Introduction, Specifications, and Pricing
With the release of the 10th gen Comet Lake series of CPU from Intel, we knew it would not be long until we saw a response from AMD. We did not necessarily expect it this quickly as you are all aware of now that AMD has been hard at work optimizing their 7nm process, and therefore they released a new XT series of CPU, which is what we are looking at today.
The CPU’s we have are enhancements to three models already in the Ryzen arsenal. The first will be the Ryzen 5 3600XT, which is essentially a 3600X with a 100MHz boost to the turbo speed. Next will be the Ryzen 7 3800XT, which is a 3800X with a 200MHz bump to the turbo speed. Lastly will be the Ryzen 9 3900XT, a 3900X with an extra 100MHz added to the turbo.
Here we see the 3600X vs. the 3600XT, and as you can see, it has all of the same features of the standard 3600X just with an extra boost to the turbo of 100MHz. Both CPU come with the Wraith Spire cooling solution in the box, so the turbo bump is where the difference lies.
The 3800X vs. 3800XT is a bit more detailed as the boost speed increases by 200MHz for the XT model. Also, the 3800X comes with a Wraith Prism model cooler with RGB lighting. The XT variant comes sans cooler as AMDs internal data shows that most users deploy aftermarket cooling at this level and above. This is not a horrible assumption, but it does take away from the value prospect for some users.
Lastly, we have the 3900X vs. the 3900XT. Once again, we see a 100MHz bump to the max turbo for the XT variant. Also, here we see that the 3900X came with the Wraith Prism model cooler with RGB lighting, and the XT variant omits this as well. This is likely the right choice overall with the stock cooling solution not always being sufficient for maximum turbo levels, and pushing users to equip better aftermarket coolers ensures better boosting performance.
All of the X vs. XT variant of CPU have the same base clock; only the turbo is bumped for the XT models. The TDP rating is also listed as the same, and we will see how close they are in our testing. PCIe gen 4 is still here, and all other aspects are identical between the XT and X series CPUs.
The XT variants of CPU are priced at the same level of the X versions, which means the X variants will either see an EOL or a price adjustment to put pressure on the Intel competing stack further.
Let’s push ahead and check out the marketing AMD has shown for these new XT CPUs, and then we will test their performance claims in our lab.
The Ryzen 5 3600XT has the full-size square box due to it shipping with the Wraith Spire cooler, but otherwise is identical to our 3600X packaging.
The Ryzen 7 3800XT comes in a slimmer packaging profile allowed by the omission of the cooler, so all you have is the instructions and CPU.
The Ryzen 9 3900XT also comes in a thinner profile package; however, it is a bit nicer presentation. While the other lower CPUs have a thin card stock box, the 3900XT has a thicker rigid box like the 3900X. There is also foam in place to hold the CPU and its instructions since that is all it comes with.
AMD 3000XT Series Marketing Points
Here we give space to the manufacturer to talk about their marketing points, and we assess them and provide our point of view on the claims.
AMD opened its briefing with the first slide detailing the fact that they have learned a lot about 7nm with the first Ryzen 3000 CPUs. Therefore they could optimize the process to squeeze a bit more form them.
AMD did not waste any time with the XT CPUs; it was upfront. They broke it down to “we were able to enable higher turbo clocks for these CPUs due to what we learned with 7nm.” As you can see here, the turbo or boost frequency was increased on each chip the same as we mentioned previously.
AMD is proud of the fact that the XT processors are drop in compatible with any current Ryzen 3000 compatible motherboard. Therefore any X570 or recently released B550 should work out of the box. However, I am sure there will be UEFI optimizations and releases coming soon to make them work better. The Up to 200MHz extra boost is something funny as that only applies to the 3800XT, but then again, I have always disliked the “up to” marketing.
Here we see where AMD shows the XT models slotting in with the existing processors. I will assume we will see a price update soon, or maybe AMD will simply EOL them as they run through inventory. Time will tell how these models work together.
Here we see the process technology claims from AMD along with their stated performance uplifts and performance leadership claims versus the competing parts. We have some chips in our testing charts, which we will be matching up against the new XT chips.
Here AMD explains the reasoning I told before about omission of stock cooling on the two higher-end XT chips. I agree with their logic, but cynical me thinks that these chips may not be able to be cooled to meet their target boost speeds adequately, and by discluding a cooler, it no longer becomes a potential fault on AMD if the product doesn’t boost on an inadequate cooling solution.
First up is the 3900XT, which is the 12-core variant with boost speeds of up to 4.7GHz now matching the top dog of the Ryzen lineup; the Ryzen 9 3950X.
Here we see the 3900XT versus the 3900X and 9900K, which are fair comparisons due to the price point. I once again don’t like the “up to” but being there are so many variables in a system, AMD keeps it safe in their messaging.
Next up is the 3800XT, which gains 200MHz on its boost speed to 4.7GHz as well versus the 4.5GHz boost of the 3800X.
Here we see the comparison between the 3800XT and the 3800X and 9700K from Intel. This once again is a reasonably even price comparison. The 3800XT shows an up to 2% -6% bump in most scenarios over the existing 3800X and a sizeable lead over the 9700K. This is not a surprise as the 9700K with anything threaded will suffer due to its non-hyperthreaded design.
The last XT CPU is the 3600XT, which is a six-core part with twelve threads. The boost speed received a nice bump to 4.5GHz versus the 3600X 4.4GHz.
Here the 3600X and XT CPUs are shown compared to the Intel 9600K, which once again the Ryzen parts show up to 30-60% leadership over the 9600K while the average X to XT gain being 2-3%.
AMD took the opportunity to tout the native X570 and B550 support for the XT model CPUs again. This is not saying that other boards won’t work, but they will need a compatible UEFI if it has support.
My testbench is strictly controlled with a fresh OS for any platform or component change. The system uses all the same components whenever possible to maintain comparable results between platforms. The ambient in the test lab is rigorously controlled at 22C +/- 1C. The CPUs are all cooled by an Alphacool Eisbaer 360mm cooler to ensure the CPUs all have the same thermal headroom. Thermal paste is also controlled with usage of Arctic cooling MX-4 with a similar spread application on every CPU.
All tests are run a minimum of three times, and any outliers are tossed, and another replacement test run will be completed to achieve our average results. The use of a TITAN RTX for the CPU testing is to ensure that the GPU is not the bottleneck for performance results, and will best represent the scaling across CPU and platforms.
- Motherboard: ASUS ROG Crosshair VIII Hero (Wi-Fi) (buy from Amazon)
- CPU: AMD Ryzen 9 3900XT
- CPU: AMD Ryzen 7 3800XT
- CPU: AMD Ryzen 5 3600XT
- Thermal Paste: Arctic Cooling MX-4 (buy from Amazon)
- Cooler: Alphacool Eisbaer LT 360mm (buy from Amazon)
- Memory: Corsair Dominator RGB 3600MHz 16GBx2 (buy from Amazon)
- Video Card: NVIDIA TITAN RTX (buy from Amazon)
- Storage – Boot Drive: Corsair MP600 (PCIe 4) 2TB (buy from Amazon)
- Network Switch: Buffalo Multi-Gigabit BS-MP2008 (buy from Amazon)
- Testbench: DimasTech Easy XL (buy from Amazon)
- Power Supply: Thermaltake 1200W (buy from Amazon)
- OS: Microsoft Windows 10 (buy from Amazon)
- Monitor: ASUS XG438 43″ 4K (buy from Amazon)
- Video Capture: Elgato 4K 60 Pro (buy from Amazon)
- Keyboard: Logitech G910 Orion Spectrum (buy from Amazon)
- Mouse: Corsair Logitech G502 (buy from Amazon)
WPrime, Cinebench, and AIDA64
WPrime, SuperPi, Cinebench, and AIDA64
WPrime is first up and being a multi-threaded benchmark. We know it will scale with any CPU we throw at it. You can manually set the number of workers or threads you want to allocate to the calculation, which we did the total thread count for each CPU to ensure we measure the maximum performance the CPU can offer.
WPrime being a threaded benchmark, we have both 32M and 1024M results. The XT CPUs do well here with the 3900XT topping the charts. Comparing price to performance even puts the new flagship 10900K in a bad place here. The 3900X and XT both have two more cores for a lower price, which makes this an uphill battle for Intel.
Cinebench is a long-standing render benchmark that has been heavily relied upon by both Intel and AMD to showcase their newest platforms during unveils. The benchmark has two tests, a single-core workload that will utilize one thread or 1T. There is also a multi-threaded test which uses all threads or nT of a tested CPU
The Cinebench R20 tests show the 3900XT topping the chart here, but more importantly, each of the Ryzen CPUs outperforms the competing Intel option at the same or nearest neighbor price point. The 9900K is still a viable CPU option, but in this test, it gets trounced by the 3800XT, which is one hundred dollars cheaper. The same applies to the 8700K versus the 3600XT.
Memory speed and latency falls where we would expect, with the 3600XT and 3800XT dropping to half write bandwidth as we see on all single CCD designs in Ryzen 3000. You will notice, as we have mentioned before, that the latency on Ryzen is always close to 20-30ns higher due to different architectural designs.
Encryption performance has been a strong point for AMD for a while, and it shows well here. As you can see, the top dog of the Intel mainstream stack, the 10900K, falls to even the 3600XT.
For FPU testing, we have 32-bit single-precision first. Here we see the 3900XT topping the charts again, but both the X and XT thoroughly beat the 10900K at a lower price. This is, of course, based on the extra cores the 3900X/XT have to play with. The 3600XT beats out the 10600K at a slightly lower price as well.
Now with the next level, we have the 64-bit double-precision test. Here once again, we see the same staggering of performance. The 3900X and XT are leading the pack. The 3800XT trounces the outgoing 9900K still, and the 3600XT leaves the 10600K in the dust.
Lastly, we have an 80-bit extended precision test, and the beating continues. This test pushed the 10900K even lower now being passed by the 3800XT, although it is within a reasonable enough margin to say they match. But that is an exciting result as the 3800XT is over 120 dollars cheaper than the 10900K, let alone the 9900K.
Handbrake, Blender, POV-Ray, CoronaRender, 7-Zip, and WebXPRT
Handbrake is up next, and we are transcoding a 4K MKV to MP4. Here we see more of the same with the 3900XT taking the lead, although the 10900K gains some ground here nipping at the 12-cores heels. The 3800XT still readily beat the 9900K, and the 3600XT outpaces the 10600K.
Blender once again shows similar scaling with the 3900X/XT beating the 10900K. The 3800XT smashes the 9900K, and the 3600XT cruises into a comfortable victory versus the 10600K.
POV-Ray or Persistence of View Raytracer allows users to create 3D graphics. We use it to test both 1T and nT performance of CPUs as it renders out as fixed workload.
Here we see the 1T result first, and the 10900K and 9900K finally get their day in the sun. The boosted single-core speeds mean the 3800XT and 3900XT are similarly staged on the chart. The 3600XT falls behind the 10600K by a slight amount.
The nT performance puts as back into more of the same as we see the 3900XT topping the chart and keeping the 10900K at bay. The 3800XT steadily beat the 9900K while the 3600XT leads the rest of the pack.
Corona Render is a photorealistic renderer that can be integrated into several popular 3D modeling applications. It can also be used as a standalone application, which is what we opted for to best control variables.
Corona Render shows the 3900XT now leading the 10900K, whereas the 3900X was matched with it. This gives AMD the lead once again. The 3800XT still steadily beat the 9900K while the 3600XT beats out the 10600K and the rest of the stack.
7-Zip is an open-source and free compression application. It works well with multi-threading and also can see gains from clock speed as well.
Here again, we see the 3900XT and 3900X beat the 10900K. Much to my surprise, though, the 10600K outpaces the 3600XT while the 9900K falls behind the 3600XT. Things did shuffle for this one.
WebXPRT is a browser-based test, and we like this test as this is one of the areas not many think to test. This also happens to be a real-world usage test that can be impacted by the mitigations which have recently rolled through and were patched.
Here we see the 10900K pull a decent win at five points over the XT CPUs. This is a 2% lead over the 3900XT and speaks to the efficiency of the 14nm process, and the clock speeds Intel has optimized.
Unigine and UL Benchmarks
Superposition from Unigine is a DX12 based benchmark. We test with the 720p LOW preset as this removes all but the most basic GPU loading, and all of the FPS performance comes from the CPUs ability to push frames to the GPU. This test is far more efficient and speed based rather than being highly threaded.
Here we see that while the XT chips are good, Superposition simply gives them a rough ride. The 3900XT falls behind even its own lower core count CPUs, while the 3800XT leading the AMD charge. The 3300X still holds a good position here, but Intel owns this one.
PCMark is a benchmark from UL and tests various workload types to represent typical workloads for a PC. Everything from video conferencing, image import, and editing, along with 3D rendering, are tested.
PCMark shows the 10900K leading the show wit the 3900XT behind it by enough of a margin to give the 10900K the win here. One note is the essentials tests that deal with application loading and other tasks that seemed to provide the XT CPUs a rough experience. As new UEFI/AGESA come out, we will have to see if this can be tweaked out, as this was not an excellent showing in this test.
3DMark Firestrike shows the 10900K leading the pack by a small margin. The 3900XT exhibits a substantial gain over the 3900X and the 3800XT hitches a ride with a close trailing position. The 3600XT steadily beat the 10600K. The CPU and combined tests show the 3900XT having a strong showing, but the graphics performance seems to be where the chips are struggling, and I am hoping some tuning can help correct this.
3DMark Time Spy
Time Spy is another 3DMark test variant, but this one is for DX12 based systems. This test can be quite stressful, and since its an entirely different load, you may be surprised to see how the results shuffle when compared to Firestrike.
Here in Time Spy, we see more of the same. The graphics performance shows similar to what we have seen form Ryzen 3000 before, and apparently, the pixel pipeline is better serviced by the higher clock speeds of the Intel parts.
In gaming tests, we are going to show 1080p results as it is still one of the most popular gaming resolutions, and it also means that the CPU will be a bottleneck for our TITAN RTX, and therefore we can see the performance scale to what the CPU can push through.
Shadow of the Tomb Raider
Here we see the 3600XT pull a 2 FPS bump from its clock speed gain, while the 3900X seemed to have lost a slight edge here. Overall the XT CPUs are bunched up mid-chart with Tomb Raider showing the Intel chips leading the pack still.
Wolfenstein seems to be another title that favors clock speed over threading or process technology. Here Intel wins again, but keep in mind that if gaming at a higher resolution, those results tighten up quite a bit.
Civilization is a quite popular game, and its performance we will test consists of standard graphical tests, which we will represent with average frame times for each resolution. We will also examine the AI turn time for each CPU to show how long each turn for AI will take based on your CPU performance and efficiency.
Civilization proves to be a more CPU bound game as we see the AI turn time benchmark pulls the XT series CPUs to the top of the chart.
Graphics performance shows similar results with the 3900XT and 3800XT pulling top spots with the 10900K in tow. The 3600XT comes in behind the 10900K and ahead of everyone else, including the 3900X.
For storage performance, we test the platform using a PCIe 4.0 Corsair MP600 2TB M.2 NVMe drive. It is tested in CrystaldiskMark 7.0.0 x64, and we average the results to ensure a good cross-section of expected performance. Do note that some platforms do not support the new PCIe gen 4, and therefore will cap out around 3400-3500MB/s.
First up is the Sequential read test, which every manufacturer touts as the “up to” speed, but in reality, you never see this in everyday usage, but I digress. Here we see the X570 platform topping 5,000MB/s or around there. The 3800XT takes the top spot formally held by the Ryzen 3100. All of the CPUs perform within what we consider average variance.
Next up is the write speed which we see it top out at 4,250-4,260 range. Once again, the Ryzen 3000 parts top out the drive with eth 3600XT and 3800XT, taking the top spot this time.
Next up is your random read, which is likely something you will experience more often in everyday computing. These are tiny files at a low queue depth. This is usually an area where OPTANE or a 3DXPoint storage solution rules, but the PCIe 4 drive does well here.
Here we see the 3600XT and 3800XT retake the top spot. The 3900XT are right on the heels of the 3100 and 3300X.
The MP600 tends to do much better here with random writes. This time we see a scaling where even the PCIe 4 platform separates. The 3900XT falls off the pace a bit here the same as we have seen before, and I can only assume its due to the CPU not boosting as frequently or not triggering to push the limits of the storage.
Clocks, Overclocking, Thermals, and Power Consumption
Here we will look at the physical and functional performance metrics for the new Ryzen 3000XT CPUs.
Out of the Box Clocks
First up is a frequency plot for the Ryzen 5 3600XT, and we see that during idle, the CPU drops to 3700MHz area. When loaded, we see the CPU jump to around 4.5GHz and hold there beside the clocks jumping from idle to load as they are needed.
The 3800XT is up next, and we see the idle speed of around 3.8GHz. The loaded speed seems to jump around 4.6GHz while I did see small blips up to the 4.7GHz turbo, it was quick and infrequent. I am not sure it was at that speed long enough to register on our logging, which has a 1000ms polling.
Lastly, will be the frequency plot for the 3900XT, which we see has an idle of around 3.7GHz. Putting the CPU under load and we see the clock speed settle into the 4.3GHz range with spikes to the 4.7GHz area readily visible. We recognize that cores two and three seem to be the preferred cores for higher binned speed as they are the ones showing higher spikes than the rest, which almost make a flat line just shy of 4.3GHz.
For power consumption tests, we use a wall meter to test the full system draw. The reason for this is it will represent what the entire system pulls versus our meter, which shows power draw on each PSU cable. The reason for this is that measuring the power draw from the EPS cable, for example, does not take into account VRM losses and, therefore, can show a much higher power draw for the CPU or other device due to an inefficient VRM design or loading range.
Idle power with the full system and TITAN RTX discrete GPU in place, we see the XT CPUs topping the chart at idle with each pulling 72 or 73W, respectively. I assume once again, this must be a tuning or teething issue with these new chips that need to be ironed out.
Loading up the CPU with a full FPU load, and we see the XT models once again showing a higher load power draw than its predecessors. The 3900XT gains six watts under load, while the 3600XT gains seventeen watts loaded versus its 3600X predecessor. I know AMD says more performance at the same power, and with the AGESA 220.127.116.11, which ASUS provided to test the CPUs, I can only assume there is some more tuning needed post-release.
Testing synthetic GPU stress only is an excellent way to show the overhead the CPU adds, as the GPU does not pull more power after it reaches steady state. Here we see more of the same with the XT CPUs pulling the highest power draw again.
Idle temps show the XT chips topping the charts by a few degrees, which now corroborates the higher power draw numbers we observed at idle and load. It looks like the UEFI update may be juicing these CPUs a, but more than AMD has intended. This is not a deal-breaker, but something to have in mind as it will be a heavier load on your cooler until this is resolved.
Loading each CPU up with a full burn-in FPU load, and we see the gaps start to widen. We waited for each platform and cooler to reach full saturation/steady-state before making any measurements. Here we see more of the same with the 3900XT and 3800X topping the chart, which makes sense as they have higher power draw and higher clocks. The 3600XT gains almost seven degrees of loaded thermals, and I can only assume this is due to its consuming more power, which in turn means more waste heat.
Overclocking Ryzen has been an exercise in futility at most junctures. Typically our analysis ends with something to the tune of just use PBO as it will do the job, and you won’t lose the higher single-core boost. But we always have to try.
I am delighted to report that overclocking with Ryzen is viable on the 3600XT as we were able to achieve a 4.6GHz all-core overclock, which gave us not only an nT gain but even a 1T boost. This is indeed one of the first Ryzen chips where we have been able to overclock and not lose performance in one of the two.
The overclock with 1.35VCore set in the UEFI, and no LLC gives an actual VCore of 1.293V. This means we are now getting more performance with less power draw.
Next up is the Ryzen 7 3800XT, which we were also able to get up to 4.6GHz at 1.35VCore and LLC to level 2 set in UEFI. This gave us an actual VCore of 1.301 loaded. This gave us a nice nT boost while we lost one point in CBR15 1T, and I would consider that close enough to count as equal, or you would likely not notice. I will note that temps at this speed and voltage with AVX loading to test stability reaches 82C average with spikes to 86C.
Lastly, we test the 3900XT, and our silicon lottery luck did not pan out, as we walled at 4.5GHz this time. We achieved this with 1.368VCore set in UEFI with LLC 4 gives us a 1.344VCore loaded. The boost to nT performance was good, but we took a pretty big hit to 1T performance dropping us down to the 3900X or 10600K level.
While pushing the overclock on the 3900XT, I had to capture a thermal image to show what pulling 200W to the CPU does with a well built VRM. As you can see, the chunky heat sinks topped out at 43C and 41C while pushing 200W to the CPU with an AVX load to test the overclock stability. This heat spread you see through the heat sinks and surrounding PCB is after over an hour of constant AVX loading.
I would say our test board is more than well equipped to handle heavy CPU loads.
The Ryzen 3000XT chips are a subtle enhancement to the existing Ryzen 3000 product stack. I do think that if you have a current Ryzen 3000 model these replace, it would not be worth it to upgrade, but if you are on an older ryzen chip or even a more former Intel platform, they have proven to be quite capable.
What we like
Overclocking: The new optimizations to the 7nm process seems to have paid dividends with clock speed capability.
PCIe 4.0: We know that PCIe 4.0 is just now starting to see capable parts coming to market, but the expansion of the bus to allow higher throughput means that not only SSD’s will benefit, but other controller or HBAs will come online allowing less PCIe lane usage for the same bandwidth.
Value: The XT series CPUs simply smash the competing Intel products at the same price point. The gaming performance is still an Intel favoring workload, but if you are running over 1080p, this deviation can be lessened substantially.
What do we think could be better?
Early teething/tuning: These chips like most launches have 11th-hour updates, and that can sometimes net great results, or it can backfire as we observed with the power testing.
Gaming performance: A lot of PC games still favor clock frequency to push pixels and Intel’s aging but matured 14nm process still gives a few swift kicks to Ryzen in gaming workloads at lower resolutions. As Ryzen develops, and with Ryzen 4000 looming with rumored higher clock speeds, this should have Intel on notice.
The new Ryzen 3000XT CPUs are great CPUs that perform computational tasks at a value that is tough to beat. However, if you are gaming at lower resolutions, you may find a more suitable solution from the blue camp. But the gap is closing enough that I think any gamer who does not consider Ryzen 3000, let alone the new XT chips, are looking past some excellent value prospects. I believe with the teething issues, and a UEFI update or two, these chips could fall into a place that will be an excellent fit for most any user.
That rounds out the new Ryzen 3000XT CPUs and our testing and thoughts on them. Stay tuned to TweakTown as we cover the latest CPUs, chipsets, and tech hardware.