Split from: Question on current-gen Intel vs. AMD processors

Discussion of all forms of processors, from AMD to Intel to VIA.

Moderators: Flying Fox, morphine

Split from: Question on current-gen Intel vs. AMD processors

Postposted on Tue Mar 18, 2014 9:38 am

Duct Tape Dude wrote:It's also important to note that aside from an arguably superior CPU architecture, Intel is making CPUs at a smaller process (22nm instead of 32nm or 28nm). In general, Intel can fit more transistors in a given space and also run them more efficiently (which means they can go faster before running into thermal limits).


Yep. On architechture side Intel is far from being superior. Intel's manufacturing process however is clearly superior.

Kougar wrote:For floating point workloads, it's because AMD halved FP-parts of its "cores" that handle FP workloads. To quote Anandtech: "As a comparison point, one core in Haswell has the same floating point performance per cycle as two modules (or four cores) in Steamroller."


That's just good thing because floating point units on processors are pretty much useless. Running legacy software is an exception of course.
Wicked Mystic
Gerbil
 
Posts: 51
Joined: Thu Oct 03, 2013 6:36 am

Re: Question on current-gen Intel vs. AMD processors

Postposted on Tue Mar 18, 2014 9:54 am

Wicked Mystic wrote:That's just good thing because floating point units on processors are pretty much useless. Running legacy software is an exception of course.

Wait, what?! :o
There is a fixed amount of intelligence on the planet, and the population keeps growing :(
morphine
Gerbil Khan
Silver subscriber
 
 
Posts: 9956
Joined: Fri Dec 27, 2002 8:51 pm
Location: Portugal (that's next to Spain)

Re: Question on current-gen Intel vs. AMD processors

Postposted on Tue Mar 18, 2014 10:11 am

morphine wrote:
Wicked Mystic wrote:That's just good thing because floating point units on processors are pretty much useless. Running legacy software is an exception of course.

Wait, what?! :o


It makes sense to use "half FPU units". Much less transistors and heat. Even low end GPU can do most jobs much faster. Processor FPU's have much less usage than 20 years ago when processors got integrated FPU's.

Intel thought same with Pentium 4 long time ago. Very weak FPU on legacy applications and somewhat competitive (but not groundbreaking) with SSE2.
Wicked Mystic
Gerbil
 
Posts: 51
Joined: Thu Oct 03, 2013 6:36 am

Re: Question on current-gen Intel vs. AMD processors

Postposted on Tue Mar 18, 2014 10:18 am

So essentially you're saying two things:

1) Modern CPUs don't really make usage of their FPUs. I'm sorry, but this is just... plain wrong. All you have to do is read one of TR's CPU benchmarks to see for yourself. Let alone we start discussing CPUs meant to be used in servers.
2) Suggesting just using the GPU doesn't really work either. Getting the work to do onto the GPU and back takes an overhead and, most importantly, specialized programming.
There is a fixed amount of intelligence on the planet, and the population keeps growing :(
morphine
Gerbil Khan
Silver subscriber
 
 
Posts: 9956
Joined: Fri Dec 27, 2002 8:51 pm
Location: Portugal (that's next to Spain)

Re: Question on current-gen Intel vs. AMD processors

Postposted on Tue Mar 18, 2014 10:32 am

morphine wrote:So essentially you're saying two things:

1) Modern CPUs don't really make usage of their FPUs. I'm sorry, but this is just... plain wrong. All you have to do is read one of TR's CPU benchmarks to see for yourself. Let alone we start discussing CPUs meant to be used in servers.
2) Suggesting just using the GPU doesn't really work either. Getting the work to do onto the GPU and back takes an overhead and, most importantly, specialized programming.


1. I said that legacy software is an exception. And those CPU benchmarks are very much legacy software. If GPU is properly used for most floating point calculations, why put powerful FPU on CPU?

2. Overhead, yes. Something like HSA is needed to reduce that. Specialized programming, yes. GPU's have been programmable for many years not but software side develop slowly.

My point is same that Intel thought 20 years ago. Putting powerful FPU on CPU is very inefficient. It takes much transistors and power consumption gets up. Only problem is that modern software is quite poor. It's 2014 and most software is still single threaded.
Wicked Mystic
Gerbil
 
Posts: 51
Joined: Thu Oct 03, 2013 6:36 am

Re: Question on current-gen Intel vs. AMD processors

Postposted on Tue Mar 18, 2014 10:41 am

Wicked Mystic wrote:And those CPU benchmarks are very much legacy software.

So, essentially, all modern software is "legacy"? Because those benchmarks are very much up to date. Never mind art software, 3D modeling software, scientific software, server software, essentially anything that relies on FPU units. See here and here for very much modern practical examples.

Wicked Mystic wrote:If GPU is properly used for most floating point calculations, why put powerful FPU on CPU?

Because it's not that simple. In order to use the GPU for something, you need to program specific code in your application to actually do that, and it has to be worth it from a performance standpoint. It's not a trivial task.

Wicked Mystic wrote:Only problem is that modern software is quite poor. It's 2014 and most software is still single threaded.

No offense, but you don't seem to completely understand how software is built, especially at a low level. Multi-threading and FPU usage have nothing to do with each other, and "just using the GPU" isn't as simple as it sounds for a large portion of practical software. Remember, just because something can be done in the GPU, it doesn't mean like it's a good idea to do so.
There is a fixed amount of intelligence on the planet, and the population keeps growing :(
morphine
Gerbil Khan
Silver subscriber
 
 
Posts: 9956
Joined: Fri Dec 27, 2002 8:51 pm
Location: Portugal (that's next to Spain)

Re: Question on current-gen Intel vs. AMD processors

Postposted on Tue Mar 18, 2014 11:04 am

morphine wrote:So, essentially, all modern software is "legacy"?


Yes. How many 64-bit games are around? Why many common applications are still 32-bit and/or single threaded?

Because almost everything is legacy. With modern software (not actual but what it should be) processors FPU's would have much less use.

morphine wrote:Because those benchmarks are very much up to date. Never mind art software, 3D modeling software, scientific software, server software, essentially anything that relies on FPU units. See here and here for very much modern practical examples.


Many of those those benchmarks are outdated. AES encryption runs fine with GPU. They did use CPU, right? There is never much use for synthetic tests. Very outdated example is Cinebench 11.5, that is over four years old. Latest Cinebench version seems to be 15.

morphine wrote:Because it's not that simple. In order to use the GPU for something, you need to program specific code in your application to actually do that, and it has to be worth it from a performance standpoint. It's not a trivial task.


It's also easier to make 32-bit games and use directx 9 instead of 10 or 11.

morphine wrote:No offense, but you don't seem to completely understand how software is built, especially at a low level. Multi-threading and FPU usage have nothing to do with each other, and "just using the GPU" isn't as simple as it sounds for a large portion of practical software. Remember, just because something can be done in the GPU, it doesn't mean like it's a good idea to do so.


Multithreading and FPU is different thing of course. However if multithreading is too hard to make, then it's no wonder that using GPU is also hard.

AMD tries to solve this problem with HSA. GPU offload is at least partially made on hardware side. If it really works well, it will lower FPU usage considerably. And then FPU is very much useless.
Wicked Mystic
Gerbil
 
Posts: 51
Joined: Thu Oct 03, 2013 6:36 am

Re: Question on current-gen Intel vs. AMD processors

Postposted on Tue Mar 18, 2014 11:12 am

Welcome back, Shining Arcanine.
There is a fixed amount of intelligence on the planet, and the population keeps growing :(
morphine
Gerbil Khan
Silver subscriber
 
 
Posts: 9956
Joined: Fri Dec 27, 2002 8:51 pm
Location: Portugal (that's next to Spain)

Re: Question on current-gen Intel vs. AMD processors

Postposted on Tue Mar 18, 2014 11:16 am

morphine wrote:Welcome back, Shining Arcanine.


Thanks. Is that some kind of special title or something?
Wicked Mystic
Gerbil
 
Posts: 51
Joined: Thu Oct 03, 2013 6:36 am

Re: Question on current-gen Intel vs. AMD processors

Postposted on Tue Mar 18, 2014 11:22 am

Wicked Mystic wrote:
morphine wrote:Welcome back, Shining Arcanine.


Thanks. Is that some kind of special title or something?

No, just someone that had the same exact ideas like you.

And who also didn't quite seem to know what he was talking about.
There is a fixed amount of intelligence on the planet, and the population keeps growing :(
morphine
Gerbil Khan
Silver subscriber
 
 
Posts: 9956
Joined: Fri Dec 27, 2002 8:51 pm
Location: Portugal (that's next to Spain)

Re: Question on current-gen Intel vs. AMD processors

Postposted on Tue Mar 18, 2014 11:32 am

morphine wrote:No, just someone that had the same exact ideas like you.

And who also didn't quite seem to know what he was talking about.


Right. I think I'll check his posts. He must have written something great.
Wicked Mystic
Gerbil
 
Posts: 51
Joined: Thu Oct 03, 2013 6:36 am

Re: Question on current-gen Intel vs. AMD processors

Postposted on Tue Mar 18, 2014 11:35 am

Hi Zaeem,

First of all, my apologies about Wicked Mystic. He receives (rather small) checks from a certain company to propagate a certain point of view and as you can see, his ignorance is almost as large as his bias.

As for Intel vs. AMD, most of the issues really boil down to nomenclature. AMD has adopted a naming scheme for "cores" that is technically correct... each core shows up independently to the OS and can (sort of) independently execute code, but in order to save money AMD made some big sacrifices to each "core" so that what AMD calls a "core" in a modern chip is not the same thing as what Intel calls a "core".

As others have pointed out, the two major areas where AMD's cores don't quite live up to the hype is in the area of instruction decoding/issuing and in the floating point units. Each of these units is shared to one degree or another between two cores in what AMD calls a "module". Each module is roughly on par with what Intel calls a "core" and in very heavily parallelized integer-only operations each AMD module is roughly equivalent to an Intel core that also enables hyperthreading. Usually AMD's cores are also going to be clocked at higher frequencies (200 - 400 MHz) to achieve the parity. So if you think of an 8-core Piledriver chip in terms of 4 modules, you start to see how it has parity with quad-core Intel chips under the right workloads, especially highly parallel integer workloads. Similarly with a chip like Kaveri, think of the "4 core" model as being roughly equivalent to a lower-clocked dual core Intel part with hyperthreading. TR's benchmarks bear out this assessment. That's just a rough comparison of performance, but it does not take into account other factors like power consumption.

One other thing to remember is that outside of buying an LGA-2011 platform chip from Intel, an Intel part is basically an overclocked notebook chip. Intel pushes lithographic processes very aggressively first to give itself a manufacturing cost advantage over competitors and second to churn out chips that use less power. The results we have seen are that Intel chips aren't getting massively faster on the desktop, although the power efficiency has gone up substantially over the last 5 years and mobile parts have become a very large part of Intel's portfolio. That's probably why you see less enthusiasm for newer Intel parts on sites like TR since the newer features aren't as valuable to some folks who want to overclock.

Unfortunately for AMD, they have stagnated even worse than Intel during the same time period since Bulldozer was in many ways slower than its predecessor and the CPU portion of Kaveri has shown itself to be only about as fast as last year's model because AMD had to lower the clockspeeds to meet the power envelope. Hopefully AMD will eventually come out with either a brand-new architecture or at least resume making 4-module chips in the future. From their roadmaps, however, this may not occur until next year or even 2016. Time will tell.
4770K @ 4.7 GHz; 32GB DDR3-2133; GTX-770; 512GB 840 Pro (2x); Fractal Define XL-R2; NZXT Kraken-X60
--Many thanks to the TR Forum for advice in getting it built.
chuckula
Gerbil Elite
Gold subscriber
 
 
Posts: 560
Joined: Wed Jan 23, 2008 9:18 pm
Location: Probably where I don't belong.

Re: Question on current-gen Intel vs. AMD processors

Postposted on Tue Mar 18, 2014 11:52 am

chuckula wrote:He receives (rather small) checks from a certain company to propagate a certain point of view and as you can see, his ignorance is almost as large as his bias.


Really? That would be appreciated. However, I haven't seen any.

As for bias, I mostly ignore benchmarks that seem to be only truth for some people.

chuckula wrote:Unfortunately for AMD, they have stagnated even worse than Intel during the same time period since Bulldozer was in many ways slower than its predecessor and the CPU portion of Kaveri has shown itself to be only about as fast as last year's model because AMD had to lower the clockspeeds to meet the power envelope. Hopefully AMD will eventually come out with either a brand-new architecture or at least resume making 4-module chips in the future. From their roadmaps, however, this may not occur until next year or even 2016. Time will tell.


Kaveri's manufacturing process was optimized for GPU. That partly explains lower clocks. Another thing is power envelope. Much faster GPU draws more power.
Wicked Mystic
Gerbil
 
Posts: 51
Joined: Thu Oct 03, 2013 6:36 am

Re: Question on current-gen Intel vs. AMD processors

Postposted on Tue Mar 18, 2014 12:23 pm

Wicked Mystic wrote:That's just good thing because floating point units on processors are pretty much useless. Running legacy software is an exception of course.

That's only true if you limit the discussion to legacy x87. The FPUs in modern CPUs (both Intel and AMD) are also responsible for handling the SSE and AVX instructions, so FPUs still matter. A LOT.

Wicked Mystic wrote:Yes. How many 64-bit games are around? Why many common applications are still 32-bit and/or single threaded?

Because almost everything is legacy. With modern software (not actual but what it should be) processors FPU's would have much less use.

Clearly, you don't even understand what an FPU is.

Even 32-bit software relies heavily on SSE these days, it is what replaced legacy x87 instructions. If CPUs didn't have FPUs, SSE would be useless because it would be even slower than legacy x87 instructions.

FPU is not a synonym for x87. It is simply the part of the CPU that handles floating point arithmetic, whether those arithmetic instructions are coded as x87, SSE, or AVX.
(this space intentionally left blank)
just brew it!
Administrator
Gold subscriber
 
 
Posts: 37677
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Re: Question on current-gen Intel vs. AMD processors

Postposted on Tue Mar 18, 2014 1:12 pm

Wicked Mystic wrote:Yes. How many 64-bit games are around? Why many common applications are still 32-bit and/or single threaded?


http://store.steampowered.com/hwsurvey/

Why would they drop ~20% of actual gamers, let alone people who use "common applications", just so they can get Wicked Mystic's "Modern Software!" sticker?

I mean, seriously...

I'm gung-ho for 64-bit in everything, but what you are saying is plainly silly.
Glorious
Darth Gerbil
Gold subscriber
 
 
Posts: 7846
Joined: Tue Aug 27, 2002 6:35 pm

Re: Question on current-gen Intel vs. AMD processors

Postposted on Tue Mar 18, 2014 1:22 pm

ronch wrote:Intel's Haswell and Ivy Bridge cores are roughly 50% faster while running at significantly lower clocks than AMD's Piledriver cores (I say Piledriver since we're talking about 8-core models). So, to illustrate, let's say a single Piledriver core found in the FX-8350 which is running at 4.0GHz produces 8 billion results per second, and an Ivy Bridge core found in the Core i7-3770K and running at 3.5GHz can manage 12 billion results per second. Note that this is for illustrative purposes only and there's no way a particular core design will always perform similarly relative to another core design with all types of code mixes. If you use all the cores in each processor, the FX-8350 may actually beat the Core i7 if we assume that the app can really scale well across many cores and not just throw a few light threads at some of the cores or some other factor related to multi-threading is limiting performance (you might want to read up a bit on Amdahl's Law). If the apps you use can well utilize as many cores as there are available, going the FX route may be a sensible choice as long as you don't mind AMD's lower energy efficiency/higher power consumption. To make their FX chips more compelling, AMD prices them very competitively and boards for FX processors generally offer more value for money. If, however, your apps generally use only a few cores then obviously most of the cores in the 8-core AMD chips are just going to sit idle while the apps run on just a few cores which can only produce 8 billion results each (again, this figure is just our example to illustrate my point). In this case, you're better off with a Core i5 or, if you're willing to spend, a Core i7. Inversely, boards for Intel processors also tend to be more expensive than similarly specced boards for AMD FX processors.

Here's the theoretical math:

If you use all available cores,

Intel = 4 x 12 = 48 billion results
AMD = 8 x 8 = 64 billion results

If you use only 4 cores,

Intel = 4 x 12 = 48 billion results
AMD = 4 x 8 = 32 billion results

Let's say your app uses 7 cores,

Intel = 4 x 12 = 48 billion results
AMD = 7 x 8 = 56 billion results

As for the common belief that 8-core FX chips only have 4 FPUs, this is simply wrong. Each dual-core module in the FX contains two 128-bit floating point engines that are accessible to both cores and both can work independently of each other. However, both 128-bit FP engines can be ganged together to work on bigger chunks of data and this feature is available to both integer cores within a dual-core module. AMD calls this FlexFP. Obviously, if both FPUs are being used by a particular integer core, the other integer core within the module in question will have to wait its turn.

I may now be wrong. But is it not also true that Intel segments their CPU architecture rather more than AMD? For example, IIRC, if you want virtual machine support or ECC memory support; you have to buy the more expensive CPU? Meanwhile AMD CPU's support these features even in the low end models.
AMD Phenom II X4 955 BE | MSI 790FX-GD70 | 4x2GB DDR3 | 3 x Intel X25-M G2 | Samsung Series 830 | XFX Radeon HD 7870 | BenQ FP241VW 24" | Grado SR80 | Logitech G710+ | SeaSonic SS-660XP 80+ Pt | Cooler Master Hyper 212+ | WinXP64 | Back-UPS NS 1250
Mr Bill
Gerbil XP
 
Posts: 420
Joined: Mon Jan 21, 2002 7:00 pm
Location: Colorado Western Slope

Re: Question on current-gen Intel vs. AMD processors

Postposted on Tue Mar 18, 2014 1:48 pm

Stop feeding the trolls, it's bad for your blood pressure!

If logic arguments are ignored and responses are illogical, incorrect and indirect, you are being baited by a pro.
<insert large, flashing, epileptic-fit-inducing signature (based on the latest internet-meme) here>
Chrispy_
Gerbil Jedi
Gold subscriber
 
 
Posts: 1843
Joined: Fri Apr 09, 2004 3:49 pm

Re: Question on current-gen Intel vs. AMD processors

Postposted on Tue Mar 18, 2014 1:54 pm

Mr Bill wrote:I may now be wrong. But is it not also true that Intel segments their CPU architecture rather more than AMD? For example, IIRC, if you want virtual machine support or ECC memory support; you have to buy the more expensive CPU? Meanwhile AMD CPU's support these features even in the low end models.

Yes, Intel segments more aggressively. Less so today than in the past, however. The higher-end features (previously reserved for Xeon) have been filtering down to some of the consumer parts.
(this space intentionally left blank)
just brew it!
Administrator
Gold subscriber
 
 
Posts: 37677
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Re: Question on current-gen Intel vs. AMD processors

Postposted on Tue Mar 18, 2014 1:57 pm

Mr Bill wrote:I may now be wrong. But is it not also true that Intel segments their CPU architecture rather more than AMD? For example, IIRC, if you want virtual machine support or ECC memory support; you have to buy the more expensive CPU? Meanwhile AMD CPU's support these features even in the low end models.


Virtual machine support - hardware virtualization - is in modern Intel chips in general (as VT-x), though VT-d (what AMD calls IOMMU support, basically allowing one to lock a piece of hardware for a virtual machine) is restricted to higher-end chips. ECC memory support's (generally?) limited to Xeons on the Intel side. AMD's Opteron and FX chips support ECC, but FM1/FM2/FM2+ chips do not support it, so it looks like AMD's behavior is beginning to parallel Intel's more closely in that regard.
AMD FX-8320 | ASRock 970 Extreme4 | 24GB Corsair Ballistic DDR3-1600 | 2048 MB MSI Geforce GTX 660 | 600W Corsair Builder Series |Xubuntu 14.04 x64 | Cooler Master CM690 | Pioneer Blu-ray RW Drive | 240 gig Mushkin SSD | 500 gig Western Digital SATA Drive
Concupiscence
Gerbil
 
Posts: 86
Joined: Tue Sep 25, 2012 7:58 am
Location: Dallas, TX

Re: Question on current-gen Intel vs. AMD processors

Postposted on Tue Mar 18, 2014 2:07 pm

Concupiscence wrote:AMD's Opteron and FX chips support ECC, but FM1/FM2/FM2+ chips do not support it, so it looks like AMD's behavior is beginning to parallel Intel's more closely in that regard.

Indeed. Until the APUs came out, all of AMD's 64-bit chips with the exception of a very few bottom-end Semprons have supported ECC. This is now changing.
(this space intentionally left blank)
just brew it!
Administrator
Gold subscriber
 
 
Posts: 37677
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Re: Question on current-gen Intel vs. AMD processors

Postposted on Tue Mar 18, 2014 2:22 pm

just brew it! wrote:That's only true if you limit the discussion to legacy x87. The FPUs in modern CPUs (both Intel and AMD) are also responsible for handling the SSE and AVX instructions, so FPUs still matter. A LOT.


Of course it matters because software is legacy. If certain piece of software is made using latest techniques available and nothing legacy? Then I suspect that very small amout of FPU power would be needed. This won't happen because PC is all about legacy and backwards compatibility. PC world just cannot abandon backward compatibility like Apple have done two times already.

just brew it! wrote:Clearly, you don't even understand what an FPU is.

Even 32-bit software relies heavily on SSE these days, it is what replaced legacy x87 instructions. If CPUs didn't have FPUs, SSE would be useless because it would be even slower than legacy x87 instructions.

FPU is not a synonym for x87. It is simply the part of the CPU that handles floating point arithmetic, whether those arithmetic instructions are coded as x87, SSE, or AVX.


Using weak FPU and no FPU at all is totally different thing. Modern processors support even MMX for compatibility reasons. AMD abandoned that and dropped 3D-NOW! support totally.

I didn't say FPU is same as x87. Already wrote about Pentium 4 x87 and SSE2 that tried to make difference.

Also we may consider it this way. Around 1995 Intel engineers suspected that high power FPU would not be needed around year 2000. Now 2014 high power FPU has much use. Were Intel engineers just stupid? I think they just underestimated power of legacy.

Glorious wrote:Why would they drop ~20% of actual gamers, let alone people who use "common applications", just so they can get Wicked Mystic's "Modern Software!" sticker?

I mean, seriously...

I'm gung-ho for 64-bit in everything, but what you are saying is plainly silly.


They would'nt because that's bad for business. However that also means they support legacy. Right?

Microsoft still has 32 bit operating systems. What if Vista would have been 64-bit only? When Windows XP support ends next month, that way we would have only 64 bit Windows systems very soon. Bad for business = support legacy. PC is all about supporting legacy.

Mr Bill wrote:I may now be wrong. But is it not also true that Intel segments their CPU architecture rather more than AMD? For example, IIRC, if you want virtual machine support or ECC memory support; you have to buy the more expensive CPU? Meanwhile AMD CPU's support these features even in the low end models.


Right. AMD usually have much more unlocked multiplier processors, almost all models support virtualization, ECC and so on. Not like Intel where you may choose between virtualization (i5-4670) and unlocked multiplier (i5-4670K).

Chrispy_ wrote:Stop feeding the trolls, it's bad for your blood pressure!

If logic arguments are ignored and responses are illogical, incorrect and indirect, you are being baited by a pro.


Usually when people are wrong (like you are), they say someone is trolling. Also your message adds nothing to discussion.
Wicked Mystic
Gerbil
 
Posts: 51
Joined: Thu Oct 03, 2013 6:36 am

Re: Question on current-gen Intel vs. AMD processors

Postposted on Tue Mar 18, 2014 2:29 pm

Wicked Mystic wrote:
just brew it! wrote:That's only true if you limit the discussion to legacy x87. The FPUs in modern CPUs (both Intel and AMD) are also responsible for handling the SSE and AVX instructions, so FPUs still matter. A LOT.

Of course it matters because software is legacy. If certain piece of software is made using latest techniques available and nothing legacy? Then I suspect that very small amout of FPU power would be needed. This won't happen because PC is all about legacy and backwards compatibility. PC world just cannot abandon backward compatibility like Apple have done two times already.

Wrong again.

AVX *is* the latest technique for running computationally intensive code on x86 processors. AVX uses the FPU.
(this space intentionally left blank)
just brew it!
Administrator
Gold subscriber
 
 
Posts: 37677
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Re: Question on current-gen Intel vs. AMD processors

Postposted on Tue Mar 18, 2014 2:34 pm

just brew it! wrote:
Wicked Mystic wrote:
just brew it! wrote:That's only true if you limit the discussion to legacy x87. The FPUs in modern CPUs (both Intel and AMD) are also responsible for handling the SSE and AVX instructions, so FPUs still matter. A LOT.

Of course it matters because software is legacy. If certain piece of software is made using latest techniques available and nothing legacy? Then I suspect that very small amout of FPU power would be needed. This won't happen because PC is all about legacy and backwards compatibility. PC world just cannot abandon backward compatibility like Apple have done two times already.

Wrong again.

AVX *is* the latest technique for running computationally intensive code on x86 processors. AVX uses the FPU.


At this point we should stop feeding him. Maybe one day if he graduates from high school he'll get run over by a real programmer and learn what the words "floating point" actually mean. In the mean time we don't want to scare him with charts like this one that show what Kaveri can['t] do in double-precision number crunching even with the GPU thrown into the mix.
4770K @ 4.7 GHz; 32GB DDR3-2133; GTX-770; 512GB 840 Pro (2x); Fractal Define XL-R2; NZXT Kraken-X60
--Many thanks to the TR Forum for advice in getting it built.
chuckula
Gerbil Elite
Gold subscriber
 
 
Posts: 560
Joined: Wed Jan 23, 2008 9:18 pm
Location: Probably where I don't belong.

Re: Question on current-gen Intel vs. AMD processors

Postposted on Tue Mar 18, 2014 2:46 pm

just brew it! wrote:
Wicked Mystic wrote:
just brew it! wrote:That's only true if you limit the discussion to legacy x87. The FPUs in modern CPUs (both Intel and AMD) are also responsible for handling the SSE and AVX instructions, so FPUs still matter. A LOT.

Of course it matters because software is legacy. If certain piece of software is made using latest techniques available and nothing legacy? Then I suspect that very small amout of FPU power would be needed. This won't happen because PC is all about legacy and backwards compatibility. PC world just cannot abandon backward compatibility like Apple have done two times already.

Wrong again.

AVX *is* the latest technique for running computationally intensive code on x86 processors. AVX uses the FPU.


Does it still have match for GPU? I think not.

chuckula wrote:At this point we should stop feeding him. Maybe one day if he graduates from high school he'll get run over by a real programmer and learn what the words "floating point" actually mean. In the mean time we don't want to scare him with charts like this one that show what Kaveri can['t] do in double-precision number crunching even with the GPU thrown into the mix.


That is problem because? Many Nvidia GPUs suffer lot when doing double precision vs single precision. Not problem because they are meant for gaming. Kaveri is very much gaming chip also.
Wicked Mystic
Gerbil
 
Posts: 51
Joined: Thu Oct 03, 2013 6:36 am

Re: Question on current-gen Intel vs. AMD processors

Postposted on Tue Mar 18, 2014 2:55 pm

chuckula wrote:he'll get run over by a real programmer and learn what the words "floating point" actually mean

Beyond the three (at least) in this thread? :lol:
There is a fixed amount of intelligence on the planet, and the population keeps growing :(
morphine
Gerbil Khan
Silver subscriber
 
 
Posts: 9956
Joined: Fri Dec 27, 2002 8:51 pm
Location: Portugal (that's next to Spain)

Re: Question on current-gen Intel vs. AMD processors

Postposted on Tue Mar 18, 2014 3:05 pm

morphine wrote:
chuckula wrote:he'll get run over by a real programmer and learn what the words "floating point" actually mean

Beyond the three (at least) in this thread? :lol:


Hehe. They just claim I said something I did not. Point where I claimed that x87=FPU for example? You cannot find that.
Wicked Mystic
Gerbil
 
Posts: 51
Joined: Thu Oct 03, 2013 6:36 am

Re: Question on current-gen Intel vs. AMD processors

Postposted on Tue Mar 18, 2014 3:29 pm

Wicked Mystic wrote:Hehe. They just claim I said something I did not. Point where I claimed that x87=FPU for example? You cannot find that.


Well, what else could a "legacy" FPU even be in x86 land? AVX was added in 2011, AVX2 in 2013. They both have FP and FP-related instructions.

Sooo....

Do you, like, just not know what the word "legacy" means? :wink:

Wicked Mystic wrote: Were Intel engineers just stupid? I think they just underestimated power of legacy.


I guess they don't know what it means either, seeing as how they keep adding new FP instructions... :o :roll: :o
Glorious
Darth Gerbil
Gold subscriber
 
 
Posts: 7846
Joined: Tue Aug 27, 2002 6:35 pm

Re: Question on current-gen Intel vs. AMD processors

Postposted on Tue Mar 18, 2014 3:47 pm

Glorious wrote:
Wicked Mystic wrote:Hehe. They just claim I said something I did not. Point where I claimed that x87=FPU for example? You cannot find that.


Well, what else could a "legacy" FPU even be in x86 land? AVX was added in 2011, AVX2 in 2013. They both have FP and FP-related instructions.

Sooo....

Do you, like, just not know what the word "legacy" means? :wink:


SSE? SSE2? MMX? 3D-NOW!?

Glorious wrote:I guess they don't know what it means either, seeing as how they keep adding new FP instructions... :o :roll: :o


In fact newest Intel processors are very strong on x87 intensive calculations. Can you explain why is that?

just brew it! wrote:Even 32-bit software relies heavily on SSE these days, it is what replaced legacy x87 instructions.
Wicked Mystic
Gerbil
 
Posts: 51
Joined: Thu Oct 03, 2013 6:36 am

Re: Question on current-gen Intel vs. AMD processors

Postposted on Tue Mar 18, 2014 3:59 pm

Wicked Mystic wrote:In fact newest Intel processors are very strong on x87 intensive calculations. Can you explain why is that?

They're only "strong" relative to older implementations. x87 still performs poorly when pitted against anything more modern like SSE. The stack-based x87 architecture imposes some severe constraints on how much optimization the compiler can do, and is incompatible with vectorization.
(this space intentionally left blank)
just brew it!
Administrator
Gold subscriber
 
 
Posts: 37677
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Re: Question on current-gen Intel vs. AMD processors

Postposted on Tue Mar 18, 2014 4:08 pm

Wicked Mystic wrote:SSE? SSE2? MMX? 3D-NOW!?


Again, AVX in 2011, AVX2 in 2013, but yet FPUs in CPUs are "pretty much useless" and only good for "running legacy software."

Let me quote you:

Wicked Mystic wrote:That's just good thing because floating point units on processors are pretty much useless. Running legacy software is an exception of course.


The Haswell New Instructions (which include AVX2) weren't available until summer last year. Thus any software that uses them is, by definition, new and *not* legacy.

What you are saying is insane.

Wicked Mystic wrote:In fact newest Intel processors are very strong on x87 intensive calculations. Can you explain why is that?


Is this supposed to be where you're going to "gotcha" me with how Intel CPUs run lots and lots of "legacy" software? :o

Yeah, OK. I'm going to stop you there.

I'm not really sure what you think the point of a computer is, but for me, and I think everyone else, the point is to get stuff done. If "legacy" software works fine, that's, well, fine.
Glorious
Darth Gerbil
Gold subscriber
 
 
Posts: 7846
Joined: Tue Aug 27, 2002 6:35 pm

Next

Return to Processors

Who is online

Users browsing this forum: Yahoo [Bot] and 2 guests