AMD Ryzen 9 7900X CPU Review & Benchmarks vs. R7 1700, 7950X, & More
video description
Date: 2022-09-27
Related videos
Comments and reviews: 14
mapesdhs
Steve, I think you need to do a deep-dive on this new stye of thermal behaviour in Zen4 , because people keep getting it horrendously wrong. Many insist on using the word throttle which is clearly wrong, as that has an accepted established meaning for existing CPU designs which does not apply to Zen4. We need new language to describe how Zen4 behaves (ideas?). Relevant questions:
1. Is it possible in the BIOS to prevent the CPU from going beyond the prescribed baseline performance when in theory the cooling capacity would allow further clock headroom?
2. When a cooler is good enough to allow a particular SKU to go beyond its baseline performance, how does that affect efficiency? This relates to Q1, ie. the ability to reign in the default behaviour so that the CPU does not stray away from optimal efficiency even if it could. In other words, for some, if it's possible to prevent the CPU from going even faster than the default then that would be a useful option if it means better efficiency.
3. Is the CPU being fed more voltage than is necessary to maintain the intended thermal behaviour? ie. what scope is there for undervolting, but what does this even mean for this new thermal design behaviour? Perhaps this will improve over time as better BIOS behaviour is rolled out.
4. Is the 95C thermal target in the BIOS? (ie. can it be changed?) Or is it hard wired into the CPU itself?
5. What category of cooler is required to achieve baseline performance for each SKU? People don't yet understand that they're going to see 95C regardless, but presumably below a certain point a lesser cooler is going to leave performance on the table. Is a basic 212 EVO or 6-pipe Snowman enough to allow a 7600X to reach at least its baseline performance? Is using a better cooler than necessary merely going to mean the CPU will automatically go faster just because it can, thus moving it away from optimal efficiency? Indeed, how wide is the efficiency curve?
6. Crazy notion: in order to force the CPU to remain in an efficienct part of the voltage curve, is therefore the easiest way to do this just to use a lesser cooler? :D Ie. instead of undervolting, etc., just use a cooler with lower potential (I don't mean a bad cooler or poor quality, just for example say 120 AIO or sensible air cooler instead of a 360 AIO).
7. Are existing materials and other aspects of current cooler designs optimal for long term use of a CPU which behaves this way? ie. where the contact plate is always going to be touching something that's 95C? Might a cooler behave better during its expected lifespan if it had a different alloy mix, different heatpipe arrangement, etc.? Any comment from the cooler manufacturers? Most users in the past probably aim for oc load temps more in the 60C to 80C, but this doesn't apply to Zen4.
Please help educate viewers, and those in the inevitable comment section flame wars, to better understand these issues, because I'm already seeing a great deal of confusion and utterly incorrect claims about how Zen4 would affect gaming performance with regard to frame spikes.
In all of your old product reviews for CPU coolers, the load temp analysis has always been based around the idea that lower temps = better, but this is not the case with Zen4. How then will this affect your cooler reviews? How does one judge the efficacy of a cooler if the load temp is no longer a useful variable? Perhaps base it on how much extra clock headroom a cooler can provide? This would make it hard to compare to old data though, if reviewers started using Zen4 for testing coolers I mean. If they want viewers to be able to retain their existing worldview about how to judge coolers then using any Zen4 part for cooler reviews is a bad idea.
reply
Steve, I think you need to do a deep-dive on this new stye of thermal behaviour in Zen4 , because people keep getting it horrendously wrong. Many insist on using the word throttle which is clearly wrong, as that has an accepted established meaning for existing CPU designs which does not apply to Zen4. We need new language to describe how Zen4 behaves (ideas?). Relevant questions:
1. Is it possible in the BIOS to prevent the CPU from going beyond the prescribed baseline performance when in theory the cooling capacity would allow further clock headroom?
2. When a cooler is good enough to allow a particular SKU to go beyond its baseline performance, how does that affect efficiency? This relates to Q1, ie. the ability to reign in the default behaviour so that the CPU does not stray away from optimal efficiency even if it could. In other words, for some, if it's possible to prevent the CPU from going even faster than the default then that would be a useful option if it means better efficiency.
3. Is the CPU being fed more voltage than is necessary to maintain the intended thermal behaviour? ie. what scope is there for undervolting, but what does this even mean for this new thermal design behaviour? Perhaps this will improve over time as better BIOS behaviour is rolled out.
4. Is the 95C thermal target in the BIOS? (ie. can it be changed?) Or is it hard wired into the CPU itself?
5. What category of cooler is required to achieve baseline performance for each SKU? People don't yet understand that they're going to see 95C regardless, but presumably below a certain point a lesser cooler is going to leave performance on the table. Is a basic 212 EVO or 6-pipe Snowman enough to allow a 7600X to reach at least its baseline performance? Is using a better cooler than necessary merely going to mean the CPU will automatically go faster just because it can, thus moving it away from optimal efficiency? Indeed, how wide is the efficiency curve?
6. Crazy notion: in order to force the CPU to remain in an efficienct part of the voltage curve, is therefore the easiest way to do this just to use a lesser cooler? :D Ie. instead of undervolting, etc., just use a cooler with lower potential (I don't mean a bad cooler or poor quality, just for example say 120 AIO or sensible air cooler instead of a 360 AIO).
7. Are existing materials and other aspects of current cooler designs optimal for long term use of a CPU which behaves this way? ie. where the contact plate is always going to be touching something that's 95C? Might a cooler behave better during its expected lifespan if it had a different alloy mix, different heatpipe arrangement, etc.? Any comment from the cooler manufacturers? Most users in the past probably aim for oc load temps more in the 60C to 80C, but this doesn't apply to Zen4.
Please help educate viewers, and those in the inevitable comment section flame wars, to better understand these issues, because I'm already seeing a great deal of confusion and utterly incorrect claims about how Zen4 would affect gaming performance with regard to frame spikes.
In all of your old product reviews for CPU coolers, the load temp analysis has always been based around the idea that lower temps = better, but this is not the case with Zen4. How then will this affect your cooler reviews? How does one judge the efficacy of a cooler if the load temp is no longer a useful variable? Perhaps base it on how much extra clock headroom a cooler can provide? This would make it hard to compare to old data though, if reviewers started using Zen4 for testing coolers I mean. If they want viewers to be able to retain their existing worldview about how to judge coolers then using any Zen4 part for cooler reviews is a bad idea.
reply
Izumi
around 7:30-7:45 Quote: So it's a non-linear gain
When I have A Task to do, I have 12 or 16 workers and every worker is exactly the same speed, I can calculate like the following:
96 Workunits / 12 = 8 minutes
96 Workunits / 16 = 6 minutes
The 12 core got 25% less cores but takes 33% more time
The 16 core got 33% more cores but takes 25% less time.
But we already agreed that each worker got the same speed. Your argumentation is that the 16 core should be able to do 33% more work but it only does 25% more work because it only is a 25% timesave. Which it is incorrect tho. The CPU does 33% more work. It only results in 25% less time.
Other example to make it clearer
8 and 16 cores.
96 workunits / 16 = 6 minutes
96 workunits / 8 = 12 minutes
The 8 core got 50% less cores but takes 100% more time
The 16 core got 100% more cores but takes 50% less time.
It should be logically clear that with double the workpower you can only half the time. So you can't argue about It's only 50% less time and then say it is 50% more workpower. That's a false conclusion.
Have a nice day!
PS: Maybe I missinterpreted you. I just wanted to point it out.
PS2: In essence. Yes, it is a non-linear function. But my point is it isn't that we got a workpower loss because we have more cores. It is just a non-linear function because that's how work/time functions work. The word predictable suggests that you meant there is a loss in workpower because there are more cores involved. And not just because that's how math works in that case.
reply
around 7:30-7:45 Quote: So it's a non-linear gain
When I have A Task to do, I have 12 or 16 workers and every worker is exactly the same speed, I can calculate like the following:
96 Workunits / 12 = 8 minutes
96 Workunits / 16 = 6 minutes
The 12 core got 25% less cores but takes 33% more time
The 16 core got 33% more cores but takes 25% less time.
But we already agreed that each worker got the same speed. Your argumentation is that the 16 core should be able to do 33% more work but it only does 25% more work because it only is a 25% timesave. Which it is incorrect tho. The CPU does 33% more work. It only results in 25% less time.
Other example to make it clearer
8 and 16 cores.
96 workunits / 16 = 6 minutes
96 workunits / 8 = 12 minutes
The 8 core got 50% less cores but takes 100% more time
The 16 core got 100% more cores but takes 50% less time.
It should be logically clear that with double the workpower you can only half the time. So you can't argue about It's only 50% less time and then say it is 50% more workpower. That's a false conclusion.
Have a nice day!
PS: Maybe I missinterpreted you. I just wanted to point it out.
PS2: In essence. Yes, it is a non-linear function. But my point is it isn't that we got a workpower loss because we have more cores. It is just a non-linear function because that's how work/time functions work. The word predictable suggests that you meant there is a loss in workpower because there are more cores involved. And not just because that's how math works in that case.
reply
Kernelpickle
AMD isn t ceding the low end, they re going to keep AM4 going for awhile longer and their previous generation parts that were high end are going to serve those budget minded customers.
They made a ton of chiplets that were ideal for 5800X CPUs and they can be had for 250 and beat Intel on price and performance at that end of the spectrum. It makes no sense to made new parts that are cheap when they don t have B series boards available in the new socket, and when they drop so will the prices and we ll start seeing more budget friendly options. AMD are always ahead of Intel on the pricing and they take their margins whenever they can get them, like right after a launch when they rule the top of the charts. They did this last time with 2000 series parts after 3000 dropped, and with 3000 series after 5000 dropped. AMD doesn t pretend their old parts don t exist on shelves new in box, so neither should anyone building a CPU between now and Xmas.
reply
AMD isn t ceding the low end, they re going to keep AM4 going for awhile longer and their previous generation parts that were high end are going to serve those budget minded customers.
They made a ton of chiplets that were ideal for 5800X CPUs and they can be had for 250 and beat Intel on price and performance at that end of the spectrum. It makes no sense to made new parts that are cheap when they don t have B series boards available in the new socket, and when they drop so will the prices and we ll start seeing more budget friendly options. AMD are always ahead of Intel on the pricing and they take their margins whenever they can get them, like right after a launch when they rule the top of the charts. They did this last time with 2000 series parts after 3000 dropped, and with 3000 series after 5000 dropped. AMD doesn t pretend their old parts don t exist on shelves new in box, so neither should anyone building a CPU between now and Xmas.
reply
Brian
Maybe it is time to get a better gaming CPU test that are more CPU demanding? All i am looking for is gaming and before you say you only need 8 cores bla bla I am building the new system to last and for Star citizen which is both GPU and CPU demanding at the minuet. eg the best cpu so far is the 5800x3d by far just hitting 60fps 1440p.
the game is still an Alpha and yes it will get better optimized but I am playing it now and will going forward. I do know it uses 100% of my 6 cores.
so I will be looking at the 7800x3d or 7950x3d or hoping 7900x3d if that one comes out ( dont give a Sh t about power or cost as long its not Nvidia daft I use a microwave to cook and a big oven for just a small item so use per Watt is only down to how much is it going to cost after a night of gaming ? 1 or 10 per night then think on how much a pint of beer is! )
reply
Maybe it is time to get a better gaming CPU test that are more CPU demanding? All i am looking for is gaming and before you say you only need 8 cores bla bla I am building the new system to last and for Star citizen which is both GPU and CPU demanding at the minuet. eg the best cpu so far is the 5800x3d by far just hitting 60fps 1440p.
the game is still an Alpha and yes it will get better optimized but I am playing it now and will going forward. I do know it uses 100% of my 6 cores.
so I will be looking at the 7800x3d or 7950x3d or hoping 7900x3d if that one comes out ( dont give a Sh t about power or cost as long its not Nvidia daft I use a microwave to cook and a big oven for just a small item so use per Watt is only down to how much is it going to cost after a night of gaming ? 1 or 10 per night then think on how much a pint of beer is! )
reply
Todd
I just gave up and ordered the 5600x. They are 50 off on NewEgg for 150 total. I got an Asrock B550 for 95 too. The main thing is it is a 65w (and you show 67.2w) and the 7600X is 105w (you show 116.4) That is almost 2x the power. I find that unacceptable. With an MSI board running 290 and 16GB DDR5 running 240. I already have 16GB DDR4 3200, but if I had to buy more I could get some G.SKILL Ripjaws V Series 16GB (2 x 8GB) DDR4 3200 for 45. So a 7600X CPU, board, and memory costs 830 and the same for a 5600x is 300. The extra performance of the 7600x is not worth an extra 530 and an extra 50w dumped in my system. This should be a nice upgrade from my Ryzen 3 1400. Now all I need if for the RTX 3060 or the 6650XT to drop in price another 100.
reply
I just gave up and ordered the 5600x. They are 50 off on NewEgg for 150 total. I got an Asrock B550 for 95 too. The main thing is it is a 65w (and you show 67.2w) and the 7600X is 105w (you show 116.4) That is almost 2x the power. I find that unacceptable. With an MSI board running 290 and 16GB DDR5 running 240. I already have 16GB DDR4 3200, but if I had to buy more I could get some G.SKILL Ripjaws V Series 16GB (2 x 8GB) DDR4 3200 for 45. So a 7600X CPU, board, and memory costs 830 and the same for a 5600x is 300. The extra performance of the 7600x is not worth an extra 530 and an extra 50w dumped in my system. This should be a nice upgrade from my Ryzen 3 1400. Now all I need if for the RTX 3060 or the 6650XT to drop in price another 100.
reply
Raphael
Thank you for the very technical reviews as usual, really like your videos!
I have 2 comments, hope they are not too dumb..
1) The efficiency test (w/h for each frame) is really useful because it objectively serves to prove/counter claims of efficiency from amd and intel, so that's a great addition.
I was wondering though if you could add other efficiency tests like these, say for certain games or other production tools, since cpu performance may vary a lot depending on what it's running
2) Sorry I am mostly a gamer and not the least knowledgeable as you guys are, but would it be possible to add other production tests? Even if not a usual software people use, but just to again show the scalability between CPUs
reply
Thank you for the very technical reviews as usual, really like your videos!
I have 2 comments, hope they are not too dumb..
1) The efficiency test (w/h for each frame) is really useful because it objectively serves to prove/counter claims of efficiency from amd and intel, so that's a great addition.
I was wondering though if you could add other efficiency tests like these, say for certain games or other production tools, since cpu performance may vary a lot depending on what it's running
2) Sorry I am mostly a gamer and not the least knowledgeable as you guys are, but would it be possible to add other production tests? Even if not a usual software people use, but just to again show the scalability between CPUs
reply
elmer
Just one note. On workloads where lower is better, its a bit misleading to say how much faster it is in %, since its calculated differently than the usual higher is better. For instance in the chromium compile benchmark you have 7900x finishing in 46.5 seconds and the 5900x finishing in 61.7 seconds. You said this is a 25% improvement because 1-46.5/61.7 = 0.25. However if you change it to how much each cpu can do per second you get 100/61.7 = 1.62 and 100/46.5 = 2.15 respectively. Which means the 7900x does 32% more work per second. I think its fine to calculate the improvement in terms of reduced compile time, however the problem is that the different benchmarks are calculated differently.
reply
Just one note. On workloads where lower is better, its a bit misleading to say how much faster it is in %, since its calculated differently than the usual higher is better. For instance in the chromium compile benchmark you have 7900x finishing in 46.5 seconds and the 5900x finishing in 61.7 seconds. You said this is a 25% improvement because 1-46.5/61.7 = 0.25. However if you change it to how much each cpu can do per second you get 100/61.7 = 1.62 and 100/46.5 = 2.15 respectively. Which means the 7900x does 32% more work per second. I think its fine to calculate the improvement in terms of reduced compile time, however the problem is that the different benchmarks are calculated differently.
reply
3XOUT
After watching your coverage and DerBaur's latest videos on these CPUs, I just don t get them at all.
I don t get temps and I don t get the design of the IHS. Seems these temps could have been lower with them not trying to design the IHS for compatibility (which it apparently still doesn't have in a lot of cases), and the open design seems to have no function at all other than getting paste or liquid metal where you really don't want it.
Seems like there are better choices within their last SKUs for the vast majority of ppl. Maybe I should even begin to consider INTEL for the future. I def do not want a CPU pumping boiling temps into my loop consistently at any point.
reply
After watching your coverage and DerBaur's latest videos on these CPUs, I just don t get them at all.
I don t get temps and I don t get the design of the IHS. Seems these temps could have been lower with them not trying to design the IHS for compatibility (which it apparently still doesn't have in a lot of cases), and the open design seems to have no function at all other than getting paste or liquid metal where you really don't want it.
Seems like there are better choices within their last SKUs for the vast majority of ppl. Maybe I should even begin to consider INTEL for the future. I def do not want a CPU pumping boiling temps into my loop consistently at any point.
reply
DrSuperGood
Blender power efficiency is probably not a good way to rate power efficiency outside of production workloads. People using CPUs to game on will not be hammering all their cores with heavy, possibly AVX, workloads.
In my opinion a more appropriate for gamers measurement would be to pick some game and limit frame rate to 60, 120 or some typical FPS which is nowhere near CPU bottleneck for the processors. This would allow CPU power consumption to be comparable directly since the times cancel out to become joules per frame. This would capture power efficiency in a realistic (gaming) workload rather than stress test or production (blender) workload.
reply
Blender power efficiency is probably not a good way to rate power efficiency outside of production workloads. People using CPUs to game on will not be hammering all their cores with heavy, possibly AVX, workloads.
In my opinion a more appropriate for gamers measurement would be to pick some game and limit frame rate to 60, 120 or some typical FPS which is nowhere near CPU bottleneck for the processors. This would allow CPU power consumption to be comparable directly since the times cancel out to become joules per frame. This would capture power efficiency in a realistic (gaming) workload rather than stress test or production (blender) workload.
reply
notachannel
it looks like for gaming if you own an AM4 board, your only sensible upgrade path if you play sims or MMOs is still 5800x3d till AMD comes out with 3D cache variants on their new gen, the cache is extremely helpful for cpu bound games such as MMOs and sims with complex physics engines to the point extra frequency of new gen barely does anything extra
people looking at CS:GO as the 'traditional' benchmark for cpu bound games don't realise that it's a game rather easy on cache because of limited amount of assets being used at the same time compared to much more complex games
reply
it looks like for gaming if you own an AM4 board, your only sensible upgrade path if you play sims or MMOs is still 5800x3d till AMD comes out with 3D cache variants on their new gen, the cache is extremely helpful for cpu bound games such as MMOs and sims with complex physics engines to the point extra frequency of new gen barely does anything extra
people looking at CS:GO as the 'traditional' benchmark for cpu bound games don't realise that it's a game rather easy on cache because of limited amount of assets being used at the same time compared to much more complex games
reply
Snafu
I'd like one, but with a lot of the motherboard manufacturers price gouging their boards, I'll have to wait for the lower priced motherboard B series, rather than shell out the 500 on the overpriced, mediocre motherboards they are gouging at launch. Some 400 motherboards that don't even have debug codes.
I wonder what the 150 boards will be but I heard they're not even real boards, that they're just a picture of a motherboard and you're just buying a box. I'll have to prepare as we move from GPU price gouging, to Motherboard price gouging.
reply
I'd like one, but with a lot of the motherboard manufacturers price gouging their boards, I'll have to wait for the lower priced motherboard B series, rather than shell out the 500 on the overpriced, mediocre motherboards they are gouging at launch. Some 400 motherboards that don't even have debug codes.
I wonder what the 150 boards will be but I heard they're not even real boards, that they're just a picture of a motherboard and you're just buying a box. I'll have to prepare as we move from GPU price gouging, to Motherboard price gouging.
reply
Kharak
okay i just gotta say that efficiency comparison from the previous videos made like no sense whatsoever, comparing the 5950 and 7950 at their max usage says nothing about efficiency when we're hitting such a wall of ever diminishing returns...
i mean a more useful comparison would be maybe overclocking the 5950 to the same level as the 7950, or lowering the 7950 to that, maybe even comparing their performance with the same power limit imposed on both
adding the 12900 to such a comparison would be nice too, to put things in perspective
reply
okay i just gotta say that efficiency comparison from the previous videos made like no sense whatsoever, comparing the 5950 and 7950 at their max usage says nothing about efficiency when we're hitting such a wall of ever diminishing returns...
i mean a more useful comparison would be maybe overclocking the 5950 to the same level as the 7950, or lowering the 7950 to that, maybe even comparing their performance with the same power limit imposed on both
adding the 12900 to such a comparison would be nice too, to put things in perspective
reply
Rabbott
Judging by the data I've seen on performance at different power limits (7950x beats or matches the 12900k on average 65watts) I would say AMD had a good balance between gaining efficiency and letting the power ramp up to compete against intel's higher power parts. It looks like the optimised the efficiency curve for laptop & server, and just let desktop crank up to 11. Judging on people's sentiment for alder lake, I don't blame them. People have voted with their wallets for what they want. Laptop should be killer though.
reply
Judging by the data I've seen on performance at different power limits (7950x beats or matches the 12900k on average 65watts) I would say AMD had a good balance between gaining efficiency and letting the power ramp up to compete against intel's higher power parts. It looks like the optimised the efficiency curve for laptop & server, and just let desktop crank up to 11. Judging on people's sentiment for alder lake, I don't blame them. People have voted with their wallets for what they want. Laptop should be killer though.
reply
deadman327
Watch a Der Bauer video on zen4 delidding the cpu. His conclusion was that the heat spreader was really thick. By design, to reach the same z height as zen3. Just so the older cooling solutions will work for zen4. He wasn't sure why amd did this because it leaves so much more performance on the table. Proven after he ran direct die cooling with dry ice. Wonder if lapping a zen4 cpu and adjusting/modding a cooling solution, how much more performance you can manage?
reply
Watch a Der Bauer video on zen4 delidding the cpu. His conclusion was that the heat spreader was really thick. By design, to reach the same z height as zen3. Just so the older cooling solutions will work for zen4. He wasn't sure why amd did this because it leaves so much more performance on the table. Proven after he ran direct die cooling with dry ice. Wonder if lapping a zen4 cpu and adjusting/modding a cooling solution, how much more performance you can manage?
reply
Add a review, comment
Other channel videos
