cache for cash

Review: Ryzen 7 5800X3D is an interesting tech demo that’s hard to recommend

Big cache benefits games, but lower clock speeds hurts everything else.

Andrew Cunningham – Apr 15, 2022 6:08 pm | 148

AMD's Ryzen 7 5800X3D. Credit: Andrew Cunningham

AMD’s AM4 socket has had a long and successful run on the desktop, ushering in the Ryzen processor lineup and helping AMD compete with and outperform Intel’s chips for the first time since the mid-2000s.

The aging socket’s time is coming to a close later this year when the Ryzen 7000-series chips are launched, but AMD is sending it off with one last high-performance processor: the Ryzen 7 5800X3D, which launches on April 20th for $449.

AMD uses a unique packaging technology called “3D V-Cache” to triple the amount of L3 cache on the processor, from 32MB for the standard Ryzen 5800X to a whopping 96MB. This new tech feels like an experiment in some ways. Unlike other Ryzen CPUs, the 5800X3D doesn’t offer overclocking or power consumption controls, and its clock speeds are a bit lower than the standard 5800X. But AMD says that the extra cache allows the 5800X3D to outrun Intel’s fastest CPUs when it comes to gaming.

We’ve run some tests on the 5800X3D to find its strengths and weaknesses and to get a sense of when you’ll notice the impact of the additional cache. This is undoubtedly an interesting processor, but its pricing and extremely specific performance advantages will limit it to a niche of a niche.

3D V-Cache, in brief

A visualization of AMD’s 3D V-Cache. Credit: AMD

Architecturally, nothing about the Zen 3 cores that power the 5800X3D has changed compared to the vanilla 5800X. We’re still talking about an 8-core, 16-thread core complex die (CCD) built on TSMC’s 7 nm process, with its 32MB of L3 cache intact. Just as Apple bulit interconnect support into the M1 Max to support the M1 Ultra, AMD created Zen 3 to support 3D V-Cache technology whenever it was ready to ship.

The main change is that AMD and TSMC have reduced the physical height of the CCD die so that a CPU package with 3D V-Cache won’t need to be physically taller than a CPU package without it. This adjustment preserves compatibility with existing CPU coolers.

The 64MB of additional L3 cache, also built on TSMC’s 7 nm process, is physically stacked on the Zen 3 CCD and connected with direct copper-to-copper bonding. The result is something that the system sees as one large pool of L3 cache that can all be treated the same—the 64MB of stacked cache isn’t an L4 cache, and the 32MB of L3 cache built into the CCD doesn’t have any performance advantage compared to the cache stacked on top.

Copper-to-copper bonding is used to fuse the CCD and the additional cache together. Credit: AMD

One side effect of this packaging technology is that the 5800X3D runs at a noticeably slower clock speed than the 5800X, and AMD doesn’t officially allow any overclocking or power adjustments when using the 5800X3D. AMD is pushing the 5800X3D mainly as a gaming processor, and that’s because games benefit more consistently from having a bigger pool of cache to play with. For workloads that care less about cache and more about clock speed—as we’ll see when we start benchmarking—the 5800X3D can be slower than the regular 5800X, which AMD freely admits.

AMD’s 8-core Zen 3 CPUs	Street price	Clocks (Base/Boost)	L3 cache	TDP	PCIe support
Ryzen 7 5700G	$280-300	3.8/4.6	16MB	65W	3.0
Ryzen 7 5700X	$299	3.4/4.6	32MB	65W	4.0
Ryzen 7 5800X	$340-360	3.8/4.7	32MB	105W	4.0
Ryzen 7 5800X3D	$449	3.4/4.5	96MB	105W	4.0

AMD played coy about whether we could expect future Zen 3 CPUs with 3D V-Cache enabled, but reading between the lines, it seems unlikely. 3D V-Cache will be one of the tools in AMD’s toolbox when it comes to boosting performance for Zen 4 and the first Socket AM5 platforms—along with a 5nm TSMC process, DDR5 support, and other architectural improvements—but I don’t get the sense that the 5800X3D will be followed up with an expanded lineup of Zen 3-based X3D chips.

AMD says that motherboards will need a BIOS update to see and use the extra 64MB of cache—look for AGESA version 1.2.0.6b or higher in the release notes. Motherboards that support other Ryzen 5000 CPUs will work with the 5800X3D, but they won’t be able to access the extra cache, defeating the purpose of spending more money on the CPU in the first place.

Performance

Our Intel and AMD test beds reuse the same EVGA Nvidia GeForce RTX 3070 FTW3 GPU, EVGA Supernova 850 P6 power supply, and Vetroo V5 CPU cooler. For this review, our DDR5-based Intel system uses this 4,800 MHz DDR5 RAM kit from Crucial. The AMD and the Intel DDR4 systems all use this 3,200 MHz DDR4 RAM kit from Crucial instead. The AMD system uses an Asus ROG Crosshair VIII Dark Hero motherboard; the Intel DDR5 system uses a Gigabyte Z690 Aorus Pro motherboard.

Admittedly, using a GeForce RTX 3070 to test CPU gaming performance is not ideal. While it’s a great GPU for most games—and, according to Steam data, considerably more common than the 3070 Ti, 3080, 3080 Ti, or 3090—game performance will still become GPU bound rather than CPU bound once you’re pushing a couple hundred frames per second at 1080p.

Our real-world game benchmarks try to stick to games and resolutions where you can tell the difference between different processors or use CPU subscores rather than average fps numbers where they’re available. But it’s worth noting that if you don’t have a 3080 or 3090-class GPU, and you’re not trying to push well over 100 fps at 1080p, your GPU will still almost always be the bottleneck for gaming performance, and you should usually save some money on your processor and spend it on your GPU instead.

Some more details on the processors tested, in addition to the 5800X3D:

A Core i7-12700, paired with DDR5 and tested at its default 65W PL1 power limit and 125W PL1 power level (chosen because it’s the same power limit as the Core i7-12700K).
A Core i9-12900, paired with DDR5 and tested at a 65W PL1 power limit and a 150W PL1 power limit (again, the same as the i9-12900K).
The Ryzen 5 5500 and Ryzen 5 5600 CPUs we just reviewed, as well as a Core i5-12400, represent some more budget-friendly gaming CPUs.

We don’t have as many results from the regular Ryzen 7 5800X as we would like to have, and we haven’t been able to test the new Ryzen 7 5700X—at $300, the latter is one of the best values in AMD’s CPU lineup right now, along with the Ryzen 5 5500 and 5600. Our review will focus primarily on comparisons between the 5800X3D and Intel chips in a similar price range. The Tom’s Hardware review features more extensive testing with the 5800X and 5700X if you’re building or upgrading an AMD system and are more interested in those comparisons.

You may also see slightly better performance from the 5800X3D if you use faster memory in your system since this will increase both your memory bandwidth and the speed of AMD’s Infinity Fabric interconnect. Our testbeds use 3,200 MHz DDR4, but 3,600 MHz DDR4 arguably offers the best combination of price and performance if you’re building from scratch.

Beginning with general-purpose CPU tests, you’ll see that the 5800X3D is outgunned not only by the higher core counts of the Core i7-12700 and Core i9-12900, but occasionally by the regular 5800X as well.

That’s the upshot of the clock speed deficit we talked about earlier—if your app doesn’t care as much about cache size, then the 5800X3D’s benefits evaporate. For anyone whose gaming rig does double-duty as a photo and video editing workstation or as a streaming PC, the 5800X3D isn’t a particularly impressive general-purpose CPU compared to other options in the $300 to $500 price range.

The 5800X3D’s benefits also weren’t readily visible in the synthetic 3DMark benchmark, with both overall scores and CPU subscores falling below the Core i7-12700 and Core i9-12900 in the Time Spy test.

The 5800X3D performs better in most games, as promised. The chip’s CPU subscores in the Forza Horizon 5 and Shadow of the Tomb Raider benchmarks are consistently at the top of the charts, as it was in the Final Fantasy XIV benchmark. There are some exceptions, though. The Ashes of the Singularity benchmarks ran better on the Core i7 and Core i9, suggesting that this game cares more about clock speeds than it does about cache.

Even for games like Borderlands 3 that were GPU bound by our RTX 3070, the 5800X3D generally improved the minimum framerates we saw in our tests. That translates to less stuttering and hitching during gameplay, something that’s just as, if not more, important to the overall experience than your average FPS game. But it’s worth noting that the average framerates in these titles were uniform across all the CPUs we tested since they’re being limited by our RTX 3070 GPU and not the processor. If you’re using anything other than top-end GPU, you won’t benefit as much from the 5800X3D’s improvements.

Power use and efficiency

Compared to our last few CPU reviews, we’ve tweaked how we’re measuring power usage, measuring CPU package power using HWInfo rather than power consumption at the wall. It doesn’t actually change our numbers much—we’re reusing the same GPU and other components across our Intel and AMD testbeds—but it does make sure the focus is on the CPU, rather than the other components.

The 5800X3D’s power efficiency as measured in our Handbrake test is decent, but generally outgunned by the Core i7 and i9 chips. A heavily multithreaded test like this one tends to show Intel’s current-gen CPUs in a favorable light because of their additional CPU cores, which make a noticeable difference despite being relatively low-performance “efficiency” cores. At their stock 65 W power limits, both the Core i7 and Core i9 chips complete the Handbrake encode in roughly the same amount of time while using less power. Raise the Intel chips’ power limits to 125 W (the default power limit of the i7-12700K and i9-12900K), and the CPUs use more power but also get the work done much faster.

On AMD’s side, the most efficient chip here is the humble Ryzen 7 5700G. Its 65 W power limit and smaller L3 cache make it noticeably slower than the other 8-core Ryzen chips, but it’s also a lot faster than the 6-core Ryzen 5 CPUs in roughly the same power envelope. Definitely keep those differences in mind if you’re building a mini-workstation or casual gaming PC without a dedicated GPU.

A note on heat

In my experience with Ryzen 7 chips in a few different motherboards, I’ve noticed they tend to run hot when you leave the settings at their defaults. This will vary from motherboard to motherboard, but as it happens with Intel CPUs, motherboard makers sometimes allow AMD CPUs to draw extra power to improve performance.

For the Ryzen 7 3700X and 5800X systems I’ve built, I’ve found that manually lowering the power settings (in AMD parlance, the power setting defaults or even a little lower can reduce temperatures by as much as 10° or 20° Celsius without affecting performance at all. You’ll find lots of threads about this on Reddit.

H.264 encode	H.264 encode temp (C)	CPU package power
Ryzen 7 5700G	69	75 W
Ryzen 7 5800X	84	122 W
Ryzen 7 5800X3D	90	109 W

I bring this up because, while the 5800X3D exhibited reasonable power consumption throughout our testing, I found it usually ran at or near its top-rated temperature of 90° Celsius during our more strenuous multicore CPU tests. The Vetroo V5 CPU cooler I like certainly isn’t the best one you can buy, but it is rated for CPUs up to 150 W, and it was still having more trouble cooling the 5800X3D than it did cooling the i9-12900 or the regular 5800X.

And because the power limits on the 5800X3D can’t be adjusted like they can on other Ryzen chips, the only way to get cooler temperatures if you want them is to invest in a better, more expensive cooler—just something to consider.

An interesting proof of concept

The Ryzen 7 5800X3D is probably best approached as a proof of concept rather than a CPU you should seriously consider for your next build.

AMD’s claims about performance are true, and it bodes well for the Ryzen 7000 series that adding extra L3 cache to the same old Zen 3 architecture can boost performance by this much in games and a few other tasks. As a tech showcase, it’s effective! And that AMD can still squeeze extra performance out of the aging AM4 socket is a testament to its versatility and longevity.

But you only really see the 5800X3D’s benefits when you’re trying to achieve extremely high framerates at relatively low resolutions. That’s fine if you’re trying to play games on a 240Hz monitor at 1080p. But at 1440p and 4K, your GPU will still usually be more important than your CPU. And for less money than the 5800X3D costs, you can get very good gaming performance along with more balanced CPU performance for creating content and doing other work.

For example, the Ryzen 7 5700X and Core i7-12700 are between $100 and $150 cheaper than the 5800X3D. Both provide pretty good gaming performance and as-good-or-better performance in non-gaming tasks that are more sensitive to clock speed and extra cores. The 5700X (and the 5600 and 5500) also provide a most cost-effective upgrade path if you want to boost an older PC with a Ryzen 1000, 2000, or 3000 CPU in it. And then there’s the Ryzen 5900X, which has the same $450 listing price but sometimes sells for less and includes 12 CPU cores instead of eight.

Even for the market AMD claims to be aiming at—high-end gamers who care about the highest possible performance more than anything else—the 5800X3D’s lack of overclocking support makes it a bit less appealing. That it can sometimes outrun the $600-and-up Core i9-12900K makes it look like a good value on paper. But the $520 Core i9-12900 and $350-ish Core i7-12700 are both cheaper than the i9-12900K while providing similar gaming performance (especially once you’ve turned their power limits up a bit from the defaults). They’re also more well-rounded chips than the 5800X3D, providing substantially better performance for multi-core productivity and content creation work.

In short, the 5800X3D is a neat chip, but it’s not what I recommend to anyone putting a system together right now.

The good

An impressive technical feat. Adding cache is clearly an effective way to boost performance in many workloads that doesn’t require adding more cores or running at higher clocks.
Drop-in upgrade for virtually all existing socket AM4 systems after a BIOS update.

The bad

Other chips, most notably the Ryzen 7 5700X and Core i7-12700, are better values for your money.
Lower clock speeds mean the 5800X3D is sometimes slower than the vanilla 5800X in tasks that don’t benefit from added cache.
No overclocking support or power limit adjustments mean you can’t really increase its performance or power efficiency.
Runs a bit hot.

The ugly

It’s hard to justify spending $450 to upgrade an AM4 system when AM5 and Ryzen 7000 is around the corner.

Ars Technica may earn compensation for sales from links on this post through affiliate programs.

Listing image: Andrew Cunningham

Andrew Cunningham Senior Technology Reporter

Andrew is a Senior Technology Reporter at Ars Technica, with a focus on consumer tech including computer hardware and in-depth reviews of operating systems like Windows and macOS. Andrew lives in Philadelphia and co-hosts a weekly book podcast called Overdue.

148 Comments