Review and test AMD Ryzen 5 5600G and Ryzen 7 5700G – i2HARD

A full test of the processor part and integrated graphics in stock and overclocking.

Review and test AMD Ryzen 5 5600G and Ryzen 7 5700G

Hello. Cool thing these your processors with integrated graphics. Is not it? More recently, the Ryzen 5600G and 5700G have appeared on sale. You look at the price – almost the same as the 5600X and 5800X, respectively. You look at the cores – the same ZEN 3. And they say about the integrated graphics that it is at the level of the GTX 1050, which, by the way, now costs 10 thousand on Avito! Solid pluses, but is everything as great as it looks? Let's figure it out.

Let's start with a comparison of the graphics subsystem. There are no changes here on paper compared to the previous generation. In the case of the 5600G, we have the same Vega 7 as in the 4650G with a frequency of 1900 MHz. Ryzen 7 similarly contains Vega 8, clocked 100MHz higher.

All the changes were made specifically to the processor part, but graphics tests are indispensable, so let's designate a test bench and rivals. Since the GTX 1050 is heard as an analogue in terms of performance, then we can compare it with it, and with the help of the Ryzen 5 4650G we will check whether there really are no changes.

Test stand:

  • Graphics Card #1: MSI GTX GeForce 1050 2G OC
  • Video Card #2: Palit GeForce RTX 3080 Ti GameRock OC
  • Processor #1: AMD Ryzen 5 3600
  • Processor #2: AMD Ryzen 5 4650G
  • Processor #3: AMD Ryzen 5 5600G
  • Processor #4: AMD Ryzen 7 5700G
  • Processor #5: AMD Ryzen 5 5600X
  • Processor #2: AMD Ryzen 7 5800X
  • Motherboard #1: ASUS ROG Strix X570-E Gaming
  • Cooling system: Gamer Storm Castle 360RGB v2
  • RAM: OLOy WarHawk RGB 16GB (2 x 8GB) ND4U0836163BCWDX
  • Storage: Silicon Power Ace A58 1TB
  • Power supply: Deepcool DQ850-M-V2L
  • Body: Open Stand

Stock iGPU benchmarks

Synthetic tests and benchmarks

Let's start with synthetics, finish with games.

Aida64, GPGPU Benchmark: both on paper and in practice, Vega 7 has identical performance in different processors. The GTX 1050 lags behind the built-in in one half of the tests, and outperforms it in the other.

Adobe Premiere Pro, our usual segment with tests lasting a little more than 5 minutes was taken as a project, except that color correction was added. And 1050 is unable to render even such a small piece of the project. When its video memory is full, an error occurs, which has been found on most Nvidia video cards for many years.

The inserts differ slightly from each other. However, there should be no difference at all between the first two rivals.

Davinci Resolve 5600G is also faster than 4650G, and the difference between them is stronger than between vega 7 and vega 8 in the modern version. And the latter, by the way, has 14% more cores and their frequency is 5% higher. The GTX 1050 runs like clockwork here and shows an almost twofold superiority over the vegs.

Assassins Creed Valhalla. Graphics preset low, Full HD, resolution modifier 70%, that is, slightly lower than 768p. Recording was carried out by a capture card from a third-party computer without affecting performance. A 2GB GTX 1050 is not enough for this game. This is signaled to us in the graphics settings menu, but we do not observe any strong penalty. However, the frametime on vegas is smoother, although they are worse in terms of average FPS. The 4650G again falls behind the Ryzen with similar integrated graphics for some reason.

Cyberpunk 2077, also at minimum settings, but the resolution modifier had to be lowered more, to 50%. The final resolution is 540p. Is it possible to play with it? Maybe.Dmitryaga ran on missions for 2 hours, however, with overclocking. You can increase the resolution by reducing the FPS to the console 30, but this is clearly an amateur. There is no tangible difference between participants, but from 5700G I would like to see a greater advantage.

Red Dead Redemption 2, again the minimum and ⅔ ⅔ from 1080p. Here it is an analogue of 720p. Above the permit modifier cannot be put for 1050 without interfering in the config file, since the game signals a lack of video memory. GT 1030 made it clear that this should not be done. As a result, 1050 breathes full chest and surpasses everyone in the average fps. VEGA 7 in 4650G for unknown reasons for the same locations sags strongly, and it looks like an incomplete load of a built -in video department, but 40 FPS can prepare the processor part 4650G, so that such strange behavior remains a mystery to us.

Resident Evil Village, Settings for at least 1080p, FSR in quality mode. We will say right away that the performance of the GTX 1050 is conditional here. Due to the lack of video memory, the FPS sags greatly and it varies greatly from the passage to the aisle. Moreover, before and after the fps Katsen, it was higher than at 3. We ran through the locations at the beginning of the game and it turned out that this was not everywhere like that. In some places, it is comfortable at 60 FPS, but it is worth it to see the wrong side of the veil – already 30. Surely, more complex locations will meet.

World of Tanks, 1080p, high -high preset. Tanks we will immediately denote the whole category of games. GTA V, DOTA 2, CS: GO – In all of them, GT 1030 fought on equal terms with 4650G accelerated. Obviously, in them 1050 will lead with a large margin. Therefore, there can be no talk of any analogues. Yes, somewhere the lack of video memory reduces the FPS to the level of veg, but somewhere it is much higher, and somewhere below. Therefore, you can not name built -in vehi and 1050 analogues.

By tests in the stock, it becomes clear that among themselves, Vehi differ by less than 10%. In general, 4650G and 5600G should not have been different, but perhaps the matter of latency of memory. If you look at the results in Aida, then ceterisis equal to its delay is 10 ns higher.

The performance of the built -in graphics 5700G is slightly disappointed. With the previously mentioned advantage in the form of an additional block of calculations and a slightly larger frequency, it has a very small advantage. The encoder, it seems, also has not suffered changes. AV1 is not supported, as well as decoding 8k videos on YouTube.

IGPU tests in acceleration

Benchmarks and synthetics

Well, maybe slow memory restrains the potential of built -in graphics? We proceed to acceleration.

For 1050, a regular Offeset of the nucleus of 135 MHz was used, and the effective memory frequency allowed itself to increase by 1600 MHz.

Our 5600- and 5700G samples are completely identical in acceleration. 4.5 GHz on nuclei at 1.3V, 4533 MHz from memory with the first timing 16, even the built -in video in all cases conquered no more than 2300 MHz. Moreover, the same acceleration of the building has 4650G. The new generation was not pleased with the great accelerated potential. Unless the old generation had worse results in nuclei and memory.The frequency of the processor nuclei at the same voltage is 4.25 GHz, and the memory is not more than 4333 MHz.

Yes, if someone missed, then the APU have two memory controllers that allow you to work with a divider 1 to 1 up to 4600 MHz in the case of 4000 series, and in 5000 this limit is even higher. So why did we stop at 4333 MHz, if the memory and the motherboard are definitely capable of more? The point is the voltage on the SOC. For hybrid processors, its relatively safe limit is higher than that of ordinary ones, but there are no clear boundaries. In the case of 4000 razins, we decided not to put above 1.2V for ourselves; For the 5000 series, by analogy with conventional processors, where the limit moved in a large way – 1.25V. Also, the acceleration of the graphic nucleus directly depends on this voltage, so the voltage on SOC, not GFX, is demonstrated in monitoring.

Again we turn to AIDA64. Only the first three measurements regarding memory demonstrate a slightly larger acceleration in the case of 5600G. For other disciplines of differences from 4650G, no again is observed.

The Render of the Scene in Premiere Pro accelerated by 28-30%. Naturally, acceleration 1050 did not help get rid of mistakes. They say that switching to another video format in the render settings menu can help.

In Davinci, acceleration gave approximately the same results for Veg: +30-31%. The increase in 1050 pleased – 17%. In the game plan, the dispersal of discrete video cards are no longer impressive for a long time, even 10% can not always be squeezed with standard cooling, and here is 17.

Tests in games

A pleasant picture is observed in the assassin. Namely: the treasured 60 FPS obtained due to acceleration, which gave more than 30% of additional personnel. It was on this basis that the settings of graphics were selected. Myld, but smooth.

In Cyberpunk also. It is difficult to see the enemy, but if you saw, there is no problem to aim. Freimtime is quite even, but the staters are still slipping on veg periodically. By the way, we did not hurt the setting up memory for the built -in graphics. By default, 500 MB is provided, and remember – almost a year ago, during the 4650G test, small staters with constant frequency constantly occurred with such a volume? Maybe they corrected the new with the driver, maybe with Bios, but now they are no longer.

In RDR2, you can also get 60 FPS by spending time overclocking and donating a resolution. Now imagine that all these games have Fidelity Super Resolution, which allows to significantly improve the picture. The craftsmen have already learned to screw him to most games, and not only to them. But since they can, why are there no developers?

There is FSR resident, and thanks to him, the clarity of the image does not particularly suffer, and FPS is pleasant. And we do not need more. With acceleration in all games, obviously, the bowl of the scales shifted. 1050 lost its weight, because the increase from its settings simply fades compared to it for APU. On average, the seventh veges increased their performance in the games. For Vega 8, the increase was a little less, because In the stock, she had a high -speed video code higher.

In tanks and similar games, 1050 did not lose its leadership. No accelerations here will help. That is, we will fix it again. For modern AAA buildings are better suited. Their microarchitecture manifests itself well in modern APIs, and the lack of a fine for a lack of video memory allows you to increase one of the most noticeable settings of graphics – the quality of the textures. The main thing is that the RAM is enough. However, if your task is to get more FPS in stale online toys, then 1050 is much better for them.

You can also add GeForce Experience to its advantages. In addition to many filters, in it you can record gameplay in excellent quality with insignificant loss of performance.

For Buildings, Radeon Relive is not available, although they may well encode the video. And the same OBS during the game can record a gameplay with a passage of personnel, plus the performance falls greatly. It is not a fact that it will come out to configure it properly, while in a company software I pressed the button and the recording went.

We study the processor part

Between themselves, we repeat, Vega 7 and Vega 8 are not particularly different. The overpayment of about 8 thousand rubles for 5700G does not look justified, but we still have a processor part that has undergone much more changes. So what is new in her?

Slide from the presentation does not quite reflect reality. For example, the maximum frequency is 50 MHz above the declared, and 65 watts is clearly not the consumption that you will encounter, because the PBO is on by default.

In our case, when using a discrete video card, both processors consumed about 80 watts with peaks up to 84 in a stress test.

With a simultaneous stress test of processor and graphic nuclei, up to 90 watts will be released.

And the boxing cooler, surprisingly, withstands. In Russia now there is no OEM version on sale, so there will be no superfluous to find out the effectiveness of the cooler. Complete coolers for 5600- and 5700G differ. Ryzen 7 has the same as 5600X, and 5600G boxing cooler is better. The gap between the fan and the radiator is much smaller, half of the air flow will not run away from the radiator. Plus, due to this, it is lower, and pre -applied thermal paste covers almost the entire processor cover. However, the speed of rotation of the fan is 500 revolutions per minute lower.

Both fans are heard at the length of the outstretched arm, but they cannot be called noisy. Tolerant.

When using a discrete video card, the stock 5700G is heated in a stress test to 85C, which is unpleasant, in games the temperature will be 10'C below. This is clearly better than the situation with the i5-11400, but given which AMD coolers can do it could be better.

As in ordinary processors, 5000 rulers, in the hybrid nuclei, switched to the new Zen microarchitecture 3. However, in addition to the unification of the L3 cache, we received its doubling relative to the past generation, that is, now we do not have a four -fold lag in terms of third -level cache, but a double. However, from other restrictions of the cornet of the eyes of the PCI Express 3.0.

Processor test in stock


Synthetics: In the test test and cache, Aida is striking not only a high latent memory of 4650G relative to 5000 series of hybrid processors, but also a lower latency of 5600x.

In Cinebench R23, the smaller cache does not manifest itself. The 5000 series of razenes has among themselves differences associated with the number of nuclei and their frequency.

In Geekbench 5, the 5600x gaps increased slightly. The lag of 4650G from his heir was reduced by a couple of percent and now 5600G is slightly less than 15%faster.

In CPU-Z, the situation is similar to Sinebench. Two additional nuclei and a little large frequency allow you to score 37% more points to an eight -core hybrid compared to six -core line.

Complete the results of synthetic tests of the process of their passage. In the blender, among stock processors, the hottest is 5600G, since it has a higher power limit compared to 5600x, and 5700G with the same heat packet has a large heat removal area. Add in a comparison of 5800x, and its 140 watts make themselves felt – the temperature of the nuclei is higher by 22 degrees. However, it should be borne in mind that with such a TDP 5700G it drops the frequencies stronger.

Let us return to this comparison in the section with acceleration in order to find out about another interesting fact. And now there will be tests in games.

Cyberpank, Preset of graphics – tracing of ultra rays, high density of the crowd and DLSS – performance. This game prefers to ignore the presence of SMT in eight -core razen, while on six -core it is actively used and gives a decent increase. As a result, 5700G is slightly weaker than 5600G. The latter surpasses 4650G by 26%, but 10 FPS, that is, 13% of additional frames are not enough before 5600x. Also, as a bonus, we tested Ryzen 5 3600 so that you have an idea of ​​what hybrids closer to: to the past or current generation of ordinary processors.

Shadow of Lara, the highest preset, 600p. The difference in game performance for the most part depends on the volume of the L3 cache. And, as we see, in Lara an additional 16 MB gave a good increase in 5600X, thanks to which even 5700G and close to it should not be. Actually, because at the battle of top -end processors, Intel loses in this benchmark. The cache is not enough. 5600G is almost 30% faster than its predecessor, but 5600X prepares personnel 36% more than 3600.

Watch Dogs Legion, Ultra Preset, DLSS – performance. It can be assumed that for dogs there are few L3 cache, or 16 MB. Of course, there is an increase from it, but not as impressive as in Lara. I also recall a large latency of memory in Aida at the same XMP in hybrids. This is probably also contributing. 4 additional stream 5700G give their increase, but it is small. It will be not enough. And even in such a game 5600G is slightly closer to the performance of 3600 than 5600X.

Starcraft II, maximum graphics settings. Here, too, 5,000 razins at the exit immediately took a leading position due to the most part of the unification of the L3 cache and the pumped block of predictions. 2 times smaller in the APU of the same ruler has a serious effect. Their performance is below by a quarter.This is the level of Raizen 3600, if it had the frequency of nuclei is the same.

Total War Saga: Troy. Ultra Preset graphics, the size of the detachments and the details of the grass are increased to an extreme level. Resolution modifier – 50%. Without appropriate acceleration of the memory of the nucleus 5700G, they cannot show themselves, which is why it is only a little faster than 5600x. 4650G is also slightly behind. A strong restriction on the part of the memory subsystem is felt.

CS: GO, minimum settings, tournament replaces. 5600X destroyed everyone. Who would have thought that some additional 16 MB-built into the memory processor can give 100 FPS out of the blue, which there are as many as 30% of additional frames. The same third 5600G is faster than 4650G. It is interesting how much faster the new razins with an enlarged cache will be.

On average, disappointing results come out in the ward. In addition to all the improvements that occurred in the usual Zen 3 processes, the hybrid ones also doubled the L3 cache, reducing the lag in terms of its volume from an analogues deprived of the four -fold to two -time. However, 5600X on average 31% faster than 3600, while the transition to the new generation brought only 26% of additional personnel from the APU. Therefore, as 4650G was closer to 2600, than to 3600, and with the 5600G Raizen, a similar situation came out.

Acceleration of memory and processors

The situation can save memory acceleration, because 5600X, like 3600 are often limited by a frequency of 3800 MHz, while at 5600G we reached 4533 MHz. The timings of processors without building are identical. Acceleration of nuclei differs. At the same 1.3v 5600x, 4650 MHz is stable at a frequency, and 3600 came out 300 MHz less.

And if the identical XMP Latenity of 5600X was the smallest, now it is in third place, and in terms of reading, recording and copying – in the latter. And yes, in hybrid processors with a monolithic crystal, the recording speed is not cut 2 times. Until now, many hayzen is for this flaw, but something has not particularly helped the Achushki.


In Cinebench, the balance of power has not changed. Acceleration of memory does not affect here, and the acceleration of the nuclei cannot be called. It is rather a simple fixation or a certain Andwolt.

In Higbench 5600x faster the namesakes with index G is no longer at 7.5, but by 2.5%. 5700G, in turn, escaped forward.

In the CPU-Z 4650G, I lost 4 points compared to the stock, since its frequency was recorded by 4250 MHz, while with PBO it could reach a value of 50 MHz higher, but in a multi-thread we received 80 points more.

We again turn to the blender to now compare the processors at the identical frequency, stress and without Power limits. Here is a surprise: a six600x hot and glider 5700G. And compared with 5800X nuclei, 5700G inside eight ice cubes. Between them, almost one and a half -fold difference in energy consumption. A good surprise.

We go to Cyberpank again and observe a pleasant picture. The 13%advantage of 5600x over 5600G decreased to 4%. Processors in the center are flush, and their gap from 4650G increased from 25 to 30%.

In Lara, the Big 5600x cache still dominates, but not with such force. On the other hand, you noticed that this cache gives a huge increase only in those games where there are so many FPS? Starcraft does not count, there the entire load hangs on two cores.Contra, tanks, DOOM, lara and many other games are already played on such processors with excessive FPS, and in more demanding games like cyberpunk or watch dogs, where you have to choose between ray tracing and normal FPS, the advantage is not so great.

In dogs, the 5700G was finally able to outperform the opponent on the right, while the 5600G is only 5% behind. Moreover, according to the statistics of rare and very rare events, it cannot be said that the difference is radically different. Plus or minus the same thing from game to game, so we don’t pay much attention to it.

An interesting situation in Starcraft. While the 5600 and 5700G gained 23% and 19% overclocking gains, respectively, the 5600X increased its average FPS by just 5 units. This is only 6%. Yes, APUs have much better memory overclocking, but this situation is quite revealing, because it demonstrates how less often the 5600X has cache misses in this game. For example, in cyberpunk and watchdogs, the increase from overclocking on this processor was 22-23% in average FPS, while in lara it was already 15%, and here it’s only 6.

Remember the stock 5700G was only marginally faster than the 5600G and 5600X? It was worth overclocking the memory, and its advantage over the six-core counterpart increased from 12 to 25%. And the 5600X, due to less memory overclocking, almost caught up with the 5600G, but it was more surprising that the 4650G quietly bypassed the ryzen 5 3600.

Guess how much overclocking the 5600X got in CS:GO. 20, 10 or 5%? Less – 2.8! Given the measurement error, which varies from launch to launch of the game, this can be entirely the merit of a fixed frequency that does not fall below 4650 MHz. Not to say that APU's shine with growth. The 5600- and 5700G got 8-9% extra frames, while the 4650G got 13%. Considering that the frequency of the cores did not change much with overclocking, it is noticeable how much higher the gain from memory overclocking becomes for processors with a smaller L3 cache.

And what do we have on average now? The 5600G is 13% faster than the ryzen 5 3600 and 11% faster than the 5600X than its numbered namesake. It turns out that with a good overclocking of the memory, we already get something in between 5600X and 3600. We would like more, but the dominance of a solid 32 MB L3 cache in some games played its role. If we were to replace Lara, Counter and Starcraft with games like dogs, troy and cyberpunk, the balance would shift towards APU.

What do two memory controllers give?

It's interesting to find out how much two memory controllers and slightly more overclocking potential of the new APUs gave in games that are more apathetic towards the cache. Let's take a 5600G with three memory settings and compare them in cyberpunk and watch dogs with the 5600X, lowering its core frequency to the same 4500 MHz so that it does not confuse the cards.

In the first game, 200 MHz did not give anything, since the timings increased slightly with them, reducing the increase to the level of error. However, compared to 3800 MHz, there is already a nice bonus in the form of 9% additional frames on average. And if not for him, then, as in stock, the 5600X would be ahead of the G version by 12%, and not by 4.

In dogs, the difference is smaller. At 733 MHz, a higher memory frequency with its corresponding timings allows you to get only 6% more FPS on average, and according to the statistics of rare and very rare events, the difference is even smaller.The difference between accelerations in the center once again reminds of the lack of the need to squeeze all the juices from iron.


What happens in the end? The processor part of the APU remained somewhere between the past and the current generation of ordinary processors of the raisen, even despite the abbreviation in the lag in the volume of L3 of the cache and a greater acceleration of memory. That is, 5600G can not be called analogue 5600X. In any case, performance is much higher than 4000 episodes.

There were no changes in the integrated graphics on paper, but we somehow received a small increase.

Ryzen 7 5700G disappointed a little. In the game plan, it is almost indistinguishable from 5600G both the graphic part and processor, but it costs tangibly more. The past suggests that each new generation of razenovs 600 is superior in games or, at least not worse than the 700 last generation, but is cheaper.

5700G is revealed in the working tasks involving all streams like the same bluebench.

The productivity is close to 5800X (especially if you remove the limits), and heat dissipation and temperature are much lower.