How good is the new Threadripper 3990X?

The AMD Ryzen Threadripper 3990X, released today, performed excellent in the reviews. With 64 cores and 128 threads, it was completely reasonable to assume this monster of a processor would only be suited for some specialized applications. And that’s a little bit true: It’s not the strongest gaming processor, or the strongest in single threaded workloads. But with some exceptions it is still strong in those - which is surprising.

In games, which tend to focus on single threaded performance, it sits a bit below the Threadripper 3970X, but still above the AMD Ryzen 5 3600:

Gaming Performance of the Threadripper 3990X: 9.27/10
See the individual benchmark results here, they also show some of the games in which the processor performs badly.

In application performance it is the strongest processor in the benchmark, with a vast distance:

Application Performance of the Threadripper 3990X: 10/10
The details are here.

Of course, no one will or should buy the Threadripper 3990X for gaming, it costs $4000. That part of its performance just shows that its cores are not that weak - this is again no FX processor, which makes it more viable for software that is not that completely parallelizable. Still, in the end the 3990X is really strong only when it can use those 128 threads. They give it excellent results in modern software benchmarks. It would be a great choice for cpu heavy applications, like professional render workstations.

The Radeon RX 5600 XT: Launch issues and benchmark performance

This AMD launch did not work that well. So here is what happened: AMD released the RX 5600 XT with a different VBIOS than originally planned. The new VBIOS raises core and memory clock, resulting in better performance. But: A last minute change is highly problematic for partner card vendors and thus consumers. The raised memory clock is hard to test and might cause issues, prompting some vendors to stick to the lower memory clock (6000 MHz instead of 7000 MHz). And that hard to notice difference is highly confusing for customers, who now can’t be sure the RX 5600 XT they buy reaches the performance shown in reviews, because those were mostly done with cards that reach the 7000 MHz.

But first, how did the card perform? In the benchmarks collected so far we saw it perform very well for its price, almost reaching the RX 5700 and beating the RTX 2060.

/blog/upload/rx5600xtchart.png

See this comparison for details.

That makes the RX 5600 XT an attractive option, especially if the price between it and the RX 5700 gets bigger over time.

How bad would it be to get a model with the slower ram? We can look at this Computerbase review of multiple partner cards, some with the slower, some with the faster memory clock. The difference was ~4%. Getting the slower card would not be optimal, but also not a big issue.

But the best 5600 XT model to get is one with the faster memory clock: The Sapphire Pulse RX 5600 XT. It comes with the new VBIOS, reaches the faster memory clock and stays cool and quiet. Best of all: It is one of the cheaper models. So at least for a while there is no reason to get something worse.

Despite the launch issues, the 5600 XT cards are now in our system and they will be used for PC build recommendations.

The almost christmas update

Over the last weeks the recommender got a couple updates I will summarize here. Most important were a couple hardware releases not properly covered in this blog. But they are in the system!

We got finally a serious competitor for the RX 570: Nvidia’s GeForce GTX 1650 Super was released at a great price of around $160 and showed very good performance. So good that it also beat the RX 580 and reached the RX 590.

AMD Radeon RX 5700 XT with 4GB or 8GB memory released shortly after. They are also graphics cards meant to replace the RX 500 cards. They use less energy and are faster than the RX 580, not necessarily faster than the RX 590. They have a hard time against the GTX 1650 Super, at least with current prices, since the 8GB version is above $200 and the 4GB version clearly slower in some games.

On the processor side of things there was the Ryzen 9 3950X. Astonishingly that processor matched or beat the older Threadripper processors, and that on socket AM4. And that is in application workloads. The real surprise here is that the 16 core processor is also exceedingly strong in games.

Good thing that Threadripper 3 got released as well and included a big performance boost, albeit with higher prices and needing new and more expensive motherboards. The 3970X has 32 cores, the 3960X 24, both of them got many architecture improvements.

At the low end there was the AMD Athlon 3000G. And we finally got enough proper benchmarks to rank the Athlon 240GE, 220GE and 200GE. Those 35W dual core processors with 4 threads do not compete with Ryzen 3 or Intel’s current i3 line, but they are valid options for low powered family PCs and could work well in a custom NAS.

The site itself got some bugfixes. For example, the cpu cooler sometimes got stuck when trying to use the cheaper/better buttons at its side. There is now a new mechanism to ignore stale prices, which so far seems to work well - it reduces the chance that a price shown here is not valid anymore when clicking through to the shop.

Have fun building and a Merry Christmas!

onli,

GTX 1660 Super - Benchmark Review

Nvidia released a new card in the budget segment. It’s another alternative to the GTX 1660 and GTX 1660 Ti, with attractive models like:

  1. EVGA GeForce GTX 1660 Super SC Ultra
  2. MSI GeForce GTX 1660 Super Gaming X

How did it fare in our meta benchmark? Not so good, have a look:

/blog/upload/gtx1660super.png

Or go directly to the always current comparison page.

It’s just nothing spectacular. The 1660 Super is a little bit slower than the 1660 Ti. It’s also supposed to be cheaper, depending on the model you pick. If sufficiently cheaper it’s a nice alternative to the Ti model. Apart from that it changes nothing on the gpu market.

The recommender will use the new 1660 Super cards from now on when appropriate, price tracking is activated.

Review of the Thermal Grizzly Carbonaut: It works, but it's not better

Thermal Grizzly Carbonaut is a carbon thermal pad intended to replace thermal paste between processor or graphics card and its heatsink.

Thermal pads sound nice in theory and not too long ago there was some hype around IC Graphite thermal pads. There is no mess, no danger of dropping some thermal paste into the cpu socket or some other sensitive area. And the Carbonaut pad promises to not get old, unlike thermal paste, which you have to replace after a while.

/blog/upload/dispalcedpad.jpg

The Carbonaut pads can be bought in multiple sizes, e.g. 31×25 mm fitting for a RTX 2080 Ti. But since the RX 580 with its ~17×13 mm gpu die is not a target size yet I had to cut it to its proper size. I bought the bigger TR4 edition to be able to test the pad in two scenarios: First on the processor, then on the graphics card. The processor is an Intel Core i5-5675C, a perfect scenario for the pad: The i5 runs cool and is not overclocked, the used Corsair H90 AIO cooler quite powerful. If the Carbonaut pad does not work here it would be basically useless. The graphics card is a Radeon RX 580 with the Rajintek Morpheus II cooler reviewed earlier, it’s a more extreme scenario in terms of heat to handle and complicatedness of installation. Both used a cheap thermal paste in the tests before applying the pad – because that’s the real competition here, not more expensive high end thermal paste targeting overclockers.

/blog/upload/carbonautbox.jpg

The Carbonaut pad arrived in a plastic bag like the ones Thermal Grizzly uses to send out their thermal paste. A small manual explains the installation (have it in the correct size, place it between thing to cool and heatsink, take care to not cause a short, it’s reuseable), the pad itself is inside a second cover. Mine was not installed properly in there, but at least that picture (the first used in this review) shows well how slim it actually is. At that point I was skeptical: Can something so flimsy actually be installed without breaking it?

The Processor test

Cutting the pad proved indeed to be difficult. It is easy to cut the very lightweight canvas-like surface, that’s not the issue. But sizing it properly, putting it on the processor without letting it touch remains of thermal paste and then not having a way to set size markers turned out to be a complicated process. The result of my throwback to Kindergarden was fine though, without wasting too much material I got the size about right, the whole integrated heatspreader (ihs) got covered.

/blog/upload/coveredcpu.jpg

Before the installation, with the cheap thermal paste, the processor did already run fine. In a 10 minute stress test (under Linux with s-tui) it showed this heat profile:

/blog/upload/cpu_before_stresstest.png

59°C, a stable 3.5 GHz, which is the multi core max load of that processor. At the end you can see ~30 seconds of idle time, when the temperature moved down very fast.

With the Carbonaut pad installed not much changed:

/blog/upload/cpu_after_stresstest.png

57°C, again a stable 3.5 GHz and the temperature moved equally fast up and down.

That’s a success! To match the performance of (albeit cheap) thermal paste was not at all excpected for the Carbonaut pad, because historically, those pads always were so much worse. But in this scenario this new kind of thermal pad worked perfectly fine. It was harder to install than thermal paste, but had I bought it precut in the size of this processor that first step would have been much easier.

The Graphics card test

The RX 580 with the Morpheus aftermarket cooler was a different story and it started with the installation. It was just much more difficult. On top of the annoying first step of having to cut the very delicate fabric, the screws to connect the heatsink to the gpu go into the back of the gpu. So you have to place the thermal pad, place the heatsink on top of it, move it 180°, maybe add the second backplate and during all of that not move the heatsink/gpu at all – because if you do, you have no idea whether the thermal pad still covers the gpu die, and in the worst case even this might happen:

/blog/upload/brokenpad.jpg

Because of the force applied during the installation the pad ripped apart, which I noticed since I had to repeat the installation because a smaller heatsink glued to the VRM fell off.

That’s really not nice, because you can’t be sure that the pad is placed properly. Unlike in the processor scenario, where good performance confirmed that. Because performance just was not great either.

The cheap thermal paste was already not perfect. During a benchmark of Witcher 3 the gpu reached a high temperature very fast, though in this short test it did not throttle:

/blog/upload/carbonaut_witcher3_old.png

With the pad temps at first looked great:

/blog/upload/witcher3_carbonaut.png

But after some time throttling was observable and the temperature did not stay that good:

/blog/upload/witcher3_carbonaut_longterm.png

Now, sure: Air flow is very important here and it’s possible that’s the main issue. But it’s just not a success. In this scenario, the Carbonaut pad primarily proved too hard to install and only secondary did not reach the performance I was hoping for; I’ll confirm at a later date whether that expectation was reasonable and edit this review accordingly.

The idea of never having to replace the thermal interface again is very appealing especially for the graphics card, where high usage and high temps are not kind to thermal paste longevity. Instead of getting that I now can’t even be sure whether the thermal pad installed was not damaged during the installation. Frankly, good thermal paste would be way better here. Performance should be better, some pastes would handle the heat fine for a long time and the installation is a lot easier.

Conclusion

Thermal Grizzly Carbonaut works and it can be a valid option for PC builders. But its usefulness really depends on the scenario. I’d only recommend it if:

  1. it comes in the size of the processor ihs/gpu die,
  2. good thermal paste is not already applied,
  3. the heatsink can be installed without having to turn the gpu around,
  4. what you want to use it on does not produce too much heat.

And those are a lot of requirements. Too many. The default sizes, made for two types of gpus (RTX 2080, RTX 2080 Ti), modern Intel and AMD Threadripper processors might just not cover what you want to cool with it and I really recommend against cutting it to the size you need manually. It’s just not worth it if you could use thermal paste instead. Which you’d also get if you really want to overclock or if the installation process of your cooler is complicated, like with basically all gpu air coolers.

But if you check all the boxes above the Carbonaut gives you a heat interface that does indeed work and that does not dry out. That’s nice and might be worth the price for you.