Retro overclocking: Intel Celeron E3300

Logically, I should have moved straight from Intel NetBurst to the Core architecture. However, I shuffled the deck a bit, and a 45nm Penryn representative was the second CPU to hit the test bench. So, I invite you to the second installment of my August-September (I'm terrible with names) benchmark session! The Celeron E3300 is installed in the motherboard socket, so let's not waste any time!

Processor

The tested sample's marking is SLGU4, and the chip was packaged in Malaysia.

[ + ]

[ + ]

The Intel Celeron E3300 chip's codename is Wolfdale (Penryn architecture). Underneath this CPU's integrated heat spreader (IHS) lies a single 45nm dual-core monolithic die, specifically the 1M modification.

Our test sample has the R0 revision, which is part of the younger Wolfdale core modification. This is great news, as R0 is the best possible stepping in terms of overclocking potential. (For context, besides the R0 revision, there's also M0, which is based on slightly hotter and less successful chips.)

⤢ ВІДКРИТИ

The Celeron E3300 features two cores running at a nominal frequency of 2500 MHz (with a 12.5 multiplier, 200 MHz bus speed, and an effective 800 MHz link speed to the chipset). This chip comes with 1024 KB of L2 cache. Its stock voltage is set at 1.312 volts, and this Celeron's TDP shouldn't exceed 65 watts. But in reality, that's nonsense. According to my measurements, it doesn't even consume 50 watts under full load. Intel likely just set very generous limits so that even less efficient chips could be cooled by standard 65-watt coolers.

Memory support for our test processor is also quite interesting. Here's the thing: the E3300 itself doesn't have an integrated memory controller on its die, so it's logical that the motherboard's chipset handles this. In our case, that means DDR2. However, if you have a more modern board, for example, one with a P45 or X48 chipset, it might also have DDR3 slots.

At first glance, it appears we have a typical representative of the 45nm Wolfdale chip family. And that's largely true, but there's one giant caveat: Intel disabled SSE4.1 instructions in this particular modification. This immediately cripples the E3300 compared to its siblings, effectively turning it into a classic Conroe, which also lacked SSE4.1.

It's still unclear to me why Intel took such a step. Even with SSE4.1 present in the E3300, nothing would have changed, as one megabyte of L2 cache already makes this chip quite weak in many tasks, even compared to low-frequency Core 2 Conroe generation processors. You'll understand what I mean shortly.

Test Bench

⤢ ВІДКРИТИ

• Processor — Celeron E3300;

• Cooler — Thermalright Ultra-120 eXtreme;

• RAM — Two 1GB Nanya Technology modules, totaling 2GB;

• Motherboard — Biostar TP35D2-A7;

• Graphics Card — MSI GTX 770 Gaming OC 2GB (~1085/7000 MHz, Power Limit 115%);

• SSD — KINGSTON 120GB SA400S37120G (Windows 7/Applications);

• HDD — Seagate 2TB ST2000DM008-2FR102 (Games);

• PSU — Chieftec GPS-1250C.

Celeron E3300 Overclocking

At its nominal voltage, the E3300 reached 3400 MHz, which is quite good considering that Wolfdale 1M bins typically receive the lower-quality Wolfdale 3M dies. So, let's start by exploring the relatively safe voltage limits for 45nm chips. I raised the core voltage to 1.44 volts, lowered the multiplier to 10.5, and achieved a stable 3924 MHz:

⤢ ВІДКРИТИ

Again, this isn't exactly a great result. Full-fledged processors like the Core 2 Duo E8400 are capable of reaching 4000-4100 MHz at a similar voltage.

Celeron E3300 Overclocking Settings for 3924 MHz in BIOS:

• Core voltage — 1.440 volts (+0.120 above nominal, multimeter showed a physical 1.425 volts);

• Bus frequency — 373 MHz;

• Bus multiplier - 10.5;

• Bus voltage — 1.350 volts (+0.100 above nominal);

• RAM multiplier in BIOS — 400 MHz (after overclocking, the effective memory frequency was 746 MHz);

• RAM voltage — 1.950 volts (+0.000 above nominal).

Now, let's talk about the FSB wall. As you can see, I managed to push the Celeron to 373 MHz; however, raising the bus frequency by even 1 MHz prevents the system from booting. So, we've hit the FSB overclocking ceiling. But, as you may have already guessed, I lowered the multiplier, which means there's still room for further overclocking. Right? Let's cut to the chase. At first glance, it might seem like we have plenty of options—we could increase the multiplier and lower the bus frequency. In practice, though, the chip became almost unresponsive to voltage increases. Only at 1.6V (which is extremely ill-advised for 45nm chips!) did the E3300 manage to pass a stress test at 4111 MHz:

⤢ ВІДКРИТИ

Does it make sense to gain just under 200 MHz for an extra 0.16 volts? That's a rhetorical question, but we're not here to baby CPUs; we're here to extract their maximum frequencies. So, I consider this a valid result, although I don't recommend anyone replicate it, especially if you want to preserve your CPU's lifespan.

Just a warning: don't be fooled by the seemingly low peak core temperatures of only 71 degrees shown in the screenshot. Core heat isn't always the sole indicator of degradation. A voltage of 1.6 volts itself is dangerous for 45nm dies, even with low core temperatures. So, proceed with caution.

Celeron E3300 overclocking settings to 4111 MHz in BIOS:

• Core voltage — 1.612 volts (+0.300 above nominal, multimeter showed a physical 1.590 volts);

• Bus frequency — 373 MHz;

• Bus multiplier - 11;

• Bus voltage — 1.350 volts (+0.100 above nominal);

• RAM multiplier in BIOS — 400 MHz (after overclocking, the effective memory frequency was 746 MHz);

• RAM voltage — 1.950 volts (+0.000 above nominal).

As an aside, the E3300's results noticeably contrast with what I once achieved on the E8400. If you don't have the time or inclination to read that article, here are some figures from my 2012 experiments:

• 1.456 volts → 4226 MHz.

• 1.664 volts → 4630 MHz.

• 1.744 volts → 4768 MHz.

So, either I got a terrible E3300 sample, or the lower-end Wolfdale variants are truly binned from completely subpar, unsuccessful dies. I'll leave the conclusion on this topic up to you.

Quick testing

After reflecting on the previous Pentium 4 631 article, I concluded that we needed to slightly expand the benchmark and game test suite for older hardware. Therefore, starting with this article in the "Retro Overclocking" series, we'll include a couple more tests: AIDA64 and Crysis. However, keep in mind that I didn't perform these new benchmarks on the Pentium 4, so AIDA64 and Crysis data will only be available for the Celeron and in future investigations.

One more thing: I'll comment on some results, while others I'll leave for your own interpretation. I think that's a fair approach.

CPU-Z benchmark

LinX

This is quite interesting: in LinX, the Pentium 4 631 chip at 4930 MHz surpassed the Celeron E3300 overclocked to 4111 MHz. The difference is quite noticeable: 7.61 versus 7.03 GFLOPS, an 8% lead. Honestly, I haven't tested processors in LinX for a very long time; for me, it's more of a stress test than a benchmark. However, it turns out that in addition to core count, frequency, fast RAM, and bus speed, LinX also greatly favors a large cache, which the Pentium 4 631 literally has twice as much of.

I must emphasize: this is the only test where the old single-core chip with a long pipeline managed to outperform the more modern dual-core processor.

AIDA64

Metro: Last Light

Benchmark settings:

• Resolution: 1280 x 1024;

• DirectX: DirectX 11;

• Quality: Low;

• Texture filtering: AF 4X;

• Advanced PhysX: Disabled;

• Tesselation: Off;

• Motion Blur: Low;

• SSAA: OFF;

An interesting situation emerged in Metro: Last Light: the stock E3300 lost to the overclocked Pentium 4 631 in terms of minimum FPS. This can't be an error, as I ran the benchmark 5 times automatically and selected the average result. So yes, the dual-core Celeron lost to the single-core, dual-thread Pentium in the minimum recorded FPS. However, the E3300 clearly wins in average frame rate, with more than a twofold lead: 38 vs 15 FPS.

As for performance gains from overclocking, it was 34% at 3924 MHz and 37% at 4111 MHz.

Crysis

Due to the peculiarities of Crysis's built-in benchmark, I had to omit the minimum FPS metric, as it consistently registered 0 at any frequency. I couldn't determine if this was a Windows 7 OS issue or a benchmark problem, but it is what it is.

The E3300's overclocking gain is 36%, while an additional overclock from 3924 to 4110 MHz adds another 2%. In reality, this yields no meaningful benefit.

Summary

The Celeron E3300 proved to be a rather controversial processor. I'm basing this on our specific sample, which exhibited relatively poor overclocking potential, especially considering what its older siblings are capable of. Interestingly, this was rarely seen in the previous generation (65nm Conroe). What I mean is that a hypothetical Celeron 420 could easily hit 3100-3300 MHz, just like its architectural counterparts.

Perhaps Intel was more aggressive in limiting cheaper E3000 processors, possibly due to a hypothetical FSB Wall, after the immense success of the E1000 and E2000 series. Alternatively, the 45nm manufacturing process might not have been mature enough when Penryn chips launched. This could have led to an abundance of less-than-perfect dies, which were then frequently allocated to budget chips. But I'm not ready to make any definitive claims.