- Joined
- Feb 4, 2009
- Location
- cypress, tx
He is talking big server side. Something 99% of people on here will never deal with.
-
Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!
AMD Zen Will Compete Favorably with Intel
- Thread starter Evilsizer
- Start date
He is talking big server side. Something 99% of people on here will never deal with.
Oh geez... yeah, this isn't really an enterprise site in the least. Though, ironically, that is what I do but in a Mainframe capacity.
- Joined
- Jul 31, 2004
Using my desktop for both gaming and software development, having a whole lot of good AMD cores to emulate a distributed environment with various VMs and containers is much preferable to a few great Intel cores.
Comparing E5-2600 series vs 6300 series (NewEgg has Intel 16-core ($1900) and AMD 16-core ($700-$900)), the performance on the Intel side doesn't seem to be 100% better (http://www.anandtech.com/show/6508/the-new-opteron-6300-finally-tested/14 and rather simplistic http://cpuboss.com/cpus/Intel-Xeon-E5-2697-v2-vs-AMD-Opteron-6328), and that's on Piledriver cores. As long as Zen doesn't cost too much more, the cost benefit is obvious. Also, for anybody doing software development on large projects, L2 cache is a killer feature when compiling anything. Intel has been skimping on L2 for years, and it shows in something like Gentoo.
Then why would you use a Xeon, which is x86?
Comparing E5-2600 series vs 6300 series (NewEgg has Intel 16-core ($1900) and AMD 16-core ($700-$900)), the performance on the Intel side doesn't seem to be 100% better (http://www.anandtech.com/show/6508/the-new-opteron-6300-finally-tested/14 and rather simplistic http://cpuboss.com/cpus/Intel-Xeon-E5-2697-v2-vs-AMD-Opteron-6328), and that's on Piledriver cores. As long as Zen doesn't cost too much more, the cost benefit is obvious. Also, for anybody doing software development on large projects, L2 cache is a killer feature when compiling anything. Intel has been skimping on L2 for years, and it shows in something like Gentoo.
Then why would you use a Xeon, which is x86?
Last edited:
OP
- Joined
- Jun 6, 2002
- Thread Starter
- #25
intels dont scale like amd? maybe its in different programs then what i tested but it doesnt have a big list of amd cpus. Intels X58, Asus RIIG with a Xeon [email protected] in cinebench R11.5 and R15. Even with the and newer gen i7(some quad core with [email protected]) in the list on R15 the x5660 didnt better then it. in the middle of moving the rig so i can get pics atm.
what kind of special, besides what seems to be a bit more binning and other types of VT enabled they are. then other types of xeons being multi cpu enabled, im not sure what else they seem to offer.Xeons are special x86, especially the new Skylake ones
OP
- Joined
- Jun 6, 2002
- Thread Starter
- #27
Ok for some reason the pics i have R15 doesnt have any AMD listed, not sure you can select one from a box some how need to look into it. R11.5 has a amd 12 core listed and the only one for some reason. That current run as you can see is at 3.6ghz with DDR3-1600 ram, now if you see the other X5660 listing. For some reason it didnt show the right cpu speed in there, that is my 3ghz with DDR3-1333 run. even with that lack of real cores for 12 threads with 400mhz more clock speed intel doesnt seem that far behind amd. One question would be which is more power efficient for doing the same work, when factoring intel beind a touch behind with handling 12 threads.
- Joined
- Jul 31, 2004
The companies buying the workstations that include Xeon CPUs and the data centers leasing Xeon boxes aren't going to be overclocking them (or be at all happy with a customer that tries to do it remotely themselves), so the overclocked benchmark score is somewhat meaningless in the wider world. Looks like the 2435 was also significantly cheaper at release time than the X5660 (though without an AMD equivalent of ARK I don't know where to find an official price listing), and a pair can still be had for no more than an X5660 now.
OP
- Joined
- Jun 6, 2002
- Thread Starter
- #29
i know they wont, but just showing what my score was with the X5660 at that speed and the lower one at 3ghz. im sure if intel wanted to back then they could have released faster hex cores no problem. the pic was to show how far things were off when he talked about multi thread ability and AMD being better that intel didnt catch up till skylake. since the X5660 is first gen i's, just different name.
- Joined
- Jun 13, 2003
- Location
- Lebanon, PA
Hard to put much stock in AMD's statement until we see some legitimate release of substantial info.
- Joined
- Dec 18, 2000
Programmers tend to be very lazy and so they will continue to reuse and modify their single-threaded code. AMD's foolishness was in thinking that they could change the coding world simply by manufacturing CPUs with lots of slow cores. Didn't happen, won't happen going forward. Most all games and other PC applications still run off of a single-threaded main module with only ancillary and background tasks assigned to other cores/threads. Even though Zen was being compared to Intel's 2013 Haswell by AMD, that is still encouraging since it will be a big jump in IPC for them. AMD's previous modular architecture with the shared FPU was doomed from the start. It only outperforms Intel in true multi-core programs which represent just a small fraction of the typical user's workload.
- Joined
- Feb 4, 2009
- Location
- cypress, tx
Right so this has been a known thing for a while now. Per core scaling has been better on AMD hence the whole AMD only does well in massively multithreaded environments using a non-biased linux based compiler.
- Joined
- Jul 31, 2004
There are programmers and then there are programmers. Some of us actually take pride in doing things right and taking advantage of available resources without making piles of bloat.
Yeah, because that's how this works. For anything requiring user input, the control thread is always waiting on the user. For anything not requiring user input, there still has to be something controlling operation. Again, due to the nature of OpenGL and DirectX APIs, games have not even had the option of applying SMP/SMT to the graphics engine. With DX12 and Vulkan, they have that option, and if you think nobody will take advantage of that, I think a bunch of programmers will be slightly offended.
1. There is no such thing as "true multi-core program". Either it uses multiple threads or it doesn't.
2. CMT outperformed SMT in lots of things that didn't make use of FP ops. Piledriver competed just fine with the same-era Intel chips for those tasks. See some 7-zip benchmarks for examples, e.g. some guy's stock 8350 only being 12% slower than EarthDog's 4.5GHz (25% moar hurts
) 3770k And I don't think this was because CMT was a bad idea, just the FP performance was weak. Intel, after all, is also sharing an FPU for "HT" threads.
- Joined
- Dec 18, 2000
All this being said, I hope Zen is a rousing success and truly wish more programmers were like you. I built AMD socket 939 and 940 dual socket systems almost exclusively from 2003 to 2006 because they were the best. Since then, I just build one now and again for the fun of it. But, if Zen is what we all hope it will be, I will give one a shot when the come out. We really need some viable competition for Intel to keep prices down and innovation up.There are programmers and then there are programmers. Some of us actually take pride in doing things right and taking advantage of available resources without making piles of bloat.
Yes, I said they tend to be lazy - I didn't say all of them were. Maybe 3 of the 50 programmers I worked with in my last 22 years were like you. The rest couldn't be bothered with actually reading all the requirements, caring about human factors, or even commenting within the code to identify when and why changes had been made.
Yeah, because that's how this works. For anything requiring user input, the control thread is always waiting on the user. For anything not requiring user input, there still has to be something controlling operation. Again, due to the nature of OpenGL and DirectX APIs, games have not even had the option of applying SMP/SMT to the graphics engine. With DX12 and Vulkan, they have that option, and if you think nobody will take advantage of that, I think a bunch of programmers will be slightly offended.
Well multi-core CPUs have been the standard for quite a few years. How long do we have to wait before this happens?
1. There is no such thing as "true multi-core program". Either it uses multiple threads or it doesn't.
You just said above that games can't function very well with multiple cores because their inherent design requirements and that the OpenGL and DirectX APIs prevent implementing SMP/SMT. That's pretty much what I was talking about.
2. CMT outperformed SMT in lots of things that didn't make use of FP ops. Piledriver competed just fine with the same-era Intel chips for those tasks. See some 7-zip benchmarks for examples, e.g. some guy's stock 8350 only being 12% slower than EarthDog's 4.5GHz (25% moar hurts
It's always a bad idea if you design a CPU architecture that you know doesn't work well with current software and then somehow expect everything to be modified to run well on it. Especially when you only have a small share of the market. AMD tried to pretend that an FX-8350 is a true 8-core processor, when it really wasn't because of the shared FPUs. Intel never claimed an i7 quad core was an 8 core, just that it had 4 cores and 8 threads.
- Joined
- Jul 7, 2013
What I'm wondering is if AMD's Octa-core Zen CPU will have a high enough price difference compared to Intel to draw people in. A 5960X is over $1000 right now. IF, note the IF there, Zen is AMD's exact equivalent of Haswell or at least it's comparable architecture for Haswell, then their octa-core would be their 5960X. I'm wondering just how much cheaper they would price it? And the hex-core? I really think they'd either have to be desperate and willing to throw money away, or they're not going to come close to really being comparable to Haswell which is obviously the most likely scenario.
Alaric
New Member
- Joined
- Dec 4, 2011
- Location
- Satan's Colon, US
If AMD thinks they have a perception problem with the buying public they may price accordingly (if they can and still make a profit). I know people who, when asked what kind of computer they have, answer "It's an Intel" because that's what the sticker on their Dell/HP/etc. says. I would think a good indicator would be how much ground they can make up with the big companies. Big numbers with Dell and the rest could give them enough volume to do some damage to Intel if AMD can stay close to their current pricing model.
- Joined
- Jul 31, 2004
It's also 140W, and given the promises so far, it's looking like an 8-core Zen will be much more power efficient. Hoping for another Athlon XP here to force Intel to drop the stupid $1000+ prices.
Last edited:
- Joined
- Jul 7, 2013
If AMD manages to make it more power efficient to the point where they can make it massively cheaper than the 5960X, less than $600 let's say, I don't see how Intel will manage to survive something like that. AMD just might regain its former glory. If Intel price matches Zen, or at least puts their prices in Zen range, it'd come down to who has better IPC PPC etc, and why do I get the feeling AMD might still push through on this end too?
- Joined
- Apr 19, 2012
See this part here is great. It's not only just an 8 core processor. It's a 16 thread capable processor. Your telling me I don't need a server board? Sweeet.
- Joined
- Jul 7, 2013
See this part here is great. It's not only just an 8 core processor. It's a 16 thread capable processor. Your telling me I don't need a server board? Sweeet.
THIS. Even if AMD doesn't get exact on a performance copy of the 5960X, their 8 core 16 thread CPU WON'T NEED A SPECIAL SOCKET INTENDED FOR SERVERS. Which would also contribute to AMD winning as the better value!
Similar threads
