
Face it Peak desktop/laptop was reached years ago, the average home and enterprise use cases have barely changed bar a bit more VC, and the only thing driving the need for more performance is the latest cruft filled blob from Microsoft be it office or windows.
So typical El Reg users aside - since most of us fall into some sort of power user cagegory - which gives me more utility? A laptop that sips power with all day battery life for workshops/meetings/commute and that starts instantly or one that can run a massively parallel data visualisation that I will never do.
Note Im not saying that AMD's arent as good as Intel in these use cases - frankly I havent a clue - but there is no point in AMD Marketing pointing out USP's for which there is a only tiny market.
>>"Sounds like AMD's boffins have hit the spot on their design briefs - I just question how much of that design brief covers real world usage."
I think it does cover most of the real world usage. It unfortunately doesn't cover mine as I need raw power and not of the graphics kind (think webserver and database usage), so I'm probably going to jump to Intel for the first time in a very long time later this year. But I think they definitely have the right approach for the mass market. Most real world usage has long since had its CPU needs met. Few are the people that are limited in their Excel usage because of a weak CPU these days. Poor disk speed yes, low memory, occasionally. CPU? Not often.
AMD have the right approach for most people - CPU usage they give a moderate bump, GPU and graphics they give a big bump. Also more specialist areas. I have a FX-8350. It's a recent upgrade from an 1100T. In terms of raw CPU power it's a bit better but not huge. However it has built-in AES support. That makes an encrypted SSD go from something like 50-60% performance hit on what it would be without encryption, to about a 5% performance hit (based on my tests using Bitlocker). See - raw power increase only modest, but real world usage increase huge. You'll see similar with things in these chips - the fast wireless speeds because of in-built support, reallly good and low-impact video decoding because of the UVD and the refinements to that. Faster memory access...
I'm not their target market as someone who is very CPU hungry. But I also recognize that I am not the majority either. The only two mass markets where there's a desperate need for more power are games and virtualization. And what do we see with AMD's new chips? Built in and better GPU architecture raising the gaming baseline significantly and for an order of magnitude less cost. And high core numbers with shared FPU and lots of support for virtualization.
>>"Will they do "instant on" as well as haswell? Do I have week long standby power?"
For the former, that's far more a factor of the rest of your system - i.e. SSD and OS (Windows 8 has its hybrid hibernate solution). For the latter, I don't know but I can't imagine Intel vs. AMD makes much difference compared to other factors. I invite numbers on that.
>>Dear AMD less TLA and FLA nerdgasms more real world use cases please.
You have it precisely the wrong way around. AMD's new chips are the way they are because they're focusing on real world use cases. A tight focus on higher FP calculations, increased clock speed and similar metrics would be the direction that ignored majority real world usage.
Gordon, I suspect the people who downvoted your post probably started wondering which planet you were on when you picked search as a workload for a CPU and hit the down vote button because they couldn't be bothered to read any further or argue with you.
Search performance is generally going to be a result of prior indexing or HDD/SSD performance if searching the disk and CPU usage is not massively relevant in either case, you know?
they could have just said: "every time we revert one of those 'improvements' we made after Phenon II, it goes faster"
Who couldn't see that coming. So many years wasted trying and failing to make less transistors do more. Hard to see how they can get back in the performance PC game and I won't be surprised if their ARM efforts also kill the low end x64 APU line before then. Sad to think I'm probably using my last every AMD desktop CPU.
Hopefully they live up to the hype and aren't ridiculously expensive. I've always been a fan of AMD. Not just because they're the underdog, or a cheaper alternative to Intel, but because they seem to be a bit less shady as a company in their practices. And it's subjective, but AMD processors seem to "feel" faster when multitasking. I hope they got the embedded software right, as this is one way I could see that this could fail. (Think of the famous Pentium bug)
I'm sure they will blow Intel "Integrated HD graphics" out of the water, but re. Intel graphics (and Mythbusters results aside), you can't polish a turd. I'd love to see a comparo with real-world equipment with dedicated graphics cards.
TDP is important. It determines the type and size of heatsink the system needs. It's especially important if you are trying to build a passively cooled system (no fan, no moving parts (SSD), no noise).
It's nice to know your system can turbo for a few seconds, relying on thermal inertia to avoid meltdown, and then down-clocking itself when not busy or in thermal distress to allow the heat to dissipate. But TDP, as the maximum long-term-average amount of power that a busy system will ever need to dissipate, is a critical parameter for designing it.
"processors in which CPUs and GPUs not only reside on the same die, but also work together in shared-memory harmony "
Sounds like what the PS4 is doing. Everything on-die sharing the same pool of 8GB of GDDR5..
*COUGH*Processor on the PS4 is AMD*COUGH*
The word 'kaveri' means 'a friend' or 'a pal' in Finnish. The story makes a disorienting and sometimes questionable read for a Finn:
"Macri, Keller, et al. worked on the back end of Kaveri as well"
But nice to see other words than just 'sauna' to become more widely used. As long as you stay of the word 'sisu' (reserved stricly for Finns only), it's okay.
The important thing is that you maintained your sense of hUMA.
Would the singular of hUMA in that case be hUMUS? (sorry, problem with a fertile imagination.....)..
But the, erm, fertilizer that often comes out of marketing sounds like the moveis:
"Most importantly," he emphasized, "it takes into account what you're doing, when you're doing it, and reacts to it in an very unique way – because only you are going to do what you do the way you do it when you do it how you do it. And we will react to that dynamically. We won't react to it statically."
kind of like, "I'm just a guy, who's a guy, being a guy"... (slimeball in Larry Crowne) ...
"One of the key things in life is to learn how to do 'nothing' well. Whether it's real life or computers, it's very important – or you'll burn yourself out for no reason."
oh, this is rich.... I imagine in marketing, they really do NOTHING very well, indeed (sorry, don't mean to disparage the sincere marketers out there...).
I hope the engineers win more of the battles going from here on out...it's been a bit of a drought since Dirk Meyer left, but I think the new folks on board should help...
Kaveri is extremely well suited for portable devices as it's quite capable, low power and properly integrated.
I recall that AMD has been claiming significant IPC increases for the last couple generations and they tend to not pan out as expected in benchmarks. Look at the AnandTech review of the A8/A10 from January, the fastest AMD part is barely competitive with the slowest Intel part on CPU heavy benchmarks. On GPU it's faster. So if you're playing video games or doing OpenCL-accelerated stuff and aren't going to buy a dedicated video card for it (which seems like a rare corner case, probably exclusively laptops) then go with AMD, but otherwise they're hard to recommend.
And this is from the person who convinced a department of a Fortune 500 company to switch to Athlon 64s because they were faster and all Intel was delivering was nonsense overpriced Pentium 4s and Itaniums.
IPC is a brick wall. The increases won't pan out. The rest of the article makes AMD's latest offering sound quite interesting, but the remarks about IPC suggest that their CTO is a bit of a noob.
For the hardware, the doubling period for IPC is measured in decades. This "big increase" was only 20% and we know that you get performance benefits like that if you double the size of the cache, something that process improvements continue to deliver every 2-3 years despite the naysayers telling us that Moore's Law has ended.
For the software, we know that Intel bet the farm on EPIC and after 15 years of flogging the horse their compiler gurus eventually conceded that they *still* don't know how to get an IPC of more than 2 or 3 out of any conventional language, and probably never will. (Academics could have told them that back in the early 90s, of course, because the problem had been studied to death in the 70s and 80s in the context of data-flow languages and auto-parallelisation.)
In summary, even if you could boost your hardware's IPC by a factor of two or more, which you can't, you wouldn't ever have any software that could exploit it.
"For the hardware, the doubling period for IPC is measured in decades. This "big increase" was only 20% and we know that you get performance benefits like that if you double the size of the cache"
Agree that we will likely never see any HUGE improvement from IPC but it's still possible for AMD to do better, we know that because Intel does better. Intel has done some interesting stuff with branch prediction (which has a knock-on effect on everything) and stuff like microop fusion which is pretty cool. I really wish AMD would do better, simply because they deserve to for inventing AMD64.
I'm sure it might be ok at the low end, and honestly like AMD's concept of a combined CPU/APU to speed up tasks - if implemented correctly it could indeed by fast.
Problem is, where is the code? There's at least one bit of custom code that speeds up part of one application but that's about it. To deliver on AMD's promises there needs to be pervasive, tangible speedups baked into operating systems rather than being vaguely comparable with Intel.
There's nothing available for Windows or Linux as far as I can see. AMD does well at cryptography and gaming but lags at almost everything else.
That's not necessarily a bad thing, if this is a reasonably performing low end chipset. Unfortunately AMD is trying to position this as a new architecture. Until they deliver on HSA code it's just marketing fluff.
Typical AMD : it's a good idea on paper, but execution is lacking.
Baking anything into the OS would be essentially worthless. When was the last time you saw your OS take more than 1-2% of your CPU time? Making that faster is not a priority.
Adoption of multithreading to speed things up was slow but we're essentially there now. Using the GPU will become commonplace eventually for tasks that would benefit from it.
The first step though is to have a standard interface to all this GPU computation stuff... OpenCL seems to be a good one but support is spotty. It's very driver dependent, people have a hard time getting it to work, and its still not supported on a lot of platforms (especially mobile). Once this all gets sorted out (which AMD seems to be trying to accelerate) then people will become more used to programming for it.