News Ryzen 5 3500X Reviewed: Promising But Out Of Reach

[Admin]

Chinese media Expreview reviews AMD's latest Rzyen 5 3500X processor.

Ryzen 5 3500X Reviewed: Promising But Out Of Reach : Read more

 

[hftvhftv]

This would have been a great chip but AMD probably didn't want to take a piece out of the 3600's market

 

[hannibal]

Also They may not have too Many chiplets that requires to cut of the hyper threading... So these are those that has been cut down as much as needed.

 

[Gillerer]

Minute FPS differences shouldn't be considered "wins" for either CPU.

 

[alextheblue]

I am disappointed a more affordable Zen 2 won't hit the US market yet. Maybe they're waiting until more 2000 inventory clears out?

 

[Gillerer]

Or they're trying to incentivize people in more affluent parts of the world to spend up and get the R5 3600, by having the next step down be the Zen+-based 3000 series APUs.

Also, at least in China the 3500X is OEM only. Reversing that for the US (or Europe) would go against the whole idea; offering a more affordable SKU for a cost-sensitive market. This means you won't be able to purchase it outside pre-built systems (and aftermarket gray resellers), even if it ever makes its way to western markets. Would you order one from Aliexpress?

It's also possible that, with the way the existing models are selling, AMD doesn't have enough faulty dies to commit to non-SMT SKUs to satisfy the demand of the entire world. This is another reason to limit such a product to a single - albeit large - market.

 

[Rdslw]

This would have been a great chip but AMD probably didn't want to take a piece out of the 3600's market

I think its a great chip, but they either don't have that much failed 3600, that need lower clock or smt disabled that make it feasible to steal market out of what still is 1'st and 2'nd ryzen series territory.
2300x up to 1200 still stay strong on that end, and they still want to get some $ from 1/2 series.
Its not good to segment the market so much, as they now ride the train of being faster than Intel for WHOLE 3'rd gen lineup.

 

[cryoburner]

This would have been a great chip but AMD probably didn't want to take a piece out of the 3600's market

It might have been alright, but I'm not sure it would really be filling much of a need. The great thing about the rest of the Ryzen 3000 lineup is that each chip is offering higher thread counts than Intel in any given price range. The Ryzen 3600 offers i7-like multithreaded performance for $100+ less. The 3700X offers i9-like multithreaded performance for $100+ less. And the 3900X offers performance comparable to Intel's current HEDT processors for hundreds of dollars less.

What would a Ryzen 3500X offer? Performance similar to an i5-9400F, for around the same price as the 9400F. Looking on PCPartPicker, all but one online retailer in the US currently has the 9400F priced under $150, with Amazon selling it for $140. AMD would need to charge no more than that for the 3500X for it to make any sense, and why do that when they can simply leave SMT enabled and sell the same processor near the $200 price point, where it already offers notably better value than the competition? At close to $150, there would be no thread or core count advantage over the 9400F, and no price advantage either. And with 7nm production being as limited as it is, pricing it much below that wouldn't make much sense. Plus, it's already possible to get last generation's Ryzen 2600 with SMT enabled for around $120, which seems like a really good deal even if it lacks the 15% higher IPC of Zen 2.

Also They may not have too Many chiplets that requires to cut of the hyper threading... So these are those that has been cut down as much as needed.

I kind of doubt that any meaningful number of chips would "need" to have SMT disabled. It should be an integral function of each core, so if the SMT of a core didn't work, most likely the entire core wouldn't work. And these are already chips that have a pair of cores disabled, just like the 3600, so presumably a defective core would just be disabled entirely.

Most likely, the SMT is perfectly functional on these chips, and is just getting disabled for market-segmentation reasons. That is what Intel does. A 9900K and a 9700K are the same silicon, just one has had its SMT disabled so that they can charge a $100+ premium for the feature.

I suspect the same applies to most of the cores AMD disables for their 6 and 12-core processors as well. Most are probably functional cores, though they are likely the lowest-binned cores on the chip. Defective cores could also be disabled though.

It is probable that binning is being used to differentiate which chips end up in the 3600, and which in the 3500X, but their clocks are pretty close. I suppose some chips might fall just short of hitting the 3600's boost clocks though.

 

[hftvhftv]

It might have been alright, but I'm not sure it would really be filling much of a need. The great thing about the rest of the Ryzen 3000 lineup is that each chip is offering higher thread counts than Intel in any given price range. The Ryzen 3600 offers i7-like multithreaded performance for $100+ less. The 3700X offers i9-like multithreaded performance for $100+ less. And the 3900X offers performance comparable to Intel's current HEDT processors for hundreds of dollars less.

What would a Ryzen 3500X offer? Performance similar to an i5-9400F, for around the same price as the 9400F. Looking on PCPartPicker, all but one online retailer in the US currently has the 9400F priced under $150, with Amazon selling it for $140. AMD would need to charge no more than that for the 3500X for it to make any sense, and why do that when they can simply leave SMT enabled and sell the same processor near the $200 price point, where it already offers notably better value than the competition? At close to $150, there would be no thread or core count advantage over the 9400F, and no price advantage either. And with 7nm production being as limited as it is, pricing it much below that wouldn't make much sense. Plus, it's already possible to get last generation's Ryzen 2600 with SMT enabled for around $120, which seems like a really good deal even if it lacks the 15% higher IPC of Zen 2.


I kind of doubt that any meaningful number of chips would "need" to have SMT disabled. It should be an integral function of each core, so if the SMT of a core didn't work, most likely the entire core wouldn't work. And these are already chips that have a pair of cores disabled, just like the 3600, so presumably a defective core would just be disabled entirely.

Most likely, the SMT is perfectly functional on these chips, and is just getting disabled for market-segmentation reasons. That is what Intel does. A 9900K and a 9700K are the same silicon, just one has had its SMT disabled so that they can charge a $100+ premium for the feature.

I suspect the same applies to most of the cores AMD disables for their 6 and 12-core processors as well. Most are probably functional cores, though they are likely the lowest-binned cores on the chip. Defective cores could also be disabled though.

It is probable that binning is being used to differentiate which chips end up in the 3600, and which in the 3500X, but their clocks are pretty close. I suppose some chips might fall just short of hitting the 3600's boost clocks though.

Some probably do fall short of the boost clocks, but after updating my BIOS, my 3600 can hit 4200Mhz all day while gaming

 

[Gillerer]

It is probable that binning is being used to differentiate which chips end up in the 3600, and which in the 3500X, but their clocks are pretty close. I suppose some chips might fall just short of hitting the 3600's boost clocks though.

The binning in this case is probably based on power or voltage; if you get dies with more than 2 cores that have exceptionally poor efficiency characteristics, you can make it easier for them to clock reasonably within the power limits by reducing the most work they can ever do - by disabling SMT. Since Windows scheduler will know to put lightly threaded workloads on the best cores, this wouldn't affect frequencies in those workloads much, either.

The way AMD needs/wants to symmetrically disable the same number of cores from each CCX, may require them to introduce an SKU such as 3500X, to harvest "lopsided" dies, where one CCX has 2 or 3 poor cores.

If you only have at most 1 really bad efficiency cores per CCX, they'd obviously get disabled, and depending on the quality of the remaining 6, be used in in a 3600, 3600X, 3900X, Threadripper or EPYC.

Some probably do fall short of the boost clocks, but after updating my BIOS, my 3600 can hit 4200Mhz all day while gaming

I don't think @cryoburner was referring to boost clocks on any CPU sold as a 3600; Rather why a die with six functional cores - a potential candidate for 3600 use - would be relegated to use in a 3500X.

 

[Olle P]

I am disappointed a more affordable Zen 2 won't hit the US market yet. Maybe they're waiting until more 2000 inventory clears out?

I've also been waiting for some Ryzen 3 based on Zen 2 to be released.

My guess is that it won't happen, at least not any time soon, because of a combination of factors:

1. The yield of Zen 2 chiplets is way too good to provide a sufficient amount of chiplets that won't be good enough for the more profitable SKUs.
2. The demand for Zen2 products is very high, with production of CPUs (putting the pieces together) being a bottleneck. The lucrative SKUs get priority.
3. The parts of a Zen 2 Ryzen (I/O chip, substrate, IHS, pins, etc) and the manufacturing of the CPU is (I guess) still to expensive for a low cost SKU, even if it's using "discarded" chiplets.

 

[alextheblue]

I've also been waiting for some Ryzen 3 based on Zen 2 to be released.

My guess is that it won't happen, at least not any time soon, because of a combination of factors:

1. The yield of Zen 2 chiplets is way too good to provide a sufficient amount of chiplets that won't be good enough for the more profitable SKUs.
2. The demand for Zen2 products is very high, with production of CPUs (putting the pieces together) being a bottleneck. The lucrative SKUs get priority.
3. The parts of a Zen 2 Ryzen (I/O chip, substrate, IHS, pins, etc) and the manufacturing of the CPU is (I guess) still to expensive for a low cost SKU, even if it's using "discarded" chiplets.

Indeed, although for #3 I would point out that the I/O chip is built on a more affordable GF process and could potentially be moved (in a future Zen revision) to 12+.

Still, it would be nice to see them release a cheaper monolithic design based on Renoir for the desktop. Including unlocked models with a disabled GPU.

 

[Gillerer]

Indeed, although for #3 I would point out that the I/O chip is built on a more affordable GF process and could potentially be moved (in a future Zen revision) to 12+.

AMD may well have designed the I/O dies to last a couple of generations, to save on development costs. With this flexible package design, they could just swap out the core dies for newer ones with Zen 3. Besides maybe small tweaks, it's only when they move to DDR5 and (probably) AM5 socket, that they'll absolutely need to completely update it.

 

[alextheblue]

AMD may well have designed the I/O dies to last a couple of generations, to save on development costs. With this flexible package design, they could just swap out the core dies for newer ones with Zen 3. Besides maybe small tweaks, it's only when they move to DDR5 and (probably) AM5 socket, that they'll absolutely need to completely update it.

Any I/O changes would require at least a partial redesign. For instance if they decided to move to a newer PCIe version, or upgrade the native USB to a newer gen. My point is when they overhaul the I/O die for any reason, it may be a good opportunity to move to a better but still not cutting edge process like 12+. You could even design it to do the opposite of what you're suggesting... build a newer I/O in preparation for Zen 3, but design it so it also supports Zen 2. Updates of the CPU cores and I/O no longer have to be joined at the hip, there's a lot of possibilities.

In addition to cost benefits of using a less expensive GF process (be it 14, 12, or 12+) for the I/O, the other huge benefit is more cores per 7nm wafer. Supply of TSMC 7nm is tight, and the chiplets really make great use of the wafers.

 

[Jo3yization]

Please at least run competitive settings if you're going to test CSGO, or show the 0.1///1% lows. Cheers.

 

[rullydotcom]

finally got some numbers,, I think i would go for intel