![[Coi] Mr.CLuBBeR](./core/images/default/default_avatar_medium.png "[Coi] Mr.CLuBBeR"){kind=avatar}
Tweet
A bridge too far: Is the big boy version of Ivy Bridge too little and too late for enthusiasts? (Ivy Bridge-E review)
Ivy Bridge-E review: The release of Intel’s Ivy Bridge-E series of chips is about as anti-climactic as you can get. It’s a chip that’s essentially based on a CPU microarchitecture already going out of style. Haswell, for the most part, has stolen its thunder.
If you’ve read any reviews of when Ivy Bridge processors came out 16 months ago you already know the story: 3D transistors, newer 22nm process and amazing improvements in performance! Well, amazing if you only count the graphics performance. On the x86 side, what we got was a decent, evolutionary upgrade. It instantly replaced the Sandy Bridge parts as our recommended part but it certainly wasn’t the 20 percent performance or more jump people have been chasing ever since the Core 2 and Nehalem Core i7 parts were introduced.
In a nutshell, Intel took the Ivy Bridge microarchitecture and integrated them into an LGA2011 package varying from four cores to six cores for consumers parts. Since LGA2011 has no ability to support integrated graphics, there’s no graphics core – only the x86 cores.
Don’t let all the negative waves get you down though. The new CPU is indeed faster and brings some new features that could prove to be potentially very cool. First up, every Ivy Bridge-E is now unlocked. On Sandy Bridge-E, Intel only unlocked the very top-end part.
Perhaps the sleeper cool feature is the per-core overclocking. The feature will give you the ability to fine tune overclocks from within the OS in real-time. With Sandy Bridge-E OS-based overclocks required a reboot. With Ivy Bridge-E, an overclock can be chosen and the overclock immediately executed. It’s possible that eventually we’ll see apps that let you overclock cores based on the applications that are launched. If a game only uses two threads, you could potentially overclock two cores to far higher clock speeds because you know the other cores won’t be hot.
Of course you don’t give a damn about anything except for the benchmark charts so click on.
Ivy Bridge-E specs compared
Click the next page to check out our Ivy Bridge-E benchmarks.
Ivy Bridge-E Benchmarks
How we tested: We used the same SSD, same GPU, same drives all running DDR3/1866 RAM on Windows 8 for our testing. For the LGA2011 testing we used an Asus Sabertooth X79 board. For LGA1155 we used an Asus P8Z77 Premium board and for LGA1155, we used an Asus Z87 Deluxe.
Since our benchmark mixes in dual, quad and hexa core parts, we wanted to gauge the performance of the new Ivy Bridge-E part so we ran the older Maxon Cinebench 10 benchmark in its single-threaded mode. Thus the differences you’re seeing here are mostly the result of the microarchitecture and clock differences between the chips, not the thread or core count. What you see matches up pretty well. The new Core i7-4960X and its Ivy Bridge cores is just about the same as the Ivy Bridge-based Core i7-3770K. The Sandy Bridge-based Core i7-3930K and Core i7-3820 also offer comparable performance to each other in single-threaded tasks. For the Haswell Haters in the room, take note of the chip’s performance in single-threaded tasks and maybe ask for some A1 Steak sauce to go with that crow.
Moving from single-threaded performance to multi-threaded performance using the same Maxon Cinebench 10 test of a CPU’s ability to render 3D models we see the hexa-core’s chirp up in performance. The Core i7-4960X IVB-E offers a pretty noticeable performance delta from the Haswell quad-core. Cinebench 10 is an older test using an older rendering engine so we’re not seeing the best scaling here going from quad-cores to hexa-core but there’s no doubt the Core i7-4960X adjudicates its place at the head of the class. Of course, some will question the price difference too. More on that later, but it’s probably best to move on to something a little newer such as the Cinebench 11.5 rendering engine.
The Cinebench 11.5 results help push the philosophy we’ve long believed matters: If the apps you push for a living can truly push the number of threads in a hexa-core chip, it’s well worth the investment for more cores. None of the quad-core parts can touch the hexa-core chips. The Ivy Bridge-E Core i7-4960X part outpaces the very fast Haswell by more than 35 percent. More interestingly is the choice for those already on the hexa-core bus. The Core i7-4960X offers a clear win over the older Core i7-3930K but probably not enough to justify an upgrade, even if it were the Core i7-4930K which ostensibly replaces the Core-i7 3930K chip. More on this later of course.
As a backup to the Maxon rendering engines, we also run POV Ray’s render which is a ray tracer. No surprise, the Core i7-4960X is again in a slide place finish while the quad-cores duke it out with Haswell, and then Ivy Bridge (LGA1155) and finally Sandy Bridge-E taking finishing. This test also made us think that maybe we shouldn’t have included the dual-core Core i3-3220 in this but it’s a good point of reference for many people.
The Fritz Chess Benchmark judges how fast a CPU is using the Fritz 12 engine is at playing chess. It spits out different results one of is the CPU’s performance relative to a 1GHz Pentium III CPU. The test is interesting but it’s also clear it doesn’t push all the cores in the hexa-core parts.
The popular 7-Zip archiving app has a built-in benchmark. We threw 8-threads at our chips to see where the pieces would land and came up with a confusing result. The good news is for Ivy Bridge-E which is the clear winner with the Core i7-3930K also representing for the hexa-cores. What we didn’t expect, however, was the performance of the Haswell chip which barely pulls ahead of the quad-core Ivy Bridge Core i7-3770K part. It’s frankly a very confusing set of results for the Haswell chip which will have to dig into at a later time or perform a restest.
PC Mark 7 is a pseudo synthetic/real-world test which attempts to simulate real-world workloads. The results we’re seeing indicate that it doesn’t seem to stress all the cores as both Sandy Bridge-E chips, the Core i7-3930K and Core-i7-3820, are virtually the same despite differences in clock and core count. Both Ivy Bridge-based chips, the Core i7-4960X and the Core i7-3770K, also pull with the same overall score despite differences in clock speed and core count again. The big winner is Haswell which has an edge against all other CPUs here by a noticeable margin. The test itself is designed to be “real world” which we’re also taking to mean not stressful apps because even the Ivy Bridge-based dual-core Core i3-3220 and its locked down clock speed of 3.2GHz gives us 80 percent the performance of the pricier chips. This shouldn’t be news to anyone though. If your chores are browsing, very light video encoding and office chores, you’d be hard pressed to see the difference between a $1,000 Extreme Edition chip and a cheapo $125 chip. Of course, you can scroll also scroll up and look at the chart that calls for more threads and see how that cheapie chip does.
The computation work load of PC Mark 7 supposedly isolates the Computation workload of a PC but the result we’re seeing doesn’t exactly agree with that as there’s not exactly much of a difference between any of the chips in our eyes. The Ivy’s and Sandy’s pretty much tie with Haswell on top, but not by much. Interestingly (or perhaps disturbingly) PCM7 seems to think a dual-core Ivy Bridge chip locked down at 3.2GHz is isn’t really that much slower than entire pack. We guess this is the kick in the butt to convert over to PC Mark 8 for all of our testing finally.
HDR Soft’s Photomatix is one of the most popular apps for creating high dynamic range photography. For our test, we take a set of nine NEF images shot with a Nikon D800 and crunch them using the app’s batch mode. Lower is better and the results show the new Core i7-4960X out pacing even that fast as hell Haswell part. That’s some good news for those who prefer cores because the Core i7-3930K couldn’t defeat the Haswell chip—this despite HDR Soft use of multi-threading in Photomatix which should give the hexa-core chips the edge. To be honest, we’re not seeing the scaling we’d like to see for the hexa-cores to pay their own way. If Photomatix HDR is your primary tool, a Haswell platform probably makes a lot more fiscal sense.
Click the next page for more benchmarks that include ProShow Producer, Stich.Efx, and more.
Proshow Producer 5
We’ve used Proshow Producer since the first dual-core chips hit the scene. Unfortunately, Proshow hasn’t really scaled with today’s hardware. It seems to top off with four-threads with quad-core chips giving the best result. Still, that doesn’t invalidate the results as it is a very popularly professional application being in use today by photographers across the world. For our workload, we shot a couple of hundred images using a Canon EOS 5D Mark II, create slideshow and export it to an MPEG2 file at 1080P resolution. The winner is the quad-core Core i7-4770K with the newer Core i7-4960X coming second. Overall though, the results are pretty flat and you’d be perfectly fine with an older Core i7-3820 or Core i7-2600K with this one.
Our last photo-related tasks involves the mother of all photos today: Gigapixel imaging. For those who aren’t up on the giga thing that’s a billion pixels. For our workload we shot 287 images using a Canon EOS 7D on a GigaPan Epic Pro and stitched the images together to make a 1.1 gigapixel panorama using Stitch.Efx 2.0. Stitch. Efx 2.0 isn’t all multi-threaded. The first third where the app aligns the images is mostly single-threaded with the final two-third exploiting more cores. Like many of the other apps we use, it’s a good representation of where today’s apps live. The truth is that few apps are capable of pushing an 8-core or even 6-core processor. The result? The Haswell-based Core i7-4770K still wins but the Ivy Bridge-E comes close. Very close. But again, like we’ve seen before, if Gigapixel imaging using Stitch.Efx 2.0 is your thing, stick with the Haswell for any new build.
This one was an easy one to forecast: Video encoding is an area where more cores still matter. Both hexa-core chips come out on top by a pretty sizeable margin. It’s enough that if you’re encoding a video and it takes three hours on your brand new Core i7-4770K, you can expect to lop off an hour if you had a Core i7-4960X part. That’s an extra hour you could have spent playing games or surfing, for, um, NSFW material for frak’s sake!
TechARP’s X264 HD 5.01 test uses the free X264 encoder library. It’s thread-heavy and we expected the hexa-core chips to win by a good margin and they do. The new Core i7-4960X smacks the Haswell Core i7-4770K across the jaw. It even gives a good showing against the hexa-core Core i7-3630K part too. Again, if you’re looking at an encode taking a few hours, extra cores will put sizeable dents into those encode times.
Pass 2 of the X254 HD 5.01 test is usually even more sensitive to thread and core count. What can we say except that if your boss has saddled your “workstation” with dual-core Ivy Bridge Core i3 part, print up this page and show him how much more work you could be getting done if he wasn’t such a cheap son of a bitch.
For our Adobe’s Premiere Pro CS6 test, we produce a video with multiple 1080P sequences and transitions and export it to a 1080P Blu-ray ready file. It’s generally pretty thread heavy and leaves dual-cores in the dust. Even quad-cores have a difficult time with this test although, Intel’s Haswell has impressive performance. The winner is still the new Core i7-4960X but not by a huge margin against the other hexa-core chip here. We’re actually a bit disappointed the Ivy Bridge-E didn’t eat the Core i7-3930K’s breakfast in this test.
It’s hard to justify the need for a quad-channel setup as we rarely see it making a difference in the workloads we run. That doesn’t mean people don’t like to see these charts though so bam! Look at how much more memory bandwidth you get with the quad-channel parts! It’s more than double the bandwidth of those little toy CPUs with their wussy dual-channel RAM configurations. Yes, we are crying for you too. Back in reality land though, we haven’t seen that memory give the LGA2011 chips an edge although, it’s possible that running dual-channel mode on the LGA2011 parts could hobble their performance. We’ll have to investigate that in the future but there’s really very little reason to run dual-channel mode on a quad-channel chip so it’s an unrealistic configuration.
First, the Core i7-4770K has fairly insane L1 cache performance. Second, the hexa-core chips have an interesting L1 cache performance advantage over their quad-core counterparts.
Haswell may have insane L1 cache performance but if the cores can’t find the data in the puny L1, it’ll have to move onto the L2 where the Core i7-4960X has the advantage over the Haswell chip.
Because no one was ever hurt by one too many charts, we run it all the way out to the L3 cache where the Ivy Bridge-E still has the edge over all. Trying to gauge where the cache performance of each chip will help or hurt it is probably well beyond us. For that, it’s probably to go visit a real smart guy like Anand at Anandtech.com to suss it out. Yeah, we did just forward you to another tech site but it’s midnight and statistics show 72.4 percent of nerds never read the stuff written under a bar chart anyway.
Click the next page to check out how Ivy Bridge-E did in some graphic-intensive benchmarks.
We’ll open up the gaming testing by running Valve’s old particle test benchmark which was originally made to show off how we all needed quad-core CPUs. It’s built using Valve’s particle engine and should be an indicator of how well each platform handles particles in a Source title. We’ve long felt the test is memory latency sensitive and cache-size sensitive as hexa-cores have long ruled the roost here. Those chips also almost always have very large caches too.
3DMark 11 is first and foremost a GPU benchmark and this, dear nerds, is how it should look on any benchmark designed to weigh GPUs. Since all of our test configurations used the same driver and the same GPUs, we should see no difference in the scores. For the most part we didn’t. Only the Core i3 with its clocks stuck down in the 3.2GHz range shows a hit and that’s not very much showing how 3DMark 11 isn’t sensitive to clock variations either.
3DMark 11 does have a physics component that stresses core count. The chips with the most cores win and the Ivy Bridge-E Core i7-4960X wins by a good clip. Interestingly, the Haswell chip doesn’t out pace the quad-core Ivy Bridge brethren at all which is a bit of a surprise.
Since many will want to know how how the new Ivy Bridge E does in the new 3DMark we’ve also included it. And like 3DMark 11, we see a well balanced graphics benchmark that is slightly more sensitive to CPU clock speeds than 3DMark 11. Still, it’s a solid gauge of pure graphics performance.
The physics component of the new 3DMark also favors core and thread count. Here, the Core i7-4960X easily takes the gold with the older Core i7-3930K taking the silver. In this newer version of 3DMark, the Core i7-4770K does better than in the older 3DMark and we award it the bronze. Dual core Ivy Bridge? Go sink your misery in a Big Mac.
We’ve recently started running the Resident Evil 6 benchmark as a gauge of “real-world” gaming performance. Unfortunately, what we can’t figure out is the results we’re getting. The Haswell Core i7-4770K wins. OK, we get it. It’s newer, has better IPC, and is the state of the art in Intel CPUs. But why does the LGA1155-based, Ivy Bridge Core i7-3770K do so much better than the LGA2011-based, Ivy Bridge E Core i7-4960X? It frankly makes no sense as the clock speeds should be the same. We though that it might be the PCIe performance of the LGA1155 and LGA1150 but we’re using a PCIe 2.0 card, not a PCIe 3.0 card for our testing. We do know that Resident Evil 6 runs perfectly fine on a dual-core since the Core i3-3220 does just fine here—the only reason it’s score was lower is probably due to its lower 3.2GHz clock speed.
We run Hitman: Absolution at a low resolution and low quality image settings to actually simulate the performance delta you’d get if you had a four-way GPU setup. That is, we run the gaming benchmark with the assumption that you have the faster GPU in the world and if something where slowing it down, it would be the CPU. In Hitman: Absolution we saw a result we didn’t expect too. The Ivy Bridge E part beats the Haswell part. Why? We doubt it’s the core count. It’s unlikely that Hitman: Absolution will stress all four cores plus the Hyper-Threading available in the other chips. With clock differences fairly small this leaves us to give the credit to either the memory bandwidth advantage or the larger cache. We’re not looking at a huge difference, but it’s still surprising to us as we usually expect the quad-core CPUs to have the advantage in the vast majority of games.
Our last gaming test is the CPU test built into Total War: Shogun 2. It simulates a field crammed solid with Samurai and is designed to stress the CPU performance rather than GPU performance. The Ivy Bridge E part gives up a fair showing by actually tying the Haswell part. The test itself, doesn’t appear to stress all six-cores available but it definitely saw the dual-core Ivy Bridge part gasping for air. This reinforces our belief that for most gaming rigs, a quad-core is more optimal over a dual-core machine today.
Click the next page to read our Ivy Bridge-E conclusion.
Ivy Bridge-E Conclusion
Initially, we wanted to have an Intel Core i7-3960X here going toe-to-toe with the Core i7-4960X since both parts are priced the same. We didn’t have one available though as both processors were hard at work testing other hardware so we fell back on the Core i7-3930K part to represent the Sandy Bridge E hexa-core parts. In the end, we realized it didn’t really matter.
The hard truth is that if you’ve already put out the big money for a Core i7-3960X or Core i7-3970X, don’t worry, you don’t need to upgrade. The differences in performance between those two and the Core i7-4960X is so small, it wouldn’t be prudent to pay $1,000 for such an upgrade.
Intel will again not include a stock cooler on its Ivy Bridge E partsBut what if you have a quad-core Core i7-3820? Your patience has rewarded you my friend. Going from the quad-core Core i7-3820 part to the mid-range Core i7-4930K will yield a sizeable performance upgrade for your thread-heavy tasks. We’ve said it before and we’ll say it again, if you compute for a living pushing pixels in 3D rendering, video encoding or other work-station level apps that will actually use the cores, it’s well worth the cash outlay for a hexa-core part. Even better, the upgrade process to the new Ivy Bridge E should be fairly painless. Just make sure your board maker has released a BIOs with support for it and drop in the new part.
This has probably lead to you believe this new chip is really just Ivy Bridge Meh. We don’t agree with that. Ivy Bridge-E is a solid part for what it is. It is simply the fastest consumer CPU out right now for thread-heavy tasks and gives Core i7-4770K a good run under more relaxing workloads. Reports of overclocking on Ivy Bridge-E from vendors are also fairly positive with most saying 4.5GHz should be capable by most of the chips out there.
We’ll also note that we didn’t have time to test the per-core overclocking ability which could be a very nice plus on the Ivy Bridge-E lineup. Per-core overclocking allows OS overclocking based on load. As the apps mature, you’ll likely see the ability to set the processor to say, overclock to 4.9GHz when X game or Y app starts because you know the workload won’t generate the thermals of your 3D rendering app or video encoder. While neat, this feature is also something that will appeal mostly to those coming to the platform or hexa-core parts. Existing Core i7-3930K, Core i7-3960X/70X users probably shouldn’t pay just for this feature.
What should I buy? It very much depends on what you do. To help break it down, we’ll give you three basic scenarios:
The 90 percent gamer
This person owns a PC primarily for gaming and really couldn’t give two iPads about encoding or 3D rendering or other thread heavy tasks. For this person, Core i7-4770K or probably even better for those on tighter budgets, the LGA-1150 Core i5-4670K, makes a lot more sense. Sure, you’ll get advice that you should “just buy” an LGA-1155 part because they overclock “better.” We think that’s bad advice because the LGA1155 platform is a deadman walking. LGA1150 will be the standard platform for the next two years and will receive all of the attention and updates. Still, LGA1150 and a Haswell chip is the best forward-looking platform for the 90 percent gamer who won’t run more than, say, three GPUs in SLI or CrossFireX.
Works for a living guy
This person pushes pixels professionally and time is money. The faster a project can be rendered, encoded and sent out means more money in the bank. Sure, gaming is also important but work takes priority on this box. For this guy or gal, Core i7-4960X or Core i7-4930K is the best investment. As a plus, you’ll get the most PCIe lanes for RAID adapters, multiple GPU configurations and other nifty stuff you need. Well, except for Thunderbolt. For some odd reason, there’s still no ability to run Thunderbolt on an LGA2011 box without integrated graphics support. What’s really confusing is that the people who would actually want to use super duper fast external I/O are actually on LGA2011 platforms running hexa-core or more CPUs.
LGA2011 is still the only consumer Intel platform that currently lets you hit 64GB which is also a plus if you truly need the RAM for your workloads.
The Average Nerd
The average nerd plays games and uses his or her PC for all things computing including editing video and photos including gaming. For this nerd, we’re still recommending LGA1150. With it you get the six SATA 6 ports, the latest Intel chipset. Unless he or she is getting paid to do the work, a Core i7-4770K or Core i5-4670K makes more sense. How would we determine which way to break it? If encoding and 3D rendering is more important, pay for Hyper-Threading. It’s well worth it. The only reason we could see for this person to go with LGA2011 is if he or she is building a box with the intention of getting a job crunching pixels all day where a hexa-core or more would make a difference. In that case, starting with a Core i7-4930K isn’t a bad idea, but you need to be able to justify that you need those extra cores, more RAM or PCIe lanes. LGA2011 is also the only consumer platform that gives you the option of running 4/6/8 cores
More...