Sandy Bridge-E Benchmarks Leaked: Disappointing Gaming Performance?

[cadaveca]

its funny everyones posiutive praise when intel releases early benchies and when AMD does it EVERYONE is critical, typical human population though, unintelligent

Haven't seen ANYONE "praising" Intel in this thread.

All I saw is everyone discussing the validity of how the benchmarks are presented, according to their own uses and needs. Nothing here is disappointing, as those htat tend to know hiow hardware works, realize that CPUs mean very little in day-to-day tasks, and most of us here game @ 1920x1080, not 1680x1050 or lower.

 

[HumanSmoke]

its funny everyones posiutive praise when intel releases early benchies and when AMD does it EVERYONE is critical, typical human population though, unintelligent

Inpai is Intel ? I know they both start with "I" and have five letters, but you should realize they aren't actually the same outfit.

Tom's Hardware had an officially sanctioned preview of SB-E, in much the same way as Anand has in the past previewed Intel products (C2E,C2D,Bloomfield etc.)...AMD had no such preview for BD.

As far as I'm aware the only early benches for BD were people not bound by NDA (OBR in particular)....Inpai would seem to be of the same variety.

Keep up that "intelligent" posting

 

[n-ster]

its funny everyones posiutive praise when intel releases early benchies and when AMD does it EVERYONE is critical, typical human population though, unintelligent

Everybody has been trying to say SB-E sucks in this thread with others defending it, which turned into a benchmark validity discussion. I see no praise of intel either.

BD failed, ESPECIALLY in the gaming department... You want us to fake it being godly?

The question is: Can Sandy Bridge-E beat Bulldozer's "awesome" multi-GPU performance? http://hardocp.com/article/2011/11/03/amd_fx8150_multigpu_gameplay_performance_review/

That's a great review, and I think that people who have the argument about how low res benchmarking isn't as useful as other types, this is a great example

That rule was forgotten when Bulldozer was launched. And remembered when Sandy Bridge-E is showing signs of floppping.

I wish people remembered that saying when Bulldozer was received negatively on launch.

I'll give you that, but there is a reason. SB-E is at least as good as SB in gaming... BD is getting beat by old hardware like i7 870s or something. TBH, BD gets handed it's ass many times by intel, that is the reason of the low price. If it weren't for the price, BD would have been practically useless except in particular situations.

For some reason, there seems to be quite a bit of confusion by various posters here over the above statement, starting with erocker:

People, I don't see how I can put it any more clearly. The idea is to isolate each individual CPU's true performance, so the last thing you want to do is give the graphics card any significant work to do. Heck, if the card could be switched off altogether (possible in Unreal Tournament 2003/4) then you'd have an even more accurate result.

And it doesn't matter if one CPU achieves 200fps and the other 1200fps (6 times faster) you're measuring performance differences between them. This difference will become plenty obvious as time goes by and games become more demanding, giving the faster one a longer useful life. So for example, when the slower one achieves only a useless 10fps, the faster one will still be achieving 60fps and be highly useable.

Of course, it's also a good idea to supplement these tests with real-world resolutions too, as there can be unexpected results sometimes.

Thanks to John Doe for replying with some excellent answers to this misunderstanding.

Will do.
It isn't, as I've explained above in this post.

Yes, of course, lol. Benchmark a bunch of older games with vsync on and they'll all peg at a solid 60fps.

Ideally, you would be looking at gaming performance without the influence of GPU, but that is called a synthetic benchmark sadly I got beat to it though lol

What are people expecting? Its a tweaked Sandybridge with 2 extra cores. Its not going to be miles better just a bit better in some tasks due to the 6 cores.

This is very true. The main advantages of SB-E are in the features and upgrade path

Except most don't realize that depending on VGA, this doesn't move the bottleneck from the GPU to the CPU...it moves it from the GPU to the memory controller. IT does NOT move the bottleneck to JUST the CPU proper.

Triangle and setup data is sent from the CPU to the GPU for every single frame, so really, by lowering resolution and increasing framerate, you are not exactly getting the same effect as it's portrayed by reviewers. You need to lower resolution, and NOT use a high powered-VGA, unless you are just testing CPU-GPU communication. It is NOT 100% just testing CPU performance, and this way of testing is SERIOUSLY flawed.

Falsely increasing the workload turns a real-world benchmark into a synthetic benchmark. Except this synthetic benchmarking practice has no correlation to real-world workloads, at all.


This is why I don't do CPU reviews. I WILL NOT review perforamcne for a CPU in the manner it is currently, by nearly every reviewer out there.

This. The more synthetic you make your game benchmark, the closer it will look like a synthetic benchmark and the further it will look like the gaming benchmarks. I think the best way is to have very CPU intensive games benchmarked if anything, but it is hard to benchmark a CPU, and technically you should benchmark the CPU in every possible way to truly be able to assess it's performance

 

[General Lee]

It's not a big surprise that SB-E doesn't fair any better in gaming. Extra cores do little when games use 2-4 threads at best. Also SB-E doesn't improve IPC nor does it raise the clocks.

The only benefit for enthusiasts is more PCIe lanes for CF/SLI and even that is of limited benefit unless going for triple/quad SLI/CF with triple monitor resolutions.

 

[HumanSmoke]

No, Inpai is benchmarking Intel. We though you were talking about that?

Not really.
Here's lashton's quote:

when intel releases early benchies

This thread concerns someone posting benches to a tech site does it not? I don't see Intel releasing benchmarks here.

If not then you were completely off-topic with the first post

How so? My first post concerned benchmarking practice in general and my particular system usage. So how is...

The 3960X for all intents and purposes is a slightly slower version of the 2600K if the app uses up to 4C/8T (3.3G vs 3.4G)...etc, etc...

...off topic ? I don't believe I wandered off into Bulldozer discussion like some.

we understand english at least as well as you do.

It's English -requires capitalization

 

[brandonwh64]

I wouldn't expect SB-E to be a game changer in the gaming world. 2600K does very nicely for gaming. If you see the BF3 CPU benches, It doesn't take much to play the game even Athlon X2's were getting as much FPS as quads.

 

[theubersmurf]

So my i7 920, running stably at 3.5 day to day, is still a viable cpu, I simply have to do without some of the modern interface like USB 3.0, Sata 3.0, PCI-e 3.0...I'm pretty much okay with that.

 

[HalfAHertz]

SB-e Should be moderately faster in games than SB at stock because if games use more than 4 cores, they can utilize the 2 extra and if they use less, SB-e can clock to a higher turbo mode because of its higher TDP. Tho once you start overclocking that difference should indeed shrink.

 

[n-ster]

Not really.
Here's lashton's quote:

This thread concerns someone posting benches to a tech site does it not? I don't see Intel releasing benchmarks here.


How so? My first post concerned benchmarking practice in general and my particular system usage. So how is...

...off topic ? I don't believe I wandered off into Bulldozer discussion like some.

It's English -requires capitalization

Lol I confused you and lashton xD

 

[HumanSmoke]

Your post was unnecessarily attacking towards cadaveca when all he did was bring an excellent point.

I did not either reference cadaveca, nor refute any of his argument. In fact I would say we are much closer in agreement than not...at least from my viewpoint.
My post was ENTIRELY in reply to lashton's argument...hence why I quoted him/her.

So, that makes two ill-directed presumptions on your part, including one as an excuse for the first.

I have a tendency to re-evaluate a stance, and to apologise if my original position is found to be in error...doesn't look like I'm in any danger of starting a trend, does it?

 

[n-ster]

I did not either reference cadaveca, nor refute any of his argument. In fact I would say we are much closer in agreement than not...at least from my viewpoint.
My post was ENTIRELY in reply to lashton's argument...hence why I quoted him/her.

So, that makes two ill-directed presumptions on your part, including one as an excuse for the first.

I have a tendency to re-evaluate a stance, and to apologise if my original position is found to be in error...doesn't look like I'm in any danger of starting a trend, does it?

ZOMG I can't read lol sorry my bad Makes much more sense now

rofl I made a fool out of myself for a bit there

 

[John Doe]

You're thinking of it wrong. One can't test the true potential of a cpu by using games, period.

Again, I don't care how many people here buy chip specifically for gaming. That has absolutely no bearing on the fact that gaming is still a TERRIBLE cpu benchmark. SB-E isn't the only thing I mention this about.

Uhm, yeah. Just because games don't load up a chip to %100, doesn't mean they're a "terrible" way of testing. The Supreme Commander bench I regularly refer to is a good way of measurement. It scales well over multiple cores while giving performance numbers in frametime, which actually is more relavent than FPS; since FPS stands for how many frames you see per-clock, and frametime = how fast an instruction will come up. So a shorter frametime equals to faster response, which is what you want out of a CPU.

 

[Wile E]

Uhm, yeah. Just because games don't load up a chip to %100, doesn't mean they're "terrible". The Supreme Commander bench I regularly refer to is a good way of measurement. It scales well over multiple cores while giving performance numbers in frametime, which actually is more relavent than FPS; since FPS stands for how many frames you see per-clock, and frametime = how fast an instruction will come up. So a shorter frametime equals to faster response, which is what you want out of a CPU.

Ummm, that's actually EXACTLY what that means.

 

[John Doe]

Ummm, that's actually EXACTLY what that means.

No, it doesn't. Per-thread performance is what matters, not total performance. The chips rarely get %100 usage, so your point doesn't make much sense. Although BD is a 4+4 core logic, it clearly loses to a quad core even when the software is heavily threaded. BD's problem for example is lack of per-thread performance.

 

[n-ster]

No, it doesn't. Per-thread performance is what matters, not total performance. The chips rarely get %100 usage, so your point doesn't make much sense. Although BD is a 4+4 core logic, it clearly loses to a quad core even when the software is heavily threaded. BD's problem for example is lack of per-thread performance.

But doesn't BD lack total performance as well?

 

[Wile E]

No, it doesn't. Per-thread performance is what matters, not total performance. The chips rarely get %100 usage, so your point doesn't make much sense. Although BD is a 4+4 core logic, it clearly loses to a quad core even when the software is heavily threaded. BD's problem for example is lack of per-thread performance.

Nope, total performance is what matters when talking about the available power in a cpu.

Whether you need that power or not is another matter altogether.

 

[cadaveca]

BD's problem for example is lack of per-thread performance.

This is the biggest mis-conception floating around currently. Alot of people will focus on the poor single-thread test results, but not one person has properly identified WHY single-thread performacne in BD is "subpar". Now that I have a chip, I think I can safely say that what I surmised about "why BD performance would not be exciting" prior to the launch, is correct.


Likewise, when it comes to SB-E, unless you truly understand where the extra performance comes in, of course it's going to look poor.

Of course, there is still NDA on SB-E, but that's not something I signed, myself. I can comment all I like about performance, and all I have to say is that I think benchmarking practice by many sites is flawed, and until those practices change, true performance comparisons with actual meaning to end users is not possible. I eagerly await reviews to see what people say. I don't expect much. And I'm NOT talking about performance.

 

[John Doe]

But doesn't BD lack total performance as well?

Yes, but it has more potential under threaded situations. It's more suited for server enviroments and such with being closer to an 8 core CPU than a 4 core chip with HT.

Nope, total performance is what matters when talking about the available power in a cpu.

Whether you need that power or not is another matter altogether.

False. It's about how you distribute that power. Single threaded, multi-threaded etc. You can have a ton of slow cores (say Quad Opteron) yet it'll get outperformed by a Sandy chip. Have a read on the IXBT link, Gulftown loses to Sandy although it has more threads (even in multi-threaded apps).

 

[Wile E]

Yes, but it has more potential under threaded situations. It's more suited for server enviroments and such with being closer to an 8 core CPU than a 4 core chip with HT.



False. It's about how you distribute that power. Single threaded, multi-threaded etc. You can have a ton of slow cores (say Quad Opteron) yet it'll get outperformed by a Sandy chip. Have a read on the IXBT link, Gulftown loses to Sandy although it has more threads (even in multi-threaded apps).

Wait, how does that contradict what I said? I said that all that matters is overall performance potential when you are talking about a cpu's potential. How does SB beating Gulftown have anything to do with this? In fact, it proves my point that per thread doesn't mean shit.

SB-E is SB with more memory bandwidth, and up to 2 extra cores. Come back with this line of reasoning when SB beats SB-E in the same tests.

 

[CyberDruid]

No, between the people here they aren't.

Also, you're wrong about the 2500k. Even in threaded situations, it's still capable of beating the Westmere architecture.

http://ixbtlabs.com/articles3/cpu/intel-ci7-limits-p2.html

If you ask me, I buy my CPU's mainly due to their gaming performance. That's what matters to people building PC's on these forums.

Sez the new guy

 

[Damn_Smooth]

its funny everyones posiutive praise when intel releases early benchies and when AMD does it EVERYONE is critical, typical human population though, unintelligent

AMD's leaked benchmarks were claimed to be "unrepresentative" by the company. Unfortunately for everybody, they were very representative.

I don't see any reason why anybody would praise AMD for the performance that those leaked benchmarks showed, or the end product itself.

SB-e is another beast altogether. It is Sandy with more cores. That is everything I was expecting from it, so I am not disappointed at all. It still has the gaming performance of SB, which is the best in the world right now, with added cores for multi-threaded apps. I don't foresee anybody that does a lot of encoding or runs a lot of programs that use more threads being let down by it.

The fact of the matter is, if you want the absolute best CPU you can buy, this will be it.

That's my opinion on this subject.

 

[John Doe]

Wait, how does that contradict what I said? I said that all that matters is overall performance potential when you are talking about a cpu's potential. How does SB beating Gulftown have anything to do with this? In fact, it proves my point that per thread doesn't mean shit.

SB-E is SB with more memory bandwidth, and up to 2 extra cores. Come back with this line of reasoning when SB beats SB-E in the same tests.

It doesn't. Point is, most apps don't scale over 6 cores, which is why Sandy outdoes Gulftown at 4 cores. So yes, per thread means a lot.

SB-E seems to have (from the given info) worse IPC than SB, and could have been worked on further to improve per-clock performance. Like Nehalem to Gulftown. But Intel didn't because they have no reason to. They're standing on top of AMD. Just add 2 more cores and leave it at that, right? Then jack up the price. There you have it, a $1000 CPU with higher TDP that, most the time, doesn't outperform a $350 chip.

Sez the new guy

Says the guy who were watching you when you were trying to talk with pipe in mouth. There was no need for that, please. I think you need to wake up a little before posting.

 

[n-ster]

It doesn't. Point is, most apps don't scale over 6 cores, which is why Sandy outdoes Gulftown at 4 cores. So yes, per thread means a lot.

SB-E seems to have (from the given info) worse IPC than SB, and could have been worked on further to improve per-clock performance. Like Nehalem to Gulftown. But Intel didn't because they have no reason to. They're standing on top of AMD. Just add 2 more cores and leave it at that, right? Then jack up the price. There you have it, a $1000 CPU with higher TDP that, most the time, doesn't outperform a $350 chip.



Says the guy who were watching you when you were trying to talk with pipe in mouth. There was no need for that, please. I think you need to wake up a little before posting.

You do know that LGA 2011's quad is supposed to be around the same price as a 2700K and the i7 3930K is supposed to be around 70% more expensive (50% more cores + unlocked + premium for being 6 core). It seems reasonable to me. The only 1000$ chip will be the extreme, but Intel would be an idiot to not release such a chip

 

[John Doe]

You do know that LGA 2011's quad is supposed to be around the same price as a 2700K and the i7 3930K is supposed to be around 70% more expensive (50% more cores + unlocked + premium for being 6 core). It seems reasonable to me. The only 1000$ chip will be the extreme, but Intel would be an idiot to not release such a chip

Intel is milking the cash cow as usual. This chip isn't like going from Nehalem to Gulftown. Gulftown improved performance, clocked up easier while working cooler. It was a solid successor over Nehalem, and still is. This chip on the other hand isn't according to the info out there. Do we have it? No. Do they? Yes. Who should I listen to, someone making assumptions here, or the guys that seem to properly preview the platform?

Extra cores of this chip, as said many times before, doesn't help unless your applications are heavily threaded. Being unlocked doesn't mean much either as it's harder to OC a bigger chip. The 2700k can do 5.2-5.4 (yes) under a good cooler. Think of it.

 

[n-ster]

Intel is milking the cash cow as usual. This chip isn't like going from Nehalem to Gulftown. Gulftown improved performance, clocked up easier while working cooler. It was a solid successor over Nehalem, and still is. This chip on the other hand isn't according to the info out there. Do we have it? No. Do they? Yes. Who should I listen to, someone making assumptions here, or the guys that seem to properly preview the platform?

Extra cores of this chip, as said many times before, doesn't help unless your applications are heavily threaded. Being unlocked doesn't mean much either as it's harder to OC a bigger chip. The 2700k can do 5.2-5.4 (yes) under a good cooler. Think of it.

Gulftown was 32nm vs the 45 of Nehalem... When Gulftown was received, many were criticizing Gulftown much like you are SB-E right now... Gulftown did not bring higher memory bandwidth or more PCI-E lanes... SB-E will, much like Nehalem did in the past.

What do you mean by that bolded part?