Reccomendations on future Dual Intel cpu rig for gaming?

C

CleanSlate

Supreme [H]ardness
Joined
Mar 28, 2003
Messages
5,258
I've been thinking that one cpu just doesn't cut it some times and I would like to stick with Intel currently.

I feel the need for more power. *obligatory grunts from tool time*

I currently have an i5 3570k with 2x4g ddr3, asus mobo, gtx 670 2g dcu2 top and I'm not familiar at all with dual/tri/quad cpu server board performance for gaming.

I realize these types of boards (AND cpus) are probably more about web server performance and other server performance areas.

Should I even bother doing some research into this area or is it just a dead end? Should I possibly wait for a good AMD solution instead?


Edit: YAY for hitting the 10 year mark!!!!!!!
 
Don't even bother. Gaming requires higher clocked cores, and current Xeons do not allow any kind of overclocking.

If you want more gaming power, get a higher end GPU, and get 2 or 3 of them.
 
The big question is what games? most games are GPU bound and/or don't take much advantage of more than four cores so building a dual socket rig is kinda pointless for them but iirc there are a handful of games that are CPU bound and highly multithreaded such that more cores can make sense.

Also bear in mind that Intel has prevented any overclocking of their dual socket capable processors this time round and the cheaper end of dual socket capable processors tend to be pretty low clocked. You are going to have to spend some serious money to significantly beat an overclocked i7-3930K even in highly multithreaded tasks.

I could be mistaken but I think we will see hell freezing over before we see AMD returning to providing high end competition.

As for multi GPU dual LGA2011 has insane ammounts of PCIe connectivity so I doubt there will be any problems there provided the board you chose has a reasonable slot configuration and is endorsed by your GPU vendor.
 
They generally don't make any high clocked 4 or 6 core dual socket xeon's so to get an equivalently clocked xeon to a high end i7 you have to get one of their high end 8 core's. The 4 core i7 is going to be around $350, the equivalently clocked 8 core xeon is around $2000, and you'd need 2 of them, and they can't be overclocked.
 
Depends on what you want to do. Not sure it would be better for gaming, but on the server side the E5-2620 is currently a steal.

The E5-2609 is 4 core, no HT, no turbo (2.4ghz). The 2620, for $100 more gets you 6 cores, 12 threads, and 2.0->2.5ghz so on a dual socket you go from 8 threads -> 24 threads for +$200.

IMHO if you want to game and have something better than i7-3770, I'd snag the E5-1650 for $580'ish at ewiz. 6 cores/12 threads & high clocks to boot. I refuse to buy the 3930k for less than $100 difference and have a krippled (no ecc support) processor. I think 3930k is $499 @ MC
 
Last edited:
plugwash said:
I could be mistaken but I think we will see hell freezing over before we see AMD returning to providing high end competition.
Click to expand...
I think a 28-nm 6-core+ Vishera would provide quite nice high end competition.
 
The board that we have that could be used as a real gaming one is the Intel® Work Station Board W2600CR2.
 
If you're not doing video editing then don't buy a dual CPU machine for a workstation, period. Single-thread performance is always more important as parallelization never scales perfectly (it's just better than nothing when you run out of single-thread performance).
 
You're already going to have trouble finding something that will best your 3570k in games, but if you really wanted to spend twice as much money for that extra 5%, go with Socket 2011.
 
Robstar said:
Depends on what you want to do. Not sure it would be better for gaming, but on the server side the E5-2620 is currently a steal.

The E5-2609 is 4 core, no HT, no turbo (2.4ghz). The 2620, for $100 more gets you 6 cores, 12 threads, and 2.0->2.5ghz so on a dual socket you go from 8 threads -> 24 threads for +$200.

IMHO if you want to game and have something better than i7-3770, I'd snag the E5-1650 for $580'ish at ewiz. 6 cores/12 threads & high clocks to boot. I refuse to buy the 3930k for less than $100 difference and have a krippled (no ecc support) processor. I think 3930k is $499 @ MC
Click to expand...
I have a dual E5-2620 system as a VM host and I would never ever want to game on it, that 2.0GHz speed kills it for anything like that. Seeing 1 CPU core pegged at max and TWENTY THREE other cores practically idle as some single threaded app chugs along sucks ass. An unlocked i5 running @ 4.5GHz would be over twice as fast as a dual E5-2620 for pretty much every game in existence and that will exist within this hardware generation.
 
If this were 2010 and we were talking about 1366 Xeons, it wouldn't be a bad idea at all since you can overclock them on boards like the EVGA SR-2.

But it's 2013 and Intel has locked down all Xeon processors. None can be overclocked, and single threaded performance is far more important for gaming.
 
-Dragon- said:
I have a dual E5-2620 system as a VM host and I would never ever want to game on it, that 2.0GHz speed kills it for anything like that. Seeing 1 CPU core pegged at max and TWENTY THREE other cores practically idle as some single threaded app chugs along sucks ass. An unlocked i5 running @ 4.5GHz would be over twice as fast as a dual E5-2620 for pretty much every game in existence and that will exist within this hardware generation.
Click to expand...

I bet the E5-2620 would be faster than my 1090T @ stock :)
 
I wouldn't focus so much on the CPU right now since you're rocking a current gen part. Instead, focus on the one thing that would have a much more tangible impact with games, like the GPU. Since you are talking about a dual socket setup, then that would imply a fairly heavy budget. Take whatever money you had been planning to spend on the full overhaul to purchase a Titan or two and, depending on what you have now, a higher res/higher refresh rate monitor(s).

I wouldn't worry so much about your CPU right now, as it's likely going to be very adequate for PC gaming over the next few years. When it does become a bit long in the tooth, hopefully we will have 6 or 8 core parts for the mainstream segment, and 8-12 core parts for the enthusiast segment..
 
KazeoHin said:
Dual socket chipsets don't allow SLI/Crossfire

/thread
Click to expand...

Well, you could go top of the line cpus and wait for a single gpu that can handle them.

Edit: but fuck it, what do I know?

Edit2: I have given some thought about going with a larger monitor- I have a triple 1080p setup right now but I dislike gaming with all three monitors due to frame rate and cumbersome nature. It can handle the three 1080p monitors even in BF3 and such decently but its not worth the extra gpu for just triple monitor gaming.

I'm torn between getting a huge like 40" monitor and selling my 3x 1080p monitors because I like the 1080p monitors for multitasking and stuff. Not quite sure where I want to go from here, honestly. Open to suggestions. I suppose I could do 2x40" monitors but even with my big ass desk that would be... cumbersome and maybe the desk is too small?





Continued discussion about monitors here:

http://hardforum.com/showthread.php?t=1755240
 
Last edited:
That's a tough call.

1. Get a 2nd 670 and keep your triple monitors.
2. Get a Titan and keep you triple monitors.
3. Get a single larger monitor with no GPU change.
4. Get a larger monitor plus a 2nd 670.
5. Get a larger monitor plus a Titan.

Decisions, decisions...
 
If you're after a 40" but still want to have extra monitors for multitasking, put the 40" then get 2 1080P monitors that can be turned into portrait mode on each side.
 
Reanimation_LP said:
If you're after a 40" but still want to have extra monitors for multitasking, put the 40" then get 2 1080P monitors that can be turned into portrait mode on each side.
Click to expand...

I've never used portrait mode and I have a feeling I wouldn't like it too much.

I was thinking I could possibly mount a big monitor for gaming and movie watching above my 3x 23" monitors on the wall that's right behind my monitors.. Holy crap that'd be sick.
 
your best bang for the buck is to keep what you have now and replace it with the 4770K when that comes out.
 
Raghar said:
No, but 28 nm 6-core would have similar thermal envelope as i7-3770K Ivy Bridge. 8-core would be too hot to be direct competitor.
Click to expand...

With MUCH worse performance.... Hardly a competitor with Intels mainstream high end products.
 
You know, I wouldn't say worse performance because you'd need to run my benchmark on: I5 (pack4C and pack6C), and 6-core AMD (pack6C). This would provide indisputable data.

People are saying about worse performance, but I seen just few FPS less in gaming benchmarks, and I didn't see any hardcore benchmarks like the my benchmark, or benchmark of something really drastic like PCSX2. (The small problem is PCSX2 supports more than 2 cores now, thus AMD can gain something from these 2-4 extra cores.) In addition, I actually never seen proper IPC, or IPS calculations (for Ivy Bridge and Vishera).

Bench something of equal price, play with multipliers and bench it again when overclocked. These locked multipliers on Intel tends to be real PITA, and in gaming it's not only about high end CPUs, it's about affordable CPUs.
 
I thought we've been over this, that your benchmark is garbage and only proves what you want it to prove. Games care greatly about single threaded performance, and that i7 is going to stomp all over any AMD solution.
 
-Dragon- said:
I thought we've been over this, that your benchmark is garbage and only proves what you want it to prove. Games care greatly about single threaded performance, and that i7 is going to stomp all over any AMD solution.
Click to expand...

You still didn't answer these two questions in the original thread.


As for performance per dollar:

Code: 
FX-6300: 0.2730214769    ?
FX-8350: 0.23491681699   0.68149929845
i5-3470:  ?              0.83734633319
i7-3770K: 0.14176772096  0.53256657082
Note CPUs above were NOT overclocked. Overclocking improves i7-3770K and FX-6300 score a bit.
Of course Crysis is quite bad game for benchmarking CPUs, theirs differences are not obvious. Majority of benchmarks like the above ends with playable speed, vs slightly faster playable speed.

There are no affordable dual-CPU AMD solutions, and cheap Xeon's single threaded performance is at E7200 level, thus dual-CPU gaming is basically out of question for majority of population. For minority, an i7-39xxX, or "my employer paid for these dual HIGH end Xeons"/"I wrote off from taxes these HIGH end Xeons", are definitely an option.
 
Last edited:
I didn't answer because they were stupid questions but I'll go ahead and answer them here.

Virtual registers have nothing to do with a CPU core getting put in an IO wait, again HT lets the OS basically "queue up" a second thread for the CPU to work on when it would otherwise be doing nothing.

As far as the second question, congratulations you've discovered one case where HT can actually hurt performance, which is probably pretty much what your ridiculous "benchmark" does. Yes if you have one very efficient thread that only has an occasional short IO wait and that second thread starts up, the first thread may have a slight performance hit while it waits to get the CPU back. In reality OS schedulers are aware of hyper threading and if they saw two processes constantly conflicting like that, it would likely move those two processes to different physical CPUs if it had them available. If the CPU is almost full do you really want it sitting there doing nothing from time to time so that one process can finish slightly faster even if the system is slower overall? If so, great, you have a tiny edge case scenario where HT doesn't benefit you, just because you are (or found) that tiny edge case doesn't mean HT is worthless, it just means that it's not perfect. For the other 99.9% of the people out there, it actually speeds things up.
 
No such thing as a stupid question.

However, AMD offerings can't compete this round with Intel offerings on any level in almost any broad way. There may be one or two benchmarks that you care about discretely but a tree doesn't make a forest if you catch my drift.
 
When you engineer a question to try to get a specific response, it's a stupid question.
 
-Dragon- said:
When you engineer a question to try to get a specific response, it's a stupid question.
Click to expand...

So, every lawyer in every court room asks stupid questions to every juror and judge.

Socrates asked stupid questions of everyone in order to get them to realize simple truths, etc etc.
 
-Dragon- said:
Virtual registers have nothing to do with a CPU core getting put in an IO wait, again HT lets the OS basically "queue up" a second thread for the CPU to work on when it would otherwise be doing nothing.
Click to expand...
But can't CPU use these virtual registers in combination of some smart out of order computation in the same way as OS is using HT cores? It seems a large virtual registry file, and some smart out of order algorithm should massively diminish importance of HT.

Basically the only difference is the existence of two RIP registers, and there are very few reasons for CPU to not interleave two threads with the same CPUID. (In fact, a core can emulate two RIP registers as a part of virtual file, at least in theory.)


As far as the second question, congratulations you've discovered one case where HT can actually hurt performance, which is probably pretty much what your ridiculous "benchmark" does. Yes if you have one very efficient thread that only has an occasional short IO wait and that second thread starts up, the first thread may have a slight performance hit while it waits to get the CPU back.
Click to expand...
The first thread is often called, event loop. The second thread is often called AI's speculative forward calculation. Starving event loop for any reason is a bad mojo.

In reality OS schedulers are aware of hyper threading and if they saw two processes constantly conflicting like that, it would likely move those two processes to different physical CPUs if it had them available.
Click to expand...
Well output of that benchmark shows attempts of a scheduler to move threads around for a while until it gives up.

Looks like the benchmarks works somehow.
 
Raghar said:
You know, I wouldn't say worse performance because you'd need to run my benchmark on: I5 (pack4C and pack6C), and 6-core AMD (pack6C). This would provide indisputable data.

People are saying about worse performance, but I seen just few FPS less in gaming benchmarks, and I didn't see any hardcore benchmarks like the my benchmark, or benchmark of something really drastic like PCSX2. (The small problem is PCSX2 supports more than 2 cores now, thus AMD can gain something from these 2-4 extra cores.) In addition, I actually never seen proper IPC, or IPS calculations (for Ivy Bridge and Vishera).

Bench something of equal price, play with multipliers and bench it again when overclocked. These locked multipliers on Intel tends to be real PITA, and in gaming it's not only about high end CPUs, it's about affordable CPUs.
Click to expand...

I say worse performance because of what every single professional review out there shows. I'm not making it up and I'm certainly not going to spend time running a random benchmark when there's countless real world tests out there that already prove which processor is the better one, and it's not AMD.
 
xIronCrossx said:
Got one in my basement right now :)
Click to expand...
Want to send it to me? I'm on this weird binge of resurrecting old components and putting them in nice new cases. Currently placing a s478 P4 2.8Ghz in a Fractal Design R4 and a s939 Athlon X2 4400+ into probably a Nanoxia Deep Silence 1.

I don't know why I'm doing this.
 
Raghar said:
You still didn't answer these two questions in the original thread.


As for performance per dollar:

Code: 
FX-6300: 0.2730214769    ?
FX-8350: 0.23491681699   0.68149929845
i5-3470:  ?              0.83734633319
i7-3770K: 0.14176772096  0.53256657082
Note CPUs above were NOT overclocked. Overclocking improves i7-3770K and FX-6300 score a bit.
Of course Crysis is quite bad game for benchmarking CPUs, theirs differences are not obvious. Majority of benchmarks like the above ends with playable speed, vs slightly faster playable speed.

There are no affordable dual-CPU AMD solutions, and cheap Xeon's single threaded performance is at E7200 level, thus dual-CPU gaming is basically out of question for majority of population. For minority, an i7-39xxX, or "my employer paid for these dual HIGH end Xeons"/"I wrote off from taxes these HIGH end Xeons", are definitely an option.
Click to expand...

socket c32 is fairly affordable...
 
You must log in or register to reply here.
Back
Top